Aldh1a2⁺ fibroblastic reticular cells regulate lymphocyte recruitment in omental milky spots

The peritoneal cavity, the largest fluid-filled abdominal cavity in mammals, contains visceral organs such as stomach, spleen, liver, and intestine. Although the peritoneal cavity is normally sterile, infection, which can occur by a pathological or traumatic loss of intestinal wall integrity, cirrhosis, pancreatitis, abdominal surgery, or peritoneal dialysis, can cause life-threatening sepsis since the peritoneal cavity serves as a conduit to other vital organs for pathogen spread (Clements et al., 2021, Vega-Perez et al., 2021). Visceral adipose tissue omentum has important immunological properties against abdominal infection and visceral organ injury (Liu et al., 2021a; Di Nicola, 2019). It is a highly mobile organ that helps to occlude the sites of peritoneal inflammation and injury (Liu et al., 2021b; Zindel et al., 2021). Although it is primarily adipose tissue, the unique feature of omentum is the presence of lymphoid aggregates, termed “milky spots,” which are located beneath the mesothelial surface (Meza-Perez and Randall, 2017). Milky spots compose unique niches to collect antigens, particles, and pathogens in peritoneal cavity and function as places to promote immune responses that in many ways resemble the follicles of secondary lymphoid organs (Koning and Mebius, 2020, Jackson-Jones and Benezech, 2020).

The development of milky spots, similar to secondary lymphoid organs such as lymph nodes and Peyer’s patches, is innately programmed, as the milky spots appear during fetal development in humans, mice, and other mammals (Shimotsuma et al., 1994; Krist et al., 1997; Solvason et al., 1992). Whereas the first phase of lymph node development requires lymphoid tissue–inducer cells that express lymphotoxin-αβ to interact with mesenchymal cells, milky spots do not require these interactions, indicating that the formation of milky spots uses mechanisms distinct from those used by secondary lymphoid organs (Rangel-Moreno et al., 2009). Additionally, postnatal exposure to microbial products greatly increases the size and number of milky spots, which is reminiscent of the formation of tertiary lymphoid organs (Cruz-Migoni and Caamano, 2016; Benezech et al., 2015). These pieces of evidence illustrate the hybrid nature of milky spots between secondary lymphoid organs and ectopic lymphoid tissues; the formation of milky spots is dictated by both hard-wired developmental programs and inflammation-mediated postnatal enhancement. However, the mechanism of milky spot formation is poorly understood.

In addition to triggering peritoneal immunity, milky spots serve as the anatomic site of cell entry into the peritoneal cavity (Meza-Perez and Randall, 2017). Surgical removal of the omentum reduces lymphocyte entry into the peritoneal cavity and abolishes neutrophil influx during the early stage of peritonitis (Berberich et al., 2008; Buscher et al., 2016). Consistently, mice deficient in chemokine CXCL13, an essential chemokine for milky spot formation, and its cognate receptor CXCR5 severely reduce the number of lymphocytes in the peritoneal cavity (Ansel et al., 2002; Rangel-Moreno et al., 2009, Hopken et al., 2004). Milky spots contain specialized blood vessels expressing markers of high endothelial venules (HEVs) that permit entry of blood-borne leukocytes into milky spots (Berberich et al., 2008; Rangel-Moreno et al., 2009). Subsequently, lymphocytes are released to the peritoneal cavity through fenestrations between the mesothelial cells that cover milky spots (Cranshaw and Leak, 1990; Mironov et al., 1979). Thus, the leukocyte transit to the peritoneal cavity, in principle, is regulated by two steps: (1) blood-borne leukocytes entering the milky spots through HEVs and (2) egress to the peritoneal cavity by transmigrating through the mesothelium.

Accumulating evidence indicates that stromal–immune cell interactions are crucial for the development, organization, and function of lymphoid organs (Krishnamurty and Turley, 2020; Krausgruber et al., 2020). Fibroblastic reticular cells (FRCs) are specialized lymphoid organ fibroblasts that form dedicated microenvironments to provide niche factors and regulate immune cell traffic (Lutge et al., 2021; Fletcher et al., 2015). FRCs are characterized by the expression of podoplanin (PDPN) and platelet-derived growth factor receptor-α (PDGFRα, CD140a) and their lack of expression of PECAM-1 and CD45 (Fletcher et al., 2015). They also express molecules common to many myofibroblasts such as desmin, CD105, and secrete chemokine CCL19 (Krishnamurty and Turley, 2020, Lutge et al., 2021). In classical secondary lymphoid organs, subsets of FRCs including T cell zone FRCs, B cell zone FRCs, pericytic FRCs, marginal reticular cells, and follicular dendritic cells have been identified based on their localization and phenotypic markers, each of which performs dedicated functions to underpin lymphoid organ microarchitecture (Li et al., 2021). On the contrary, milky spots of omentum display a reduced structural complexity lacking neither clear segregation of B and T cell zones nor a well-defined follicular dendritic cell network (Rangel-Moreno et al., 2009; Perez-Shibayama et al., 2019). Although it has emerged that FRCs in milky spots are crucial for triggering abdominal immune responses (Perez-Shibayama et al., 2018), understanding of phenotypic and functional diversity of FRCs in the milky spots is under-explored. Here, we identified a novel subset of milky spot FRCs, characterized by the expression of retinoic acid–converting enzyme, Aldh1a2, as well as the endothelial cell marker, Tie2. Genetic depletion of Aldh1a2⁺ FRCs led to a substantial reduction in the number of T cells and conventional B2 B cells in milky spots due to the impaired recruitment of these lymphocytes from circulation. Mechanistically, we found that Aldh1a2⁺ FRCs are required for the induction of homeostatic chemokine CXCL12 displayed on HEVs in a manner dependent on retinoic acid. Lastly, a progressive decline in the number of lymphocytes in the peritoneal cavity was observed by Aldh1a2⁺ FRC ablation. Altogether, these results demonstrate a subset of FRCs that are an integral component of milky spot formation through the regulation of lymphocyte recruitment.

Multiple studies have shown that retinoic acid, a lipophilic molecule derived from vitamin A, is abundantly present in omentum, and we previously showed that retinoic acid regulates functional polarization of macrophages through the induction of transcription factor GATA6 (Maruya et al., 2011; Okabe and Medzhitov, 2014; Jayakumar et al., 2022). Retinoic acid controls a broad array of immune functions including lymphocyte/myeloid cell differentiation, class switching of B cells, lymphocyte homing, and lymphoid organization (Erkelens and Mebius, 2017), suggesting the role(s) of retinoic acid not only in macrophage regulation but also in other immunological functions in omentum. Omentum was characterized by the high expression of Aldh1a2 gene, which encodes a rate-limiting enzyme regulating retinoic acid synthesis, whereas Aldh1a1, another important enzyme for retinoic acid synthesis, showed a distinct expression pattern (Fig. 1 A and Fig. S1 A). Although previous studies reported that mesothelial cells and fibroblasts in omentum expressed Aldh1a2 gene (Buechler et al., 2019; Jackson-Jones et al., 2020), the cellular characterization and contribution of each cell type in omental biology remains to be determined. To characterize Aldh1a2-expressing cells in omentum, we utilized fluorescent substrate for aldehyde dehydrogenase (ALDH) enzymes, which assesses retinol metabolism capacity (Buechler et al., 2019). Whereas ALDH activity was detected in non-hematopoietic cells (CD45⁻) of all adipose tissues (omentum, perigonadal, and inguinal fat), we found a fraction of CD45⁻ cells in omentum exhibited the highest ALDH activity (Fig. 1 B). This fraction was further identified as CD45⁻CD140a⁺CD105⁺ALDH^high cells (Fig. 1 C and Fig. S1, B–D). In this population (gated as P1 in Fig. 1 C), the expression of Aldh1a2 mRNA, but not two other retinaldehyde dehydrogenases (Aldh1a1 and Aldh1a3), was enriched (Fig. 1 D), suggesting that selective expression of Aldh1a2 endows the strong ALDH activity. We next determined the localization of ALDH1A2-expressing cells in omentum. Omentum was predominantly occupied by adipose tissue in which scattered distribution of milky spots was observed (Fig. S1 E). Immunofluorescence staining of whole-mount omentum revealed that the majority of cells expressing ALDH1A2 were found to localize in milky spots (Fig. 1 E and Fig. S1 F). These results indicate that a fraction of stromal cells in milky spots exhibit high retinoic acid–metabolizing capacity via the expression Aldh1a2 gene.

To characterize the cellular phenotype of Aldh1a2-expressing cells, we next performed RNA sequencing (RNA-seq) analysis on sorted omental stromal populations (P1–P6 in Fig. 1 C). We identified 41 genes that had the expression of at least twofold higher in P1 compared to all five other populations (Fig. 2 A), and the selective expression of Phex, Postn, Ptx3, Cd143, Adm, and Tie2 in P1 was confirmed by quantitative PCR (Fig. 2 B). Gene ontology analysis revealed that highly expressed genes in P1 are associated with endothelial phenotypes including “vasocontraction” and “blood vessel development” (Fig. S2 A), whereas P1 was clearly distinguishable from PECAM-1⁺ endothelial cells that exhibited low ALDH activity (Fig. 2 C). In contrast, P1 constitutively expressed other well-defined endothelial cell markers, TIE2 and VCAM-1 (Fig. 2 D; Goncharov et al., 2020), suggesting a phenotypic link between P1 and endothelial cells. We also found P1-expressed PDPN, which is known to be expressed in lymphatic endothelial cells, mesothelial cells, and FRCs (Fig. 2 D). Meanwhile, the expression of mesothelial cell markers MSLN, Lrrn4, and Upk3b was not detected in P1 at protein and mRNA levels (Fig. 2, B and D), indicating that P1 does not represent the cells that are generally considered mesothelial cells.

In secondary lymphoid organs, PDPN⁺PECAM-1⁻CD140a⁺VCAM-1⁺CD105⁺ cells, which correspond to P1, P2, and P3 (Fig. 2, C and E), are generally referred to FRCs (Perez-Shibayama et al., 2018). Indeed, they were enriched in the expression of FRC-associated genes including Ccl19, Col4a1, Col14a1, Dpt, and Cd140a (Fig. 2 B and Fig. S2 B). Additionally, ALDH1A2⁺ cells were found in Desmin⁺ filamentous network of FRCs in milky spots (Fig. 2 F; Zeng et al., 2011). These results indicate that omental Aldh1a2⁺ cells are a subset of FRCs. Although the expression of FRC-associated genes was also detected in P2 and P3 at mRNA and protein levels, TIE2 was exclusively expressed in P1 (Fig. 2 E). Notably, the cells displaying both ALDH activity and TIE2 were absent among FRCs (CD45⁻PDPN⁺PECAM-1⁻) in lymph nodes, suggesting that Aldh1a2⁺ cells represent a subset of FRCs distinct from the cells that are present in secondary lymphoid organs. To further delineate the cellular phenotype of Aldh1a2⁺ FRCs, we analyzed single-cell RNA-seq (scRNA-seq) data on Cd140a⁺ cells of omental CD45⁻CD41⁻TER119⁻PECAM1⁻PDPN^+/− stromal cells (Fig. S2 C; obtained from Jackson-Jones et al., 2020). We found Aldh1a2 expression was enriched in a fraction of Ccl11⁺CD140a⁺ fibroblasts (Cluster 1 in Fig. S2, C and D), which was distinct from previously described omental FRCs that were characterized by the expression of Ccl19, Matn2, and CD140a (Cluster 4 in Fig. S2, C and D). Consistent with RNA expression analysis (Fig. 2 B), Aldh1a2⁺Ccl11⁺CD140a⁺ cluster was enriched with the expression of Ptx3, Cd143, and Adm, supporting that this cluster corresponded to Aldh1a2⁺ FRCs (Fig. S2 E). Moreover, we identified genes that were highly expressed in Aldh1a2⁺Ccl11⁺CD140a⁺ cluster including Clec3b, Gsn, and Mfap5 (Fig. S1 E). These results suggest that at least two distinct FRC subsets, Aldh1a2⁺ FRCs and Ccl19⁺ FRCs, are present in omentum.

Among the genes that are selectively expressed in P1 (Fig. 2 B), Periostin (Postn) is known to be expressed in restricted cell lineages including myofibroblasts, osteoblasts, and colonic epithelial cells (Kanisicak et al., 2016). To determine in vivo function(s) of Aldh1a2⁺ FRCs, we utilized Postn promoter to selectively target Aldh1a2⁺ FRCs in the omentum and generated mice in which a diphtheria toxin receptor (DTR) was genetically targeted into the first coding exon of Postn gene (Fig. S3 A; Furukawa et al., 2006). Heterozygous mice were healthy and born at the expected Mendelian ratio, and we hereafter used Postn^Dtr/+ heterozygotes for experiments (referred to as Postn-DTR mice). A single injection of low-dose (100 ng/mouse) diphtheria toxin (DT) through the i.p. route efficiently ablated P1 (3.1 ± 0.8% for WT vs. 0.4 ± 0.2% for Postn-DTR among CD140a⁺ population; Fig. 3, A and B). While DT administration had only marginal effect on the frequency of P3–P6, it ablated P2 to some degree (6.7 ± 0.8% for WT vs. 3.5 ± 0.7% for Postn-DTR among CD140a⁺ population; Fig. 3 B and Fig. S3, B and C), suggesting that P2 could be precursors of P1 or vice versa. The depletion of ALDH1A2⁺ FRCs in the milky spots was confirmed by immunofluorescence staining (Fig. 3 C). These results indicate that DT injection in Postn-DTR mice selectively ablated Aldh1a2⁺ FRCs.

Given that FRCs are an integral component of the organization and function of the lymphoid organs (Acton et al., 2021; Cremasco et al., 2014; Kapoor et al., 2021), we next explored whether the ablation of Aldh1a2⁺ FRCs affected the structure of milky spots. To this end, we performed immunostaining of whole-mount omenta with anti-CD45 antibody and the area of CD45⁺ cell aggregates was determined by Z-stack imaging. Although no obvious change in the size of milky spots was observed at 1 d after DT injection, Postn-DTR mice severely reduced the size by 5 d after DT administration (Fig. 4, A and B, left). Moreover, the milky spots of Postn-DTR mice showed low follicle density (Fig. 4 B, right) and possessed fewer numbers of B220⁺ or CD3⁺ lymphocytes than that of WT mice (Fig. 4 B, middle). In contrast, Desmin⁺ network was still observed in Postn-DTR mice (Fig. S4 A). To further assess the alteration to omental leukocyte populations, we performed an scRNA-seq of cells isolated from omenta of WT and Postn-DTR mice at 5 d after DT injection (three pooled mice per group). After quality control and adjusting for technical noise, CD45⁺ hematopoietic cells (7,997 cells for WT mice and 5,233 cells for Postn-DTR mice) were determined by unsupervised classification visualized by using t-distributed stochastic neighbor embedding (t-SNE), and we identified 16 cell types based on the expression of canonical cell marker genes (Fig. 4 C and Fig. S4, B and C). Omental leukocytes were predominantly composed of T and B lymphocytes that were designated with six T cell subsets including naive CD4 T cells (Cd3e⁺Cd4⁺Cd62l⁺Cd44⁻), CD4 T cells (Cd3e⁺Cd4⁺Cd62l⁻Cd44⁺), regulatory T cells (Cd3e⁺Cd4⁺Foxp3⁺), CD8 T cells (Cd3e⁺Cd8a⁺), γδT cells (Cd3e⁺Cxcr6⁺Trgv6⁺), and natural killer T cells (Cd3e⁺Nkg2d⁺Nk1.1⁺), and three B cell subsets that were B2 B cells (Cd19⁺Cd79a⁺Ighd⁺), B1a B cells (Ighm⁺Cd43⁺Ighv11-1⁺Ighv11-2⁺), and B1b B cells (Ighm⁺Cd43⁺Ighv1-11⁺; Barington et al., 2022; Prohaska et al., 2018; Fig. 4 C). Among them, the proportion of all T cell subsets was markedly reduced in Postn-DTR mice (Fig. 4 D). The proportion of conventional B2 B cells was also reduced in Postn-DTR mice whereas that of B1a and B1b B cells showed the opposite trend (Fig. 4 D). Consistently, Postn-DTR mice showed a substantial reduction in the cell numbers of CD4 T, CD8 T, and B2 B cells as well as total CD45⁺ leukocytes, although that of B1 B cells and macrophages was not altered (Fig. 4 E). These results suggest that ablation of Aldh1a2⁺ FRC led to the selective loss of lymphocyte subsets in omentum. We also found that genes such as Rps21, Tmsb10, Ly6d, and Cd52 were highly expressed in T cells and B2 B cells of WT mice compared with that of Postn-DTR mice, as illustrated in the volcano plot (Fig. S4 D). Interestingly, Cd52 has been shown to be induced by retinoic acid in lymphocytes (Li et al., 2003), suggesting the involvement of retinoic acid in the functionality and activation of these lymphocytes. The reduction of these lymphocyte numbers in omenta of Postn-DTR mice was unlikely due to off-target depletion of lymphocytes since Postn was barely detected in omental leukocytes under homeostatic settings (Fig. S4 C), DT did not affect the frequency of circulating lymphocytes in peripheral blood (Fig. S4 E), and lymphoid structures and ER-TR7⁺ fibroblastic network of lymph nodes and spleen were intact in Postn-DTR mice (Fig. S4 F).

T cells and B2 B cells constitutively recirculate between the omentum and other lymphoid tissues, whereas macrophages and B1 B cells are permanent residents in omentum and peritoneal cavity and recirculate with relatively slow rate, respectively (Ansel et al., 2002; Berberich et al., 2008). This suggested that the selective reduction of T cell and B2 B cell numbers in Postn-DTR omenta was possibly due to the impaired recruitment of circulating lymphocytes to milky spots. To test this possibility, GFP-expressing splenocytes were i.v. transferred at 24 h after DT injection, and the positioning of donor splenocytes in the omentum was determined by fluorescence imaging. At 6 h after transfer, donor splenocytes in WT mice abundantly accumulated in the region of milky spots with only a few GFP⁺ cells scattered in the surrounding regions (Fig. 4 F). Meanwhile, milky spots of Postn-DTR mice had significantly fewer numbers of GFP⁺ splenocytes. Consistently, the numbers of GFP⁺ cells in omentum, but not in inguinal lymph node, significantly reduced in Postn-DTR mice (Fig. 4 G). Collectively, these data indicate that ablation of Aldh1a2⁺ FRCs led to the selective impairment of lymphocyte recruitment to the milky spots. Interestingly, majority of F4/80⁺ macrophages were found outside of the lymphoid aggregates that were marked with B220 (Fig. S4 G), suggesting that the conventional lymphocytes are continuously recruited to the milky spots whereas other leukocytes such as macrophages reside primarily outside of the milky spots.

The above data suggested that Aldh1a2⁺ FRCs directly or indirectly recruit circulating lymphocytes to the milky spots. To delineate the molecular mechanism underlying lymphocyte recruitment, we performed RNA-seq analysis of omenta at 5 d after DT injection and identified 39 genes that were downregulated at least 0.5-fold in Postn-DTR mice compared with WT mice (Fig. 5 A). Among these genes, chemokine Cxcl12, which recruits a diverse range of CXCR4-expressing leukocytes (Campbell et al., 2003), was significantly downregulated in omenta of Postn-DTR mice, whereas other homeostatic chemokines such as Cxcl13, Cxcl16, Ccl19, and Ccl21 were comparably expressed between omenta of WT and Postn-DTR mice (Fig. 5 B). Given that CXCL12 is involved in the formation of ectopic lymphoid tissues (Fleige et al., 2014) and its receptor CXCR4 was broadly expressed in omental lymphocytes including T cells and B2 B cells (Fig. S4 C), we investigated the functional role of Cxcl12 in milky spot formation. To this end, WT C57BL/6 mice were i.p. injected with AMD3100, a specific antagonist of CXCR4, and the size of the milky spots was determined by immunofluorescence staining for CD45. We observed that the administration of AMD3100 significantly reduced the size of milky spots (Fig. 5, C and D), consistent with the previous observation (Abe et al., 2009). Additionally, the administration of AMD3100 reduced the numbers of B2 B cells, but not B1 B cells and macrophages, in omentum (Fig. 5 E), similar to those seen in Postn-DTR mice. Collectively, these results indicate that Aldh1a2⁺ FRCs are required for the induction of CXCL12 which is, at least in part, the mechanism regulating milky spot formation.

We next addressed CXCL12-producing cells in omentum. It has been shown that HEVs in lymph nodes display CXCL12, which promotes extravasation of blood-borne leukocytes into the lymph node parenchyma (Okada et al., 2002; Pablos et al., 2003; Santiago et al., 2006), and these observations prompted us to determine the localization of CXCL12 on omental HEVs. Consistent with the previous studies (Berberich et al., 2008; Rangel-Moreno et al., 2009), PECAM-1⁺ vessels expressing HEV marker, peripheral node addressin (PNAd), were exclusively present inside milky spots (Fig. 6 A). Additionally, we observed predominant localization of CXCL12 in the luminal side of HEVs in WT milky spots (Fig. 6 B). In contrast, CXCL12 was absent on HEVs in omenta of Postn-DTR mice (Fig. 6 B), despite PECAM-1⁺ cells expressing PNAd and another HEV marker, mucosal addressin cell adhesion molecule 1 (MAdCAM-1), were equally present in omenta of WT and Postn-DTR mice (Fig. S5, A and B). CXCL12 on HEVs can be produced by endothelial cells themselves or can be produced by surrounding stromal cells and, subsequently, transported to HEVs by transcytosis (Rustenhoven et al., 2021; Weiss et al., 2013). The expression of Cxcl12 mRNA in omentum was found in PECAM-1⁺ endothelial cells, Aldh1a2⁺ FRCs (P1), and conventional FRCs (P2 and P3; Fig. 6 C). Additionally, the expression of Cxcl12 mRNA in both endothelial cells and FRCs was downregulated in Postn-DTR mice (Fig. 6 D). Thus, the decline of Cxcl12 expression in whole omental tissues of Postn-DTR mice could be reflected by the reduction in gene expression in endothelium and conventional FRCs and the ablation of Aldh1a2⁺ FRCs.

Multiple studies have shown that retinoic acid induces the expression of Cxcl12 (Liu et al., 2011; Hu et al., 2017; Kim et al., 2007), suggesting the possibility that Aldh1a2⁺ FRC-derived retinoic acid induces the expression of Cxcl12 gene in endothelial cells and/or FRCs. To test this possibility, we sorted PECAM-1⁺ endothelial cells and PDPN⁺CD140a⁺ FRCs from omenta of WT C57BL/6 mice and stimulated with all trans retinoic acid (ATRA), the most abundant form of retinoic acid in vivo, for 24 h. We observed ATRA stimulation increased the expression of Cxcl12 in endothelial cells about 3.1-fold, whereas it had only marginal effect on the expression in FRCs (Fig. 6 E). To further explore the role of retinoic acid in the expression of omental Cxcl12 gene in vivo, WT C57BL/6 mice were i.p. injected with pan-retinoic acid receptor inverse agonist BMS493. Interestingly, BMS493 treatment was able to diminish the CXCL12 display on HEVs in milky spots (Fig. 6 F and Fig. S5 C). Altogether, these results suggest that homeostatic display of CXCL12 on HEVs in the milky spots is primarily regulated by retinoic acid–dependent induction of Cxcl12 gene in endothelium.

Given that the milky spots of omentum are proposed as the source of peritoneal cells (Ansel et al., 2002; Barington et al., 2022), we next explored the role of Aldh1a2⁺ FRCs in peritoneal lymphocyte composition. To this end, GFP-expressing splenocytes were adoptively transferred i.v. or i.p. at 2 d after DT injection, and the number of donor cells in peritoneal cavity or spleen was determined by flow cytometry at 24 h after transfer, respectively (Fig. 7 A). Although the depletion of Aldh1a2⁺ FRCs did not affect the lymphocyte homing from peritoneal cavity to spleen, that from circulation to peritoneal cavity was significantly diminished (Fig. 7, B and C). Consistently, flow cytometry of peritoneal lavage cells at 6 d after DT injection revealed that Postn-DTR mice had 4.4-fold fewer CD4 T cells, 2.8-fold fewer CD8 T cells, and 4.5-fold fewer B2 cells compared with that of WT controls (Fig. 7 D). These results indicate that Aldh1a2⁺ FRCs are required for the recruitment of T cells and B2 cells to peritoneal cavity.

The reduction in T cell and B2 cell numbers in Postn-DTR mice seemed to be primarily caused by poor lymphocyte recruitment to the milky spots (Fig. 4, E and G). In addition to these conventional lymphocytes, Postn-DTR mice had 2.4-fold fewer B1 B cells in the peritoneal cavity (Fig. 7 D), whereas the number of omental B1 B cells was not affected in Postn-DTR mice (Fig. 4 E). This suggests the possibility that DT had some impact on the lymphocyte egress to the peritoneal cavity from omentum or other sources (Berberich et al., 2008; Jia et al., 2020). In contrast, the number of peritoneal resident cells such as macrophages was not affected in Postn-DTR mice (Fig. 7 D). Consistently, the expression of GATA6, which regulates tissue-specific transcription program and localization of peritoneal macrophages (Okabe and Medzhitov, 2014), was not affected in CD11b⁺ F4/80⁺ MHC-II^{− or low} large peritoneal macrophages of Postn-DTR mice (Fig. 7 E). These results indicate that Aldh1a2⁺ FRCs were not essential for macrophage polarization and localization in peritoneal cavity.

In the present study, we identified a subset of FRCs that was uniquely present in omental milky spots. They were characterized by the enriched expression of gene-encoding retinoic acid converting enzyme, Aldh1a2, as well as endothelial cell–associated genes, including Tie2. Ablation of Aldh1a2⁺ FRCs was not immediately accompanied by the collapse of the lymphoid aggregates, and population of B1 B cells and macrophages was maintained, which suggested that lymphoid infrastructure was maintained despite a lack of Aldh1a2⁺ FRCs. In contrast, the ablation of Aldh1a2⁺ FRCs was associated with the impaired recruitment of circulating lymphocytes to the milky spots, and our results indicated Aldh1a2⁺ FRCs are required for display of CXCL12 on HEVs. Thus, Aldh1a2⁺ FRCs are an integral component of milky spot formation, which is mediated by the recruitment of circulating lymphocytes in a manner dependent on CXCL12.

Although retinoic acid is abundantly present in omentum under homeostatic conditions, the role of retinoic acid in the development and maturation of milky spots remained underexplored. In lymph nodes and Peyer’s patches, retinoic acid initiates lymphoid organ development through the induction of homeostatic chemokine CXCL13 in stromal organizer cells (van de Pavert et al., 2009). However, the ablation of Aldh1a2⁺ FRCs in omentum did not alter Cxcl13 expression (Fig. 5 B). By contrast, our results indicated Aldh1a2⁺ FRC-derived retinoic acid induced CXCL12 in endothelial cells, which is essential for constitutive recruitment of circulating lymphocytes into the milky spots. On the opposite side, Cxcl12 mRNA was also expressed in omental FRCs in a manner independent of retinoic acid. CXCL12 is a key regulator for the development of ectopic lymphoid follicles and, notably, CXCL12 in FRCs is induced by inflammatory signals such as IL-17 during the development of bronchus-associated lymphoid tissue (Fleige et al., 2014; Pandey et al., 2017; Meza-Perez and Randall, 2017). Collectively, this evidence suggest that retinoic acid–dependent CXCL12 induction in endothelium preserves the milky spot formation under homeostatic conditions, whereas FRC-derived CXCL12 might be rather prone to the expansion of lymphoid structure during the inflammatory settings (Fig. 8).

Retinoic acid is essential for the maintenance of peritoneal macrophages through the induction of transcription factor GATA6 (Okabe and Medzhitov, 2014). However, we did not observe a significant change in the peritoneal macrophage number as well as GATA6 expression in Postn-DTR mice (Fig. 7, D and E). A previous study has shown that Wilms’ tumor 1 (WT1)–expressing omental stromal cells, including fibroblasts and mesothelial cells, are the source of retinoic acid, and depletion of WT1⁺ cells reduced the frequency of GATA6⁺ macrophages in the peritoneal cavity (Buechler et al., 2019). We found that the expression of Wt1 mRNA was rather lower in Aldh1a2⁺ FRCs among the stromal populations (Fig. S2 B), suggesting WT1⁺ cells and Aldh1a2⁺ FRCs are distinct populations. Given that mesothelium is in particular the first contact with peritoneal macrophages that drift with peritoneal fluid flow, mesothelial cell–derived retinoic acid may be prone to the maintenance of peritoneal macrophage population. Collectively, distinct sources of retinoic acid, Aldh1a2⁺ FRCs and WT1⁺ mesothelial cells, may endow the distinct regulating mechanisms in omental biology.

Postn gene was selectively expressed in Aldh1a2⁺ FRCs among omental stromal populations (Fig. 2 B). Indeed, DT administration in Postn-DTR mice selectively impaired the lymphocyte recruitment to the milky spots, but not to the other lymphoid organs (Fig. 4 G and Fig. S4 F). However, Postn expression was reported in cells of mesenchymal lineage present in other tissues, such as stomach, colon, and bone (Kanisicak et al., 2016; Morra and Moch, 2011). We observed that some Postn-DTR mice, after DT injection, exhibited cecal enlargement that might be caused by tissue damage in stomach and/or colon. Therefore, systemic effects of Postn-expressing cell ablation on the milky spot formation cannot be ruled out. Additionally, this made it difficult to study long term effect of Aldh1a2⁺ FRC ablation. Although this study illuminated the role of Aldh1a2⁺ FRCs in omentum and peritoneal cavity, these immune compartments also contribute to building systemic immunity including the production of serum natural antibodies and intestinal IgA (Ansel et al., 2002; Kunisawa et al., 2007). Further study is needed regarding the roles of Aldh1a2⁺ FRCs in these immune responses.

All mice were bred in specific pathogen–free conditions, and experiments were approved by the Institutional Animal Care and Use Committee of Osaka University. Littermate controls were used for all experiments when feasible. C57BL/6J mice were purchased from CLEA Japan. CAG-EGFP mice (RBRC 00267) were obtained from Riken BioResource Research Center (Okabe et al., 1997), Mac-Gata6 KO was described previously (Okabe and Medzhitov, 2014), and these mice were backcrossed to B6. Postn-DTR mice were generated by CRISPR/Cas-mediated gene targeting in C57BL/6 zygotes. In brief, CRISPR RNA: 5′-ACG​GAG​CTC​AGG​GCT​GAA​GA-3′ was used as targeting guide (Integrated DNA technologies) and mutated DTR sequence was inserted by microinjection (Saito et al., 2001; Furukawa et al., 2006).

Rabbit monoclonal antibodies against ALDH1A2 (E6O6Q; Cell Signaling Technology), Desmin (Y66; Abcam), and CXCL12 (79018; R&D) were used. Antibodies against B220 (RA3-6B2), CD3 (17A2), CD4 (GK1.5), CD8a (53-6.7), CD11b (M1/70), CD16/32 (93), CD19 (6D5), CD23 (B3B4), CD45 (30-F11), CD105 (MJ7/18), F4/80 (BM8), IgM (RMM-1), MAdCAM-1 (MECA-367), PECAM-1 (390), PNAd (MECA-79), PDPN (8.1.1), TIE2 (TEK4), VCAM-1 (429), and Rat IgG2a,κ isotype control (RTK2758) were obtained from BioLegend. Antibodies against CD140a (APA5) and Msln (295D) were obtained from BD Biosciences and MDL, respectively. ATRA was obtained from Sigma-Aldrich.

For DT-mediated cell ablation, mice were i.p. injected with 100 ng of DT (Sigma-Aldrich). For CXCR4 inhibition, 200 μg of AMD3100 (Sigma-Aldrich) was i.p. injected every day for five times and omenta were harvested 24 h after the last injection. For inhibition of retinoic acid signaling, 500 nmol of pan retinoic acid receptor inverse agonist BMS493 (Tocris) was i.p. injected every day for three times and omenta were harvested 24 h after the last injection.

Tissues were separated, minced, and incubated in PBS containing 0.2 U/ml of Liberase TM (Roche) and 20 μg/ml DNase I (Roche) in the presence of calcium and magnesium. After digestion, the suspension was further mechanically disrupted by pipetting and filtered through 70-μm cell strainer. Peritoneal exudate cells were harvested by injecting 8 ml of PBS containing 10% fetal bovine serum and 2 mM EDTA into peritoneal cavity. Single-cell suspensions were preincubated with antibody against CD16/32 to block FcγRII/III receptors and stained on ice for 10 min with antibodies conjugated with fluorochrome. Flow cytometry was performed on an LSRFortessa (BD Biosciences), and aldehyde dehydrogenase activity was determined by ALDEFLUOR (StemCell Technologies).

Total RNA was reverse-transcribed using PrimeScript RT Master Mix (Clonetech) containing random hexamer and oligo (dT) primer. Alternatively, SMART MMLV Reverse Transcriptase (Clonetech) with oligo (dT) primer was used. Aliquots of the products were amplified using BioRad CFX96. The primers used for the mouse genes were as follows: Gapdh, 5′-GTT​GTC​TCC​TGC​GAC​TTC​AAC-3′ and 5′-CCA​GGG​TTT​CTT​ACT​CCT​TGG-3′; Rpl13a, 5′-GGC​TGA​AGC​CTA​CCA​GAA​AGT-3′ and 5′-TCT​TTT​CTG​CCT​GTT​TCC​GTA-3′; Aldh1a2, 5′-CAC​AGG​AGA​GCA​AGT​GTG​TGA-3′ and 5′-TAG​TTG​CAA​GAG​TTG​CCC​TGT-3′; Aldh1a1, 5′-GCA​CTC​AAT​GGT​GGG​AAA​GT-3′ and 5′-TTT​GGC​CAC​ACA​CTC​CAA​TA-3′; Aldh1a3, 5′-AAA​CCC​ACG​GTC​TTC​TCA​GAT-3′ and 5′-CTT​TGT​CCA​GGT​TTT​TGG​TGA-3′; Phex, 5′-GGG​TTT​ATC​CTT​GGC​TGA​GAC-3′ and 5′-AGG​TGA​ATG​CCT​CAA​GAT​GTG-3′; Postn, 5′- AAC​CAA​GGA​CCT​GAA​ACA​CG-3′ and 5′- GTG​TCA​GGA​CAC​GGT​CAA​TG-3′; Ptx3, 5′-CCT​GCT​TTG​TGC​TCT​CTG​GT-3′ and 5′-TCT​CCA​GCA​TGA​TGA​ACA​GC-3′; Cd143, 5′-CAG​TGT​CTA​CCC​CCA​AGC​AT-3′ and 5′-TTC​CAT​CAA​AGA​CCC​TCC​AG-3′; Adm, 5′-GTC​GTG​GGA​AGA​GGG​AAC​TAC-3′ and 5′-GGT​AGC​GTT​TGA​CAC​GAA​TGT-3′; Tie2, 5′-CAG​GCC​TGG​AAA​TAC​ATT​GAA-3′ and 5′-GGC​AGG​AGA​CTG​AGA​CCT​CTT-3′; Msln, 5′-AGT​CAG​GGA​GGT​TCT​GAG​GAG-3′ and 5′-AGG​GGT​CTC​TGG​AGA​TGT​GTT-3′; Lrrn4, 5′-TGA​GTT​CCT​TTG​GTC​CTT​GG-3′ and 5′-TAA​AGC​AGG​CTC​ACA​CAT​GG-3′; Upk3b, 5′-AGA​ATC​CCA​ACT​CCA​TTG​ACA​C-3′ and 5′-ATG​TAA​CGT​TTT​CCC​ATG​AAG​G-3′; Ccl19, 5′-CAC​CAC​ACT​AAG​GGG​CTA​TCA-3′ and 5′-TCT​TCT​GGT​CCT​TGG​TTT​CCT-3′; Col4a1, 5′-GAG​AGA​CAG​GAC​CCT​TTG​GAC-3′ and 5′-GGA​CAC​AGT​GGG​TCA​TCT​GTT-3′; Col14a1, 5′-CTG​CCC​ACA​CAG​CTA​GTG​AA-3′ and 5′-GCC​AGA​CCT​TCT​GTG​AGA​GG-3′; Cxcl12, 5′-GCT​CTG​CAT​CAG​TGA​CGG​TA-3′ and 5′-TAA​TTT​CGG​GTC​AAT​GCA​CA-3′; Cxcl13, 5′-TCT​GGA​AGC​CCA​TTA​CAC​AA-3′ and 5′-TTT​GTA​ACC​ATT​TGG​CAC​GA-3′; Cxcl16, 5′-TCC​GTG​AAC​TAG​TGG​ACT​GCT-3′ and 5′-GGA​AGA​GTG​GAG​TGC​TGA​GTG-3′; Ccl21, 5′-ATG​TGC​AAA​CCC​TGA​GGA​AG-3′ and 5′-TCC​TCT​TGA​GGG​CTG​TGT​CT-3′; Dpt, 5′-CCA​CTA​TGG​GGA​AGA​CAT​GG-3′ and 5′-CCT​TCA​CCC​GGA​CTT​CTG​TA-3′; Wt1, 5′- ACA​CCA​AAG​GAG​ACA​CAC​AGG-3′ and 5′- GCG​CAA​ACT​TTT​TCT​GAC​AAC-3′.

Omental cells were sorted by FACS Aria III (BD Biosciences) as indicated in the specific experiments. RNA was purified using RNeasy Micro Kit (QIAGEN) and RNAprotect Cell Reagent (QIAGEN). Sequencing libraries were constructed using NEBNext Ultra RNA Library Prep Kit for Illumina (NEB), and single-end sequencing was performed with Illumina NextSeq 500 with 76 bp reads. Gene expression heat map was created by using Cluster 3.0 and Java TreeView software. The accession numbers for the RNA-seq dataset reported in this paper are GEO GSE224315 and GSE224023.

Single-cell suspension isolated from the omenta from three mice per group (WT and Postn-DTR, 5 d after DT injection) were run on 10X chromium (10X Genomics) and then through library preparation following the recommended protocol for the Chromium Single Cell 3′ Reagent Kit (v2). Fragment size of the libraries was confirmed with an Agilent 2100 Bioanalyzer. Libraries were sequenced on Illumina NovaSeq 6000 as a single-end mode. The raw reads were processed by Cell Ranger 3.1.0 (10X Genomics). Gene expression–based clustering was performed using the Seurat R package and filtering was applied to remove low-quality cells by excluding cells expressing <200 or >7,500 unique genes, as well as cells with >25% mitochondrial gene expression. The Seurat SCTransform function was used for normalization, and data were integrated without performing batch-effect correction as all samples were processed simultaneously. The accession number for the scRNA-seq dataset reported in this paper is GEO GSE224483.

Single-cell suspensions of splenocytes from CAG-EGFP mice were prepared by homogenization of the spleen by using the frosted ends of the glass slides, followed by the filtration through 70-μm cell strainer. 5 × 10⁷ cells of splenocytes were injected i.v. or i.p., and peritoneal exudates or spleens were harvested after 24 h injection, respectively.

For immunofluorescence of whole-mount tissues, omenta were dissected and fixed with 4% paraformaldehyde and blocked with PBS containing 5% bovine serum albumin in the presence of 2 μg/ml of antibody against CD16/32 (BioLegend). Specimens were stained with Can Get Signal immunostain Immunoreaction Enhancer Solution A (Toyobo) containing antibodies and mounted with Prolong Glass (Thermo Fisher Scientific). Samples were analyzed with a confocal laser scanning microscope (Leica STELLARIS 5 or Olympus FV-3000). Images were created using Leica Application Suite X or Olympus FV31S-SW.

The milky spot size was semiautomatically determined by BZ-X800 analyzer (Keyence). Recruitment of GFP⁺ splenocytes into the milky spots was statistically analyzed by using BZ-X800 analyzer (Keyence). The number of GFP signals in full-focused images from the multiple sectioning images (0.4 μm pitch) was automatically counted by BZ-X800 analyzer.

All experiments were performed at least twice. Results were statistically analyzed by an ANOVA test or Student’s t test using GraphPad Prism 9. No statistical methods were used to predetermine sample size. Mean ± SD is shown for bar graphs, and P values of <0.05 were considered statistically significant. Sample sizes are detailed in the figure legends.

Fig. S1 shows the expression of Aldh1a1 gene among tissues, cellular characterization of ALDH-producing cells in omentum, and localization of ALDH1A2⁺ cells in the omentum. Fig. S2 shows gene expression analysis of Aldh1a2⁺ FRCs. Fig. S3 shows targeting construct of Postn-DTR allele and population analysis of omental stromal fractions. Fig. S4 includes Desmin⁺ FRC network, scRNA-seq analysis, and lymphoid organization of WT and Postn-DTR mice. Fig. S5 shows HEV populations in WT and Postn-DTR omenta and quantification of CXCL12 signal in the lumen of HEVs.

We thank members of the Okabe laboratory. We thank Dr. Miyachi (Kyoto University) and Dr. Kitano (Kyoto University) for the generation of Postn-DTR mice, Dr. Kondo (Kyoto University) for assistance with RNA-seq analysis, Genome Information Research Center (Osaka University) for RNA-seq analysis and scRNA-seq analysis, and Dr. R. Medzhitov (Yale University) for critical reading and discussion of the manuscript.

This work was supported by Japan Society for the Promotion of Science grant numbers 19H03137 and 19K22531, and Japan Science and Technology Agency grant number JPMJPR1941. Open Access funding provided by Osaka University.

Author contributions: Y. Okabe conceived the project idea and designed all experiments. Y. Okabe and T. Yoshihara performed experiments. Y. Okabe wrote the manuscript.