Single-cell atlas of the small intestine throughout the human lifespan demonstrates unique features of fetal immune cells

Abstract

Proper development of mucosal immunity is critical for human health. Over the past decade, it has become evident that in humans, this process begins in utero. However, there are limited data on the unique features and functions of fetal mucosal immune cells. To address this gap, we integrated several single-cell ribonucleic acid sequencing datasets of the human small intestine (SI) to create an SI transcriptional atlas throughout the human life span, ranging from the first trimester to adulthood, with a focus on immune cells. Fetal SI displayed a complex immune landscape comprising innate and adaptive immune cells that exhibited distinct transcriptional programs from postnatal samples, especially compared with pediatric and adult samples. We identified shifts in myeloid populations across gestation and progression of memory T-cell states throughout the human lifespan. In particular, there was a marked shift of memory T cells from those with stem-like properties in the fetal samples to fully differentiated cells with a high expression of activation and effector function genes in adult samples, with neonatal samples containing both features. Finally, we demonstrate that the SI developmental atlas can be used to elucidate improper trajectories linked to mucosal diseases by implicating developmental abnormalities underlying necrotizing enterocolitis, a severe intestinal complication of prematurity. Collectively, our data provide valuable resources and important insights into intestinal immunity that will facilitate regenerative medicine and disease understanding.

INTRODUCTION

The establishment and maintenance of mucosal immunity is empirical to intestinal homeostasis, whereas maladaptation of the process leads to inflammatory conditions, such as necrotizing enterocolitis (NEC) and inflammatory bowel disease. The development of system’s biology approaches such as single-cell ribonucleic acid (RNA) sequencing (scRNA-seq) and mass cytometry by time of flight (CyTOF), among others, has allowed us to study these processes at an unprecedented granularity, permitting the identification of novel cell types, cellular interactions, and cellular neighborhoods. In the case of mucosal immunity, building on previous pioneering work by Joseph McCune and others has led to a paradigm shift in our understanding of the developmental state and function of human mucosal immune system at birth from that of naive and immature^1–5 to one that is dominated by T cells capable of effector memory function, albeit potentially serving functions distinct from what is required of adult immune cells.^6–14 Experiments from the 80s, 90s, and early 2000s identified the presence of adaptive immune cells in the human gut at the end of the first trimester,^15–18 with an increase in T- and B-cell receptor diversity throughout gestation.^6,19 More recently, several groups demonstrated the incredible diversity of the human in utero mucosal immune system^6,20–23 and identified that by the end of the second trimester, the human small intestine (SI) is dominated by memory-like cluster of differentiation (CD)4 T cells.^6–10,24 However, the developmental trajectory and the potential functions of in utero mucosal immunity have not been fully elucidated.

Studies suggest that fetal immune cells process features distinct of their adult counterparts. For example, fetal natural killer (NK) cells, unlike mature adult NK cells, are highly responsive to cytokine stimulation and antibody-coated target cells but respond poorly to human leukocyte antigen (HLA) class I–negative target cells.²⁵ Compared with adult dendritic cells (DCs), fetal DCs differ markedly in response to allogeneic antigens, promoting regulatory T-cell (Tregs) induction.²⁶ Michaëlsson et al.¹² reported that depletion of fetal Treg cells resulted in vigorous CD69⁺ T-cell proliferation and interferon (IFN)-γ production upon antigen stimulation, whereas the depletion of adult Treg cells did not lead to the T-cell proliferation, suggesting that fetal T cells have higher regulatory capacities. Halkias et al.⁸ identified CD4⁺PLZF⁺ memory T cells in the human fetal intestine with a high capacity to rapidly produce T helper (Th)1 cytokines, such as IFN-γ and tumor necrosis factor (TNF)-α, that are essentially absent from the healthy adult intestine,⁸ suggesting that these PLZF⁺CD4⁺ T cells are a polyfunctional effector population, similar to adult intestinal CD4⁺ T cells that are poised to produce Th1 cytokines after stimulation.²⁷ Finally, fetal intestinal CD8⁺ memory T cells show impaired cytotoxic effector capacities compared with adult intestinal CD8⁺ T cells, potentially contributing to food and commensal tolerance during foreign antigen exposure after birth.²⁸ However, a systematic comparison of fetal and postnatal mucosal immune cells from a system’s biology perspective has not been undertaken. As such, emergent unique properties of the fetal mucosal immunity are largely unknown.

In the present study, we build on previous scRNA-seq datasets, both from our group, including late second trimester [21- and 23-weeks’ gestational age (GA)] and neonatal samples and publicly available SI datasets to establish a transcriptional atlas consisting of over 300,000 cells covering the human life span from 8 weeks gestation to 70 years of age, including weekly coverage of the first and second trimesters. Using this atlas, we establish the developmental trajectories of various intestinal immune cells; perform differential gene analysis between fetal and postnatal samples; and identify gene expression, signaling, and cell-cell interaction patterns that highlight the unique properties of fetal immune cells. In addition, we define altered developmental trajectories of immune cells from NEC samples that diverge from the healthy developmental patterns identified in the previously mentioned atlas.

RESULTS

Transcriptional landscape of human SI throughout the human lifespan

To assess how SI cellular heterogeneity is established and maintained, we integrated several scRNA-seq publicly available datasets from our group and others spanning the human lifespan (8 weeks GA to 70 years of age)^{21–23,29–33} (Fig. 1A and Supplementary Table 1). In total, the integrated dataset comprised 350,485 high-quality cells without significant differences in the transcript counts per cell between fetal and postnatal samples (Supplementary Fig. 1A). For optimal dataset integration, we performed batch correction using the BBKNN algorithm,³⁴ followed by Leiden clustering in Scanpy (v1.9.2)³⁵ that revealed major immune and non-immune cell types constituting the intestinal mucosal environment (Figs. 1B–D and Supplementary Table 2).

Fig. 1.

Cellular atlas of human SI across lifespan. (A) Schematic of study design and analysis workflow. (B) UMAP visualization of the cellular composition of the human SI colored by cell types. (C) UMAP plot colored by age groups, indicated by gestational weeks or postnatal years with cohort numbers. (D) Cell numbers per age groups. (E) Bar plot of relative proportions of cell types at each developmental stage. (F,G) Line plots of relative proportions of cell types at each developmental stage. The shaded area represents the variation between the donors. (H) Heatmap showing scaled mean expression score of gene sets grouped by developmental stage. Log-normalized expression for each gene set is scaled (z score). GA = gestational age; ILC = innate lymphoid cell; NEC = necrotizing enterocolitis; NK = natural killer; scRNA-seq = single cell ribonucleic acid sequencing; SI = small intestine; UMAP = Uniform Manifold Approximation and Projection.

We first wanted to determine how the global SI landscape differs between the various developmental stages by identifying differences between major cellular populations. Cellular composition of the intestinal mucosa across the developmental stages (fetal, neonatal, pediatric, and adult) was notably different. Fetal samples were enriched for mesenchymal and neuronal cells, whereas epithelial cells were more abundant in postnatal samples (Figs. 1E–G and Supplementary Fig. 1C), albeit with some skewing in some postnatal samples, such as 25–30 years and 50–55 years, that might be due to tissue processing reasons. However, we opted to keep all the data because the trends in population differences between developmental age groups were not largely affected with or without the skewed data. The proportion of immune cells in the fetal samples increased throughout gestation with the later fetal stages (21–23 weeks GA), closely resembling the postnatal samples. Although, in low abundance, myeloid cells were the first immune cells to appear in the human SI at ∼8 weeks’ gestation, followed by innate lymphoid cells (ILCs, 0.1% of total cells) and B cells (0.2% of total cells) that were detected by 9 weeks’ gestation and then by T cells (0.1% of total cells) and NK (0.2% of total cells) cells that were detected by 12 weeks’ gestation (Figs. 1E–G and Supplementary Table 3). These data are consistent with work from many groups that have demonstrated that the intestinal immune system is established early on in human development.^6,20,21,36 The SI is phenotypically heterogeneous, comprising three morphologically distinct regions: the duodenum, jejunum, and ileum.³⁷ The distribution of innate, adaptive, and innate-like immune cells varies in different segments of the adult intestine.³⁸ We broadly compared the cellular profiles between proximal regions (duodenum and jejunum) and distal region (ileum) to determine if this was also the case in early samples. Interestingly, we found that early in development, there were high similarities in cellular compositions (Supplementary Fig. 1D) and transcriptomic signatures (Supplementary Fig. 1E), where samples from different regions strongly clustered by age rather than by region, suggesting that the cells shared similar properties across location. On the other hand, adult samples exhibited slightly more regional differences (Supplementary Figs. 1D and 1E). Notably, plasma B cells dominated proximal regions in adult samples, as has been previously reported with higher numbers of immunoglobulin (Ig) A⁺ cells located in the duodenum than in later segments.³⁹ Importantly, the adult samples were collected in two different ways (either from intestinal mucosal biopsies or from mucosal tissue from transplant donors; Supplementary Table 1), where it is possible that sampling might affect the regional differences. B cells and plasma cells were very rare in prenatal samples and drastically increased in postnatal samples, especially in pediatric samples, again consistent with previous work.⁶ Given their rarity prenatally and our focus in early immune cells, B cells and plasma cells were removed from the further analysis.

To elucidate the pathways enriched for at various developmental stages, we performed pathway enrichment analysis at the various developmental stages. We identified significant upregulation of several growth- and metabolism-related pathways obtained from Gene Set Enrichment Analysis datasets (Supplementary Table 4), including proliferation, glycolysis, MYC signaling, cell cycle–related targets of E2F transcription factors (TFs), and mTOR signaling in the prenatal SI (Fig. 1H and Supplementary Figs. 1F and 1G). This was also accompanied by significant down-regulation of apoptosis-related p53 signaling in fetal cells (Fig. 1H and Supplementary Figs. 1F and 1G), consistent with the robust prenatal cellular proliferation required for the development of a mature SI. In contrast, the genes involved in inflammation were significantly upregulated in postnatal cells and remarkably higher in neonatal cells (Fig. 1H and Supplementary Fig. 1F and 1G).

Collectively, these data suggested that immune cells first inhabit the SI in the first trimester, with enrichment for growth- and proliferation-related genes prenatally.

Differential gene expression between pre- and postnatal intestinal cell types

To understand the functional differences within the major intestinal cellular population between the various stages, we first subsetted the intestinal cells into immune and non-immune populations and then compared within those populations (Supplementary Fig. 2A). Innate immune cells, including myeloid cells, NK cells, and ILCs, were more abundant in fetal samples than in postnatal samples (Supplementary Figs. 2B and 2C). Non-immune cells revealed high heterogeneity between age groups (Supplementary Figs. 2B and 2C).

Hierarchical clustering computed by Pearson correlation-based distance demonstrated that fetal first trimester immune cells were most similar to immune cells from the second trimester and distinct from postnatal samples (Supplementary Figs. 2D and 2E). As such, we combined the samples into fetal or prenatal samples and postnatal samples for further differential analysis.

We first examined the differences between prenatal and postnatal myeloid cells (Supplementary Table 5; myeloid cells). Among myeloid cells, tissue-resident–associated markers, SPP1, F13A1, and FOLR2,⁴⁰ and those associated with yolk sac–derived macrophages, LYVE1, CSF1R, CX3CR1, ADGRE1 (encoding F4/80),⁴¹ were upregulated in fetal cells (Fig. 2A). In addition, fetal cells expressed higher levels of anti-inflammatory genes, such as MAF⁴² and CD28⁴³ (Fig. 2A). In contrast, genes commonly associated with inflammation, S100A8, S100A9, CXCL8, IL6, IL1A, and IL18,⁴⁴ were higher in postnatal cells (Fig. 2A).

Fig. 2.

Differential expression analysis between fetal and postnatal samples. (A) Volcano plots showing differentially expressed genes between fetal immune cells (upregulated genes) and postnatal immune cells (down-regulated genes). (B) UMAP plot of HPGDS expression in T cells. (C) The frequency of HPGDS+ T cells at each developmental stage from scRNA-seq dataset. Each dot represents an individual donor. Data are presented as mean ± SEM. ** p < 0.01; *** p < 0.001. (D) Immunofluorescence staining of CD3 (green), HPGDS (red) on fetal (21 weeks gestational age) and adult small intestinal tissue. The yellow square highlights the region magnified on the right. Yellow arrows represent HPGDS⁺CD3⁺ T cells. <Scale bars are 90 μm> (E) Heatmap of TF regulons identified by SCENIC analysis, plotted by score of centered AUC for each regulon within cell types across age groups. (F) Comparison of cell-cell interaction strength among fetal and postnatal cells. (G) Heatmap of cell-cell interaction strengths in fetal cells compared to postnatal cells. The top colored bar plot represents the sum of column of values displayed in the heatmap (incoming receptor signaling). The right colored bar plot represents the sum of row of values (outgoing ligand signaling). AUC = area under the curve; ILC = innate lymphoid cell; NK = natural killer; scRNA-seq = single-cell ribonucleic acid sequencing; SEM = standard error of the mean; SI = small intestine; TF = transcription factor; UMAP = Uniform Manifold Approximation and Projection.

We next focused on the lymphoid compartment, including T cells, NK cells, and ILCs. Among T cells, naive T-cell markers, such as CCR7, SELL, KLF2, S1PR1, LEF1, and TCF7,⁴⁵ were upregulated in fetal T cells, whereas memory signatures, especially those associated with CD8⁺ T cells, were enriched for in postnatal samples, with higher expression of cytotoxic-associated genes including granzymes (GZM) GZMA, GZMB, GZMH, PRF1, and NK cell receptors (KLRD1, KLRC1, KLRC2, KLRC3, KLRK1). Similarly, memory-like associated genes, such as CCL5, CCL4, and IFNG, and exhaustion markers, including LAG3, CD160, and TIGIT, were enriched for in the postnatal T cells (Fig. 2A and Supplementary Table 5; T cells). Interestingly, CD28, costimulatory molecule required for T-cell activation, was enriched for in prenatal T cells (Fig. 2A). In addition, interferon (IFN)-stimulated genes (ISGs) such as ISG15, IFI16, MX1, GBP1, GBP2, GBP4, and IRF2 had higher expression in fetal T cells, even in the setting of lower overall IFNG expression by fetal T cells than postnatal T cells (Fig. 2A). Consistent with previous reports, there were other unique features of fetal T cells, including ZBTB16 (encoding PLZF) expression⁸ that was specifically enriched for in the fetal intestine (Fig. 2A and Supplementary Fig. 2F). Furthermore, we identified that the expression of a prostaglandin associated gene, HPGDS, was almost exclusively limited to fetal T cells compared with postnatal T cells (Figs. 2A and 2B), where there was a significant increased frequency of HPGDS⁺ T cells in fetal samples compared with pediatric and adult samples (Fig. 2C). This finding was further confirmed by the enrichment of HPGDS and CD3 protein co-expression in fetal SI compared with postnatal samples (Fig. 2D and Supplementary Fig. 2G).

Next, focusing on NK cells that, in contrast to CD8⁺ T cells which had increased cytotoxic activity postnatally, had enrichment of GZM genes, GZMA, GZMH, GZMK, and GZMM and well-known NK cell effector function genes IFNG, PRF1, CCL3, CCL4, XCL1, and XCL2 in fetal rather than postnatal cells (Fig. 2A and Supplementary Table 5; NK cells). These data may suggest a critical function for NK cells in the developing intestine during a period of relatively low abundance of CD8⁺ T cells. However, a higher expression of resident memory markers, CXCR6, IL7R, and ANXA1,⁴⁶ were detected in postnatal rather than fetal NK cells.

Overall, ILCs were more abundant in the prenatal samples (Supplementary Figs. 2A–C and Supplementary Table 5; ILCs) and prototypical markers of lymphoid tissue inducer (LTi) cells, LTA, LTB, and CCR6, that are necessary for ILC function in lymphoid organogenesis⁴⁷ and tissue residency-associated genes, CD69, CXCR4, CXCR5, were upregulated in fetal cells, whereas inhibitory molecules, LAG3, CD226, HAVCR2 (encoding TIM-3), PDCD4, CTLA4, LAYN, were highly expressed in postnatal cells, indicating their activated state⁴⁸ (Fig. 2A).

Furthermore, we used singel-cell regulatory network inference and clustering (SCENIC)⁴⁹ to detect the differences in TF regulatory networks across immune cell types in the fetal and postnatal groups (Fig. 2E). Distinct regulons (TFs and their target genes) were observed across fetal and postnatal groups in a cell type–specific fashion. Notably, the regulons of CCAAT/C/EBP TF family members, CEBPA, CEBPB, and CEBPD, were enriched in myeloid cells and especially upregulated in fetal myeloid cells. These C/EBPs can bind to the promoter regions of numerous genes expressed in myeloid cells, such as CD14 and CSF3, to regulate myeloid cell differentiation.⁵⁰ EOMES regulon, the key checkpoint of NK cell maturation,⁵¹ was enriched in fetal NK cells, whereas T-cell development-associated regulons of TCF7, LEF1, and SOX4^52,53 were highly expressed in fetal T cells.

To determine whether there were global differences in intestinal non-immune cells, we next examined their transcriptional differences. In the endothelial cells, FCR γ chain FCER1G and PRND were upregulated in fetal cells. PRND encodes the glycosylphosphatidylinositol-linked protein, doppel, that promotes angiogenesis,⁵⁴ whereas venous cell markers, ACKR1, VWF, and CPE, and migration-associated genes, CD9 and S1PR1, were enriched in postnatal endothelial cells (Supplementary Fig. 2H and Supplementary Table 5; endothelial cells). In the epithelial compartment, genes associated with earlier developmental stages and stemness, including stem cells (LGR5 and ASCL2), proximal progenitors (FGG and FGB), proximal early enterocytes (TF, AFP, and VTN), and distal early enterocytes (SLC26A2 and CKB) were highly expressed in fetal cells, whereas mature enterocyte markers such as proximal enterocytes, ACE and FABP2, and distal enterocytes, FABP6 and IL32, were enriched for in postnatal cells. Similarly, mucin-related genes (MUC20, MUC2, MUC17, MUC4, MUC12, MUC3A, MUC1, and MUC6) were also higher in postnatal cells, suggesting that the mucin layer covering the intestinal epithelium develops postnatally (Supplementary Fig. 2H and Supplementary Table 5; epithelial cells). Moreover, collagen genes, COL11A1, COL6A6, COL21A1, COL9A1, and COL23A1, that are involved in the epithelial-mesenchymal transition⁵⁵ were enriched in fetal mesenchymal cells, whereas chemokines such as CXCL14, CXCL1, CXCL6, CXCL2, CXCL3, CX3CL1, CCL11, CCL2, CCL8, CCL13, CCL19, and CCL21 were enriched for in postnatal compared with prenatal mesenchymal cells. These chemokines and their receptors regulate migration and residence of immune cells, suggesting that there continues to be an influx of immune cells postnatally⁵⁶ (Supplementary Fig. 2H and Supplementary Table 5; mesenchymal). Finally, fetal neuronal cells were enriched for inhibitory motor neuron associated genes, such as GAL, VIP, and ETV1, and an excitatory motor neuron associated gene, BNC2, that respectively evoke muscle contractions and relaxations to control peristalsis.⁵⁷ Although, glial cells associated genes (CRYAB, FGL2, GFRA3, and RXRG) were higher in postnatal samples (Supplementary Fig. 2H and Supplementary Table 5; neural cells). Our data supported the finding that enteric neurogenesis occurs in early life and declines with aging, in contrast to gliogenesis.⁵⁸ The glial cell network of the gut mucosa develops after birth in mice.⁵⁹

To determine whether there were developmental stage–specific cellular interactions in the SI, we performed receptor-ligand pairing analysis in prenatal versus postnatal samples. Overall, there were more than double the strength of interactions between fetal cells and postnatal cells (Fig. 2F). In particular, fetal non-immune cells and immune cells displayed stronger ligand sending signals and receptor receiving signals, respectively, suggesting that a highly active cell-cell crosstalk supports the SI maturation prenatally (Fig. 2G and Supplementary Table 6). For example, fetal endothelial cells had increased interactions with all other cell types, with the highest strength of interactions with ILCs, likely secondary to ILC’s importance in recruitment of immune cells into the intestine and gut-associated lymphoid tissue generation during this period. Furthermore, fetal mesenchymal cells had the strongest ligand-receptor interaction with all cell types but in particular with T cells and ILCs demonstrating their importance in intestinal immune development as previously suggested⁶⁰ (Fig. 2G and Supplementary Fig. 3A). These interactions were dominated by macrophage inhibitory factor -CD74/CXCR4 and -CD44/CD74 interactions on NK cells and ILCs, promoting their survival and proliferation^61,62 (Supplementary Fig. 3B), consistent with the observed upregulation of proliferation signals in fetal cells (Fig. 1D). We also detected MDK/PTN-NCL-driven interactions with T-cell signaling that also promotes cell migration and survival.⁶³ Whereas postnatally, proinflammatory chemokines, CXCL1, CCL2, and CCL11, were enriched in mesenchymal cells and interacted with ACKR1 on endothelial cells (Supplementary Fig. 3A). Finally, the interaction between SPP1, mainly expressed in neuronal cells, and adhesion molecules (ITGA5, ITGB1, ITGA4, and CD44) that regulate cell migration, polarization, and activation⁶⁴ were upregulated on numerous postnatal cells (Supplementary Fig. 3B).

Myeloid cells change across gestation

To understand the function of fetal immune cells in more granularity, we subsetted just the fetal immune cells. We first focused on the individual immune lineages in more detail. To define the developmental trajectory of myeloid cell that were the first immune cells to inhabit the intestinal mucosa as stated previously, we examined the various subtypes of myeloid cells across the first and second trimester in the fetal SI. Numerous macrophages, monocyte, and DC subtypes were identified consistent with previous data^6,65 (Figs. 3A and 3B, Supplementary Fig. 4A, and Supplementary Table 7). Macrophages were the predominant myeloid population throughout gestation with diverse subsets (Supplementary Fig. 4B). Seven different subsets were identified based on their transcriptional profiles, including LYVE1⁺ CX3CR1⁺, LYVE1⁺ CX3CR1⁻, CX3CR1⁺CLEC10A^hi, CX3CR1⁻CLEC10A^lo, CX3CR1⁻CLEC10A^hi, MMP9^int, and MMP9^hi (Fig. 3A and Supplementary Fig. 4A), and LYVE1⁺ macrophages, also with high expression of CSF1R,^41,66,67 FOLR2, CD163, and F13A1,⁶⁸ indicative of their yolk sac origin⁶⁹ (Fig. 3C). Interestingly, even during the fetal development, not all macrophages expressed LYVE1, suggesting that even in utero, multiple waves of hematopoiesis contribute to the diversity of intestinal macrophages (Supplementary Figs. 4A and 4C). Consistent with our understanding of LYVE1⁺ macrophages, their transcriptional analysis revealed a tissue-resident phenotype with high expression of residency markers, LYVE1 (associated with perivascular macrophages),⁷⁰ GAS6, SPP1, TIMD4, and COLEC12⁷¹ (Fig. 3C). These cells were abundant in fetal samples but rare in postnatal samples (Supplementary Fig. 4C). LYVE1⁺ macrophages were subclustered into two subsets by the expression of fractalkine receptor, CX3CR1, canonical LYVE1⁺ CX3CR1⁺, and non-conventional LYVE1⁺ CX3CR1⁻ macrophages (Supplementary Fig. 4D). The upregulation of transcriptional activators (KLF2 and KLF6),⁷² activation/immune recruitment genes (CCL4, CCL3, CXCL8, NFKBIA, NFKBIZ, IL1B, DUSP1, FOS, JUN, and CD83),^56,73 and increased the major histocompatibility complex class II (MHCII) gene set (Supplementary Table 4) expression score in the non-conventional LYVE1⁺ macrophages suggested that the these macrophages may facilitate cellular communications and recruitment (Figs. 3C and 3D). A similar pattern of fractalkine receptor expression was observed in another set of macrophages that were CLEC10A^hi (encoding lectin receptor, MGL) (Fig. 3C). Overall, CX3CR1⁺ macrophages were detected earlier in gestation than CX3CR1⁻ macrophages, indicative of their precursor potential as reported⁷⁴ (Fig. 3E). CX3CR1⁺ CLEC10A^hi macrophages were marked by ADGRE1 (encoding F4/80) (Fig. 3C) and appeared early in gestation but were enriched later in gestion, likely representing a transitional state (Fig. 3E). It has been reported that CXCR3⁺F4/80⁺ cells are present in the lamina propria in mice.⁷⁵ To analyze the differentiation trajectory of various macrophage populations, we used potential of heat diffusion for affinity-based trajectory embedding (PHATE), an unsupervised, nonlinear dimensionality reduction method to preserve the local and global structures of the data⁷⁶ (Fig. 3B and Supplementary Fig. 4E). An additional trajectory analysis was performed using dynamic scVelo to disentangle cell kinetics (Supplementary Figs. 4F and 4G), which used the ratio of spliced to unspliced reads to infer RNA velocities.⁷⁷ The RNA velocities were projected onto the forced-directed graph that indicates developmental trajectories in a lineage structure (Supplementary Fig. 4F). The inferred latent time represented the cell’s internal clock underlying cellular biological processes as the cells were differentiating (Supplementary Fig. 4G). The latent time was grouped based only transcriptional dynamics to indicate the direction of motion. The trajectory analyses and differential abundance analyses (Fig. 3E) revealed that the two CX3CR1-negative macrophage populations (LYVE1⁺CX3CR1⁻ and CX3CR1⁻CLEC10A^hi, respectively) grouped together because the terminal branch of one of the trajectories and both appeared later in gestation, determined by the higher positive log fold change (FC) values (Fig. 3E), indicative of their more mature states. Finally, CX3CR1⁻CLEC10A^lo macrophages had the highest MHCII gene set expression score (Fig. 3D), implying the highest antigen presenting potential.

Fig. 3.

Myeloid cells across gestation. (A) UMAP visualization of fetal myeloid cell types. (B) The PHATE visualization colored by cell types. (C) Dot plot of selected genes expression in macrophage and monocyte cell types. Here and in later dot plot figures, color represents normalized mean expression of marker genes in each cell type, and size indicates the proportion of cells expressing marker genes. (D) Violin plot of MHCII gene set expression scores in macrophages. (E) Violin plot of log FC in cell abundance across gestational ages. Higher positive value means that cells were more enriched in later gestation and vice versa. (F) Representative multiplex RNAscope combined with immunofluorescence images of fetal small intestine (21 weeks gestational age). The yellow square highlights the region magnified on the right. Red arrows represent NC slan⁺ monocytes identified by MMP9⁺S100A4⁺IL3RA^low, pink arrows represent pDC (IL3RA⁺S100A4⁺MMP9⁻), yellow arrows represent MMP9⁺ macrophages (MMP9⁺IL3RA⁺S100A4⁻). <Scale bar represents 90 μm> Here and in later RNAscope images, RNA signals were visualized as dots. (G) Dot plot of selected genes expression in dendritic cell types. cDC = conventional dendritic cell; FC = fold change; IL = interleukin; MMP = matrix metalloproteinase; NC = nonclassical; pDC = plasmacytoid dendritic cell; PHATE = potential of heat diffusion for affinity-based trajectory embedding; RNA = ribonucleic acid; UMAP = Uniform Manifold Approximation and Projection.

Matrix metalloproteinase (MMP) is produced by numerous cell types, including macrophages,⁷⁸ degrade the extracellular matrix components, and are critical for cellular migration and tissue remodeling. We identified two groups of SI macrophages that were enriched in MMPs, including MMP1, MMP9, MMP10, and MMP12, designated as MMP9^int and MMP9^hi macrophages (Fig. 3C). MMP12 has been shown to be critical for macrophage migration in the lamina propria. as well as transepithelial migration and wound healing.⁷⁹ These macrophages appeared late in gestation (Fig. 3E) and group close to mature DCs in the PHATE analysis (Fig. 3B), suggestive of their highly mature state.

In addition to macrophages, monocytes were also present in fetal SI. Monocyte precursors (mono-neutrophils) characterized by the expression of neutrophils-associated genes (MPO, PRTN3, and AZU1) and proliferation gene, MKI67, were detected early in gestation (Fig. 3E and Supplementary Fig. 4A). Interestingly, we observed the expression of MS4A3 (Fig. 3C), which is specifically and transiently expressed by granulocyte-monocyte progenitors in the bone marrow,⁸⁰ suggesting that monocyte development, at least, partially, can occur in the fetal SI. Two subsets of non-classical (NC) monocytes were characterized by the expression of FCGR3A (encoding CD16)⁸¹ (Supplementary Fig. 4A). These monocytes were distinguished by their expression of SECISBP2L (encoding slan) (Fig. 3C), a phenotypic marker that has been mostly used to separate intermediate monocytes (slan⁻) from nonclassical monocytes (slan⁺).⁸² We found that the NC slan⁻ monocytes were present earlier in gestation (Fig. 3E), reaffirming their intermediate phenotype.

Interestingly, MMP9^int and MMP9^hi macrophages together with NC slan⁺ monocytes formed a trajectory inference away from other macrophages, indicating their unique phenotype and potentially suggesting that the NC slan⁺ monocytes give rise to MMP9⁺ macrophages (Fig. 3B and Supplementary Figs. 4E–G). These cells shared a similar expression pattern of cytokine receptors, IL7R and IL3RA, and interleukin (IL)-1 signaling-related genes (IL1R1, CXCL8, IRAK2, IRAK1, TANK, and MAPK13), likely contributing to host defenses against pathogens.⁸³ The genes involved in the regulation of angiogenesis, LGALS3,⁸⁴ GPNMB,⁸⁵ CYP1B1,⁸⁶ THBS1,⁸⁷ were also highly expressed in these cells, pointing to their role in mucosal blood vessel formation (Fig. 3C). Finally, these cells highly expressed Schwann cell-associated genes, EMP1 and PMP22⁸⁸ (Fig. 3C), suggestive of Schwann cell phagocytoses during active remodeling of the mucosa occurring throughout development. Similar populations were also reported in human colonic macrophages.⁷¹ Fluorescence in situ hybridization (RNAscope) staining combined with immunofluorescence (IF) confirmed the presence of MMP9⁺ macrophage (MMP9⁺IL3RA⁺S100A4-) cells and NC slan⁺ monocytes (S100A4⁺ MMP9⁺IL3RA^low) in the fetal SI (Fig. 3F).

Diverse DCs were also detected in the fetal samples with heterogeneity in their transcriptional profiles (Fig. 3A). Among these, complimentary to the data demonstrating CD4⁺ T-cell predominance in fetal SI (Fig. 4A), type 2 conventional DCs (cDC2) cells marked by the expression of CD1C, CD1E, ITGAM (encoding CD11b), FCER1A, and CLEC10A⁸⁹ were the predominant DC population throughout gestation (Figs. 3E and 3G). This is in direct contrast to murine neonatal intestine that has recently been shown to be dominated by cDC1 cells, with delayed maturation of cDC2 resulting in delayed T-cell priming.⁹⁰ cDC2s were unexpectedly closely connected to NC slan⁻ monocytes in the trajectory tree (Figs. 3B and Supplementary Fig. 4E), possibly suggesting that the cluster represented a mixture of cDC2s with monocyte-derived DCs because we observed weak expression of monocyte-derived DC genes, such as MRC1 (encoding CD206), FCGR3A, FCGR1A (encoding CD64), CD14, and CD209,⁹¹ in the cluster (Fig. 3G). However, the absence of the complement receptor C5AR1 (encoding CD88) (Supplementary Fig. 4A) suggested that the majority of cDC2s and cDC1s were of non-monocytic origin.⁹¹ Interestingly, we found endogenous expression of IFNL1 (IFN-λ) in cDC1 cells (XCR1⁺CLEC9A⁺), although at low levels (Fig. 3G). It has been reported that IFN-λ is produced by leukocytes but not epithelial cells in the gut mucosa at a steady state, contributing to anti-viral mucosal defenses,^92,93 while the IFN-λ production by DCs was likely playing a similar role. Anti-inflammatory pattern-recognition receptor TLR10⁹⁴ was highly expressed in cDC1 and cDC2 (Fig. 3G), potentially modulating the balance between pro- and anti-inflammatory responses in utero. Plasmacytoid DC (pDC) and migratory DCs appeared later in gestation (Fig. 3E) and segregated out in the trajectory tree by their unique mature migratory signatures, CCR7, CXCR4, CXCR3, and SELL (Figs. 3B and 3G and Supplementary Fig. 4E). Migratory DCs revealed immunoregulatory signatures with the expression of CD274 (encoding PD-L1), PDCD1LG2 (encoding PD-L2), and CD200. They also expressed high levels of the Th2 response genes IL4R, IL4I1, CCL17, CCL22 and BCL2L1, as well as activation markers CD80, CD83, and CD86^26,95 (Fig. 3G). The presence of mature DCs supports previous observations that fetal DCs are capable of antigen presentation.²⁶ In addition, we observed a strong expression of IL15 and its receptor, IL2RG, in migratory DCs. It has been reported that IL-15 mediates the survival of mature DCs and, in turn, affects the memory fate of CD8⁺CD44^hi T cells.⁹⁶

Fig. 4.

T-cell compartment across gestation. (A) UMAP visualization of fetal T cells, NK cells, and ILCs cell types. (B) Violin plot of log-fold change in cell abundance across gestational ages. Higher positive value means that cells were more enriched in later gestation and vice versa. (C) The proportion of overall T cells, NK cells, and ILCs across gestational ages. Data are represented as mean ± SEM. (D) Dot plot of selected genes expression in ILC3 cell types. (E) Dot plot of selected genes expression in NK cell types. (F) Violin plot of cytotoxic gene set expression scores in NK cell types. (G) Heatmap showing scaled mean expression of selected genes, grouped by CD4 and CD8 memory T cells across developmental stages. (H) Heatmap of TF regulons identified by SCENIC analysis, plotted by score of centered AUC for each regulon in memory T cells across age groups. (I) The respective proportion of fetal-like T cells that expressed genes ID3, SMC4, PLAC8, and adult-like T cells that expressed genes JUNB, FOS, S100A4 in memory T cells. AUC = area under the curve; IL = interleukin; ILC = innate lymphoid cell; NK = natural killer; NKT = natural killer T cell; SEM = standard error of the mean; TF = transcription factor; Tmem = memory T cell; Treg = regulatory T cell; UMAP = Uniform Manifold Approximation and Projection.

The T-cell compartment is dynamic across gestation

The subclustering of fetal T cells, NK cells, and ILCs revealed numerous diverse cell types (Fig. 4A, Supplementary Fig. 5A, and Supplementary Table 8). Differential analysis across gestation revealed that proliferation- and glycolysis-associated genes, such as PCNA, CDK2, CDK4, LDHA, and ALDOA, were upregulated early in gestation, indicative of their increased self-renewal potential. However, IFN signaling and MHCI genes, such as IFI44L, ISG15, and HLA-A, were enriched later in gestation (Supplementary Fig. 5B).

We then assessed our dataset for developmental trajectories of the various lymphocyte populations. Consistent with their previous known role in lymphoid development, ILCs were the earliest lymphoid cells present in the SI and were enriched earlier on in gestation and gradually decreased in their abundance across gestation (Figs. 4B and 4C and Supplementary Fig. 5C). ILC3s were subdivided into four subsets by the expression of NCR2 (encoding NKp44) and CCR6 (Supplementary Fig. 5D). NCR⁺CCR6⁺ ILC3 and NCR⁺CCR6^lo ILC3 displayed a phenotype reminiscent of LTi cells with a high expression of RORC, KIT, IL7R, TNF, ITGB7 (encoding beta chain of α4β7 integrin), LTA, LTB, and TNFSF11 (encoding RANK)²¹ and the chemokine receptor CXCR5, which binds CXCL13 secreted by mesenchymal cells for the initiation of lymph node formation⁹⁷ (Fig. 4D). Indeed, CXCL13 expression was detected in our atlas in fetal mesenchymal cells (Supplementary Fig. 5E). It has been reported that CCR6⁺ LTi cells can cluster in cryptopatches of the lamina propria via the CCR6 ligand, CCL20.^98,99 CCL20 was expressed by a significant proportion of NCR⁺CCR6^lo ILC3s and a small proportion of NCR⁺CCR6⁺ ILC3s (Fig. 4D). In addition, the differential analysis revealed that these NCR⁺ ILCs had a lower expression of a residency gene, CXCR6, higher expression of KLF2, a gene associated with migration, and increased IL1R1, a gene that drives the steady state production of CSF2, a key determinant of myeloid lineage differentiation.¹⁰⁰ Interestingly, NCR⁻CCR6⁺ cells that appeared in earlier gestation (Fig. 4B), although in low abundance, uniquely expressed NK cell signatures, NKG7, GZMA, PRF1, CD244, as well as CD69 and IL17A (Fig. 4D), indicating that these cells were more likely to be precursors that could differentiate into ILCs or NK cells. This was supported by the trajectory inference where NCR⁻CCR6⁺ cells were clustered with other ILC3 and slightly connected to ILC2 and NK cells (Supplementary Fig. 5F). The differential abundance analysis demonstrated a shift from CCR6⁺ ILC3 that were detected in early gestation to CCR6^lo ILC3 later in gestation again, suggesting that CCR6 expression is associated with a precursor phenotype¹⁰¹ (Fig. 4B).

Diverse NK cell populations were also identified with relatively stable abundance across gestation (Fig. 4C). Consistent with previous reports, a small proportion of CD16⁺ NK cells (FCGR3A^hi NK) and a large proportion of CD56⁺ (encoded by NCAM1) (CD69^hi NK and ITGA1^hi NK) were identified in the fetal SI (Fig. 4A). FCGR3A^hi NK cells displayed a potent effector phenotype with the highest expression score for genes associated with cytotoxicity (Supplementary Table 4), followed by CD69^hi and ITGA1^hi NK cells (Figs. 4E and 4F). ITGA1^hi NK cells exhibited a tissue-resident phenotype, with the highest expression of tissue residency markers, ITGA1 (encoding CD49a), ITGAE (encoding CD103), and CXCR3; in contrast, FCGR3A^hi NK cells highly expressed a migration gene KLF2 (Fig. 4E).

T cells were the last major immune cell type to appear in utero and gradually increased across gestation (Figs. 4B and 4C). These included CD4 and CD8 naive and memory T cells, double negative T cells, NKT cells, and γδ T cells (Fig. 4A). Expression of gut homing receptor CCR9 is essential for the preferential homing of T cells to the gut¹⁰² so we further investigated CCR9 expression in fetal T cells. We found that CCR9 was detected in fetal T cells, particularly, in the CD8 memory and γδ T cells but was relatively low in naive T cells (Supplementary Fig. 5G). This was validated by the RNAscope combined with antibody staining, where CCR9 staining was enriched in CD3⁺CD45RA⁻ memory T cells and lower inCD3⁺CD45RA⁺ naive T cells (Supplementary Fig. 5H). Our findings are consistent with published data suggesting that outside of the thymus, CD8 T cells express more CCR9 than CD4 T cells.^103,104Moreover, a subset of Tregs and CD4 cells do express CCR9 and this was also observed in this dataset.¹⁰⁵ Increased expression of CCR9 on memory CD8 T cells is observed upon the priming of CD8 T cells in the mesenteric lymph nodes (or Peyer’s patches), leading to upregulation of CCR9 and migration of CD8 T cells to the SI.^106–108

Among memory T cells, CD4 memory T cells were the most abundant and were especially so during later stages of gestation (ranging from 1.6% early on to 67.9% later in gestation), consistent with previous reports⁶ (Supplementary Figs. 5C and 5I). In contrast, complimenting the low abundance of cDC1s in fetal SI (Supplementary Fig. 4B), CD8 memory T cells were present at a much lower frequency in fetal SI (0.8%–9.3%). This contrasts with what was observed in postnatal samples, where CD8 T cells were more abundant (Supplementary Fig. 5I). Fetal CD4 memory T also revealed the strongest T cell receptor (TCR) signaling signature that was calculated using associated genes (Supplementary Table 4), suggestive of active TCR signaling (Supplementary Fig. 5J). Subclustering of memory CD4 T cells into helper subsets by the expression of CXCR3 and CCR6 revealed that Th1 (CXCR3⁺CCR6⁻) cells were the predominant subset in utero, as previously described,⁶ followed by a smaller proportion of Th17 (CXCR3⁻CCR6⁺), Th1/17 (CXCR3⁺CCR6⁺), and Th2 (CXCR3⁻-CCR6⁻) cells (Supplementary Figs. 5K–M).

The differential analysis of CD4 and CD8 memory T cells gene expression across the lifespan identified distinct fate-determining signatures unique to early-life (fetal and neonatal) and later-life (pediatric and adult) T cells (Fig. 4G and Supplementary Table 9). Consistent with Farber group’s finding,¹⁰⁹ stem-like TFs, ZBTB16, LEF1, SOX4, and IKZF2 and regulons of these target genes were differentially enriched in early memory T cells (Fig. 4H). In addition, chromatin modifier, SMC4,¹¹⁰ T-cell response regulator, PLAC8,¹¹¹ early T-cell development-associated gene, SOS1,¹¹² costimulatory molecule, CD81,¹¹³ and T-cell activation gene, CD38,¹¹⁴ as well as PIK3RA1 (phosphoinositide-3-kinase, regulatory subunit 1 alpha) were upregulated in early-life memory T cells (Fig. 4G). Moreover, we observed a higher expression of ZNF683 (encoding Hobit, the homolog of Blimp-1) in fetal CD8 memory T cells, whereas the expression of PRDM1 (encoding Blimp-1) was higher in later-life memory T cells. The inhibitor of differentiation (ID) family of TFs, ID3 and ID2, were almost exclusively expressed in early-life or late-life samples, respectively (Fig. 4G). Conversely, activation-associated genes, activator protein 1 family (JUN, JUNB, FOS, and FOSB),¹¹⁵ along with their target gene regulons (Fig. 4H), CD69, inflammatory gene S100A4,¹¹⁶ cytotoxic genes PRF1, GAMA, GZMH, GZMB, and exhausted genes CD160, LAG3, were enriched in later-life memory T cells (Fig. 4G). Consistently, the negative regulator of T cell activation JUND regulons was enriched in early memory T cells¹¹⁷ (Fig. 4H). We further checked the proportion of early (fetal-like) and late (adult-like) memory T cells, characterized by the three of highest expressed genes, respectively (Fig. 4I). Early memory T cells (quantified by the expression of ID3, SMC4, and PLAC8) were highest during the fetal stage and decreased rapidly after birth. However, these rare cells could be detected even in the adult intestine indicating their long-lived potential and giving credence to the layered immune hypothesis^118–120 or suggesting that fetal-like cell can be produced later in life. In contrast, the “adult” signature (quantified by the expression of JUNB, FOS, and S100A4) was low at birth and increased rapidly thereafter, with highest expression in adult tissue. Interestingly, the proportion of adult cells was higher in fetal samples than in neonatal samples, perhaps suggesting that these are shorter-lived memory cells.

NEC is associated with abnormal SI immune development

Alterations in the gut immune system are thought to play an important role in the pathogenesis of inflammatory disease, including NEC. To investigate if the inflammation observed in NEC is related to defects or alterations to normal SI mucosal immune development, we integrated our previous dataset^29,30 consisting of 6972 cells from the NEC SI samples with the current dataset (Figs. 5A and Supplementary Figs. 6A and 6B). The differential abundance analysis revealed that the frequency of major T-cell compartments (T/NK/ILCs) and myeloid cells were comparable, whereas non-immune cells were relatively decreased in NEC samples compared with fetal and neonatal samples under consideration of factors of age and tissue regions (Supplementary Fig. 6C). Among immune cells, consistent with our previous observations, there were profound immune alterations associated with NEC, including expansion of some myeloid populations²⁹ (Fig. 5B and Supplementary Figs. 6D and 6E). We had previously identified a subgroup of monocytes/macrophages that were enriched for in NEC samples as inflammatory monocytes/macrophages that are double-positive for CD16 and CD163 and infiltrate the intestine from the peripheral blood.³⁰ In the longitudinal developmental atlas, we observed a unique group of monocytes that we termed NC slan⁺ monocytes that were also double-positive for CD16 (encoded by FCGR3A) (Supplementary Fig. 4A) and CD163 (Fig. 3C) and likely represent similar cells to the ones we previously identified to be enriched in the NEC samples. During normal fetal development, the NC slan⁺ monocytes, MMP9⁺ macrophages. and migratory DCs start to ingress into the SI in late 2^nd trimester (Fig. 3E). NC slan⁺ monocytes were strikingly enriched in the cell frequency and cell numbers in the NEC samples (Fig. 5C and Supplementary Figs. 6D–G), suggesting that although ingress of these cells into the SI is part of normal development, in infants who develop NEC, there is an increased infiltration of these cells. Conversely, MMP9⁺ macrophages and migratory DCs were significantly decreased in the NEC samples compared with fetal and neonatal samples (Fig. 5C and Supplementary Figs. 6D–G). Consistently, our previously mass CyTOF data revealed that CCR7⁺CD103⁺ DCs, which closely resembled the designated migratory DCs by the unique expression of CCR7 (Fig. 3G), were significantly reduced in NEC compared with fetal samples and with a similar trend compared with neonatal cases.³⁰ Our previous CyTOF and imaging mass cytometry (IMC) data further supported the enrichment of monocytes in NEC samples.³⁰

Fig. 5.

Transcriptional signatures of immune cells in NEC. (A, B) UMAP visualization of all cell types (A) and immune cell types (B) after adding NEC samples. (C) Violin plot of log FC in cell abundance of immune cells under two factors (age and tissue regions). Higher positive value means that cells were more enriched in NEC samples and lower negative values means that cells were enriched in fetal and neonatal samples. (D) Volcano plots showing differentially expressed genes between combined NC slan⁺ monocytes, MMP9⁺ MΦ, and migratory DCs in NEC samples (upregulated genes) and fetal and neonatal samples (down-regulated genes). (E) Representative multiplex RNAscope combined with immunofluorescence images of fetal (21weeks gestational age) and NEC small intestine. The pink arrows represent NC slan⁺ monocytes (IL1B⁺IL3RA^lowS100A4⁺). <Scale bar represents 90 μm> (F) Heatmap showing mean expression of module 1 gene set and selected genes of module 1 with dendrogram in naive T cells across developmental stages and NEC. (G) Heatmap showing scaled mean expression of selected module gene sets and selected genes from each module with dendrogram in CD4⁺ memory T cells across developmental stages and NEC. cDC = conventional dendritic cell; DC = dendritic cell; FC = fold change; ILC = innate lymphoid cell; NC = nonclassical; NEC = necrotizing enterocolitis; NK = natural killer; pDC = plasmacytoid dendritic cell; RNA = ribonucleic acid; Tmem = memory T cell; Treg = regulatory T cell; UMAP = Uniform Manifold Approximation and Projection.

To determine whether, in addition to alteration of abundance, the transcriptional program of myeloid cells was also altered by NEC, we identified differentially expressed genes (DEGs) in the combined NC slan⁺ monocytes, MMP9⁺ macrophages, and migratory DCs between the combined fetal and non-NEC neonatal samples and NEC samples. We found that antigen presentation associated genes, such as CD80, CD86, SECTM1, CD74, CIITA, and MHCII, and regulatory genes, PDCD1LG2 and SOCS2, were significantly down-regulated in the NEC samples. Although inflammatory genes such as CCL3, CCL4, CCL20, CXCL2, CXCL8, IL1A, IL1B, IL6, CSF2, CSF3, S100A6, S100A10, GNG11, and TNFAIP6¹²¹ were significantly enriched in the NEC samples (Fig. 5D and Supplementary Table 10). For the shifts in the monocyte/Mφ populations in NEC, we performed RNAScope with IF to visualize the presence of NC slan⁺ monocytes (S100A4⁺IL1B⁺IL3RA^low) (Fig. 5E and Supplementary Fig. 6H) and MMP9⁺ macrophages (MMP9⁺IL3RA⁺S100A4⁻) in fetal and NEC SI (Supplementary Figs. 6H and 6I). Validating the scRNA-seq data, we observed that there was a slight increase in NC slan⁺ monocytes abundance and a significant increase in NC slan⁺ monocytes as the percentage of all S100A4+ cells in NEC compared with fetal samples (Supplementary Figs. 6J and 6K). On the other hand, MMP9⁺ macrophages had a strong trend toward a decrease in their overall cell numbers in the NEC samples (Supplementary Fig. 6L and 6M). In addition, the ratio of NC slan⁺ monocytes to MMP9⁺ macrophages was shifted in the NEC samples, where fetal samples were dominated by MMP9⁺ Mφ and NEC samples by NC slan⁺ monocytes (Supplementary Fig. 6N). However, we cannot exclude the possibility that these differences in monocyte/Mφ abundance and proportions were secondary to the bias of sectioned tissue regions or limited number of sample cases.

Given the observation of altered antigen presentation capacity of the myeloid cells in NEC samples, we next turned our attention to T cells. There was an increase in naive CD4⁺ and CD8⁺ T cells and decrease in CD4⁺ and CD8⁺ memory T cells in the NEC samples, consistent with our previous results²⁹ (Fig. 5C and Supplementary Figs. 6D and 6E). Because myeloid cells are required for memory T-cell generation, the significant down-regulation in MHCII genes on NEC-associated myeloid cells is consistent with decreased proportion of memory T cells in NEC samples and suggests a possible defect in memory generation. To further determine if there were unique features of NEC-associated naive T cells, we compared their transcriptional signature with naive T cells from other developmental groups. NEC-associated naive T cells exhibited a high similarity to naive T cells from fetal samples (Supplementary Fig. 7A). To further define the unique features of naive T cells associated with various developmental stages, we identified six different gene modules in naive T cells using Hotspot (v0.9.0) (Supplementary Fig. 7B and Supplementary Table 11), whose expression varied in similar ways among cells of similar developmental stages.¹²² Particularly, fetal and NEC naive T cells were found to be enriched in module 3 defined by IFN signaling genes (IFI44L, GBP4, STAT1, GBP2, GBP1, and MX1) (Fig. 5F) suggesting that fetal and NEC-associated naive T cells are transcriptionally similar and develop in a similar environment.

However, when similar analysis was performed for the CD4 memory T cells, we found that NEC-associated memory CD4 T cells were transcriptional more similar to those found at later developmental time points (pediatric, adult) rather than early developmental time points (fetal, neonatal) (Supplementary Fig. 7C), potentially suggesting that NEC-associated myeloid cells were priming T cells differently than what is observed during normal mucosal development. We additionally identified eight gene modules differentiating between CD4 memory T cells associated with different developmental time points (Supplementary Fig. 7D and Supplementary Table 12). We found that module 2 was enriched in fetal and neonatal T cells and consisted of a number of genes we had previously shown to be associated with fetal memory T cells (Fig. 5G). Interestingly, module 2’s overall signature and a number of genes, including ID3, PIK3RA1, and SOS1, were decreased in NEC-associated memory CD4 T cells and almost absent in pediatric and adult cells (Fig. 5G). Conversely, two modules—a module enriched for in activation-associated genes including JUNB, FOS, FOSB, and CD69 module 1 and module 4 enriched for effector genes PRDM1, IL22, cytokine receptor IL12RB1, and a resident memory gene CXCR6—were enriched in later samples (pediatrics and adult). Both modules were also increased in NEC samples (Fig. 5G). Moreover, the differential analysis revealed that NF-κB signaling genes, RELA, RELB, NFKB1, and NFKB2, were uniquely upregulated in NEC memory CD4 T cells compared with all other samples (Fig. 5G).

To determine whether receptor-ligand interactions were altered in NEC, we performed receptor-ligand pairing analysis that demonstrated NEC-associated specific interactions between myeloid and T cells (Supplementary Fig. 7E). Consistent with what was previously discussed, MHC-II-CD4 T cell and MHC-I-CD8 interactions were significantly down-regulated in NEC samples (Supplementary Fig. 7E). We also performed RNAscope to examine the expression of HLADR expression in fetal and NEC small intestinal samples and observed that the expression of HLADR was significantly decreased in NEC samples (Supplementary Figs. 7F and 7G), suggesting the reduction of antigen presenting capacity in NEC samples. This is consistent with our previous study where we observed a reduction in HLADR+ monocytes in NEC compared with control samples using IMC.³⁰ Conversely, interactions between CD55/ICAM1 on NC slan⁺ monocytes with ADGRE5 on Treg cells were upregulated in NEC samples (Supplementary Fig. 7E), consistent with our previous IMC results that the interaction between monocytes and Treg cells was especially enriched in NEC samples.³⁰ It has been reported that this interaction was involved in T-cell regulation, blocking the interaction resulted in the inhibition of proliferation and IFN-γ secretion in T cells.¹²³

Collectively, we found that NEC tissue had an expansion of myeloid cells that were present in fetal and neonatal tissue under normal developmental conditions but had reduced antigen presenting and regulatory capacity coupled with an increased inflammatory capacity that likely resulted in decreased proportions of memory T cells and their abnormal activated profiles.

DISCUSSION

The gastrointestinal tract is the largest immune organ in the human body that continuously balances tolerance to commensal organisms and protection from pathogens. Moreover, numerous diseases, such as NEC and inflammatory bowel disease, arise from dysregulation of intestinal immunity. As such, mucosal immune homeostasis is critical to human health. A better understanding of how mucosal immunity is developed and maintained would allow an improved understanding of mechanisms that could lead to pathology and thereby improved therapeutics. Several single-cell datasets have investigated the mucosal immune system at an unprecedented granularity^20–23,33. However, numerous gaps remain. None has profiled the human intestinal immune system across the entirety of the human life spectrum with majority lacking neonatal samples. Moreover, most of the data exploration were focused on the non-immune system.^21–23,33 With the goal to fill these gaps in mind, the current study provides a comprehensive single-cell atlas of the developing human small intestinal immune system from 8 weeks gestation to 70 years of age, with an emphasis on the fetal immune development. By comparing prenatal with postnatal cells, we identified that fetal SI cells display unique properties compared with postnatal samples, especially pediatric and adult samples, to meet the developmental needs. However, we cannot exclude that some of the age-associated differences we observed in cellular composition were due to either biological or technical effects, such as differences in tissue region collections, tissue processing methods, and sequencing saturation.

Mesenchymal and neuronal cells were highly enriched in fetal intestine compared with postnatal samples. In addition, we found an upregulation of collagen genes in fetal mesenchymal cells, suggesting an active epithelial-mesenchymal transition status in fetal SI.¹²⁴ Moreover, we found that fetal mesenchymal cells displayed much stronger ligand-receptor interactions with almost all other cell types other than postnatal samples. They were predicted to interact with all other intestinal cell subtypes through the MDK/PTN-NCL signaling among other ligand-receptor interactions, a pathway that promotes cell migration and survival.¹²⁵ This signaling pattern was also detected in fetal neuronal cells and highlights the importance of non-immune to immune interactions to the establishment of the intestinal immune system; in contrast, proinflammatory chemokines, such as CCL2 and CCL11, were found to be upregulated on postnatal mesenchymal cells. Interestingly, these chemokines interacted with endothelial cells by the expression of ACKR1 (also known as DARC), instead of their conventional chemokine receptors, a pattern that has previously been suggested to promote leukocyte migration to sites of inflammation.¹²⁶

Focusing on immune cells, myeloid cells were the first immune cells to ingress into the intestine and were detected at the earliest GAs. Of these, LYVE1⁺, yolk sac–derived macrophages, were the earliest to appear and were almost exclusively found in fetal samples. This is consistent with observations that in mice, yolk sac–derived macrophages that are abundant in fetal intestinal tissue, after birth, are rapidly replaced by the bone marrow–derived monocytes that are recruited to and differentiate into macrophages in the gut.^127,128 Interestingly, not all fetal macrophages expressed LYVE1 and several different populations of monocytes were observed to ingress into the fetal intestine at various GAs, suggesting a more complex origin for fetal SI macrophages. For example, MMP9⁺ macrophages were completely LYVE1-negative, seemed to differentiate from NC slan⁺ monocytes by PHATE analysis, entered the SI subsequent to the yolk sac–derived macrophages, and were enriched in the late second trimester and neonatal sample, all suggesting that these macrophages differentiate from a different hematopoietic source than the yolk sac and are potentially from fetal liver-derived monocytes.¹²⁷ In particular, IL7R (encoding CD127), a marker exclusively associated with the lymphoid lineage in hematopoiesis,¹²⁹ was uniquely expressed in these cells. CD127 expression has been reported in mouse fetal monocytes that is drastically down-regulated in the postnatal stage,^130–132 indicating a unique function of IL7Rα in fetal myelopoiesis. Interestingly, we also found the IL7R-expressing myeloid cells in NEC samples, along with the upregulation of inflammatory cytokine, such as IL1B, suggesting that these cells retained inflammatory phenotypes within the highly inflammatory tissue environments, as reported in patients with COVID-19.¹³³ Interestingly, another monocyte population we termed mono-neutrophil was present in the fetal SI, subsequent to the yolk sac–derived macrophages and expressed genes normally associated with bone marrow hematopoiesis, suggesting that even monocyte development is occurring in the fetal intestine.

As previously reported by several groups,^{6,20,21,26,134} there was a large variety of DCs, including cDC1, cDC2, and migratory DCs with transcriptional signatures, suggestive of maturity present in fetal SI tissues, especially during the 2^nd trimester. This was supported by the presence of phenotypic memory T cells also enriched in the 2^nd trimester samples. Consistent with the high abundance of CD4 over CD8 memory T cells, cDC2s were more prevalent than cDC1s during the fetal period. This is in direct contrast to a recent paper by Torow et al.⁹⁰ that demonstrated that cDC2s are not expanded until after weaning in neonatal pups, which leads to the low abundance of memory CD4 T cells in murine intestine until after weaning. Migratory DCs were present in late 2^nd trimester and neonatal samples and rapidly decreased in later-life samples. These cells were clearly separated from other DCs in the PHATE trajectory, suggesting their unique transcriptional profile and developmental origin. We found that these cells were especially enriched in the immunoregulatory signatures. It has been reported that these DCs promoted Treg induction in response to maternal antigens in utero,²⁶ indicating their importance in mediating immune homeostasis during gestation. Interestingly, these cells were also found largely expanded in tumor environment in adult mice and humans, where they play an important role in the anti-tumor responses.⁹⁵ They were significantly decreased in NEC samples compared with non-NEC neonatal samples, suggestive of loss of tolerance.

On the adaptive immune front, we found an upregulation of IFN signaling in fetal T cells. It has been reported that ISGs enriched T cells appeared in the late stages of human,¹³⁵ mouse,¹³⁶ and pig¹³⁷ thymocyte development in the absence of infection, suggesting that IFN signaling is involved in the maturation of T cells. However, the source and type of IFN production has not been defined yet. The decreased expression of IFNG in fetal T cells compared with postnatal T cells suggests that other IFN subtypes might be involved. Type III IFN (IFN-λs) are potential candidates because we observed an endogenous expression of IFNL1 in fetal cDC1 cells. Type I IFNs were barely detected in the current SI atlas. However, expression of type I IFNs is notoriously challenging to detect in single-cell datasets; therefore, the lack of expression does not completely rule out their contributions. However, it has been previously shown that in vivo treatment of human T cells with IFNL induced minimal ISG expression,¹³⁸ suggesting that it was unlikely the driver of the ISG in utero signatures. The lack of IFNL response is potentially due to the expression of a short IFN-λ receptor splice variant on various immune cells, which inhibits the IFN-λ1 effects at a high concentration.¹³⁹ Nevertheless, it has been reported that IFN-λ modulates DCs to facilitate downstream T-cell polarization and effector function during infection.¹⁴⁰ As such, the unique roles of IFN-λ and other IFNs on ISG-enriched T cells in fetal intestinal homeostasis needs further investigation.

Numerous previous studies have demonstrated that phenotypical memory T cells are present in the SI in utero,^{6,8,10,24,109} although the potential antigens priming these T cells have remained elusive. Several studies have suggested that fetal intestine contains a microbiome,^10,141 although this remains highly controversial. However, microbial-associated molecules have been reported in fetal intestine in mice and humans. Li et al.¹⁰ found a diverse set of microbially derived metabolites in fetal intestine,¹⁴² Kaisanlahti et al.¹⁴³ detected the bacterial extracellular vesicles in the amniotic fluid of healthy pregnant women, which exhibited similarities to extracellular vesicles found in the maternal gut microbiota, and Agüero et al.¹⁴⁴ found that maternal-microbiota-derived molecules partly bound to immunoglobulin G stimulated the innate immune maturation of the fetal gut in mice. Although all postnatal SI samples contain a microbiome, it varies drastically between full-term and preterm infants, including those who develop NEC.¹⁴⁵ How the microbiome alters intestinal immunity is an active area of investigation and should be further evaluated in the setting of NEC.

However, how these cells differ from postnatal memory T cells has not been well understood. The SI atlas highlighted several features unique to early-life T cells. We identified that the ratio of CD4⁺ to CD8⁺ memory T cells in early and later life is flipped with the maturity of CD8 T cells lagging behind that of CD4 T cells and leading to CD4 predominance in the second trimester samples. Interestingly, NK cells from the same developmental windows had increased effector function compared with later NK cells, and one subset of ILCs that appeared in early gestation also displayed a cytotoxic phenotype, suggesting that these cells at least partially compensate for low CD8 T-cell function early in life.

Consistent with recent studies, we found that fetal memory CD4 T cells were dominated by PLZF⁺ T cells, which have the capacity to rapidly produce Th1 cytokines, such as IFN-γ and TNFα.⁸ Conversely, CD8 memory T cells made up most memory T cells found in the postnatal intestine, especially pediatric and adult samples, with increased cytotoxic capacity. We also found a unique expression of HPGDS (encoding hematopoietic prostaglandin D synthase) in fetal memory T cells. It has been reported that hPGDS-expressing CD4⁺ lymphocytes demonstrate a typical Th2-like cytokine pattern with the production of PGD2, which may be involved in various aspects of Th2-related immune responses similar to mast cells.¹⁴⁶ However, up to now, they have only been associated with inflammatory disease states in tissue.^146–148 What role they play during development is yet to be identified.

Building on recent work from Donna Farber’s group,¹⁰⁹ we found that memory T cells marked by a stem-like gene signature were enriched in neonatal samples and displayed reduced activation and effector phenotypes. The stem-like gene signature was even higher in fetal memory T cells than neonatal T cells. Although the abundance of memory T cells expressing the stem-like signature decreased rapidly with advancing age, we were able to detect “fetal-like” memory T cells even in adult samples in support of layered development of the immune system^109,119 and the long-lived potential of fetal T cells. Even more striking, we were able to detect adult-like memory T cells in fetal samples that were reduced in neonatal samples, potentially indicating that these cells are short-lived and highlighting the vulnerability of the neonatal developmental state that had a lower abundance of fetal-like and adult-like memory T cells. In addition, we identified reciprocal expression patterns of some TFs in memory T cells between early versus late developmental time points. For example, ID3, IKZF2 (encoding Helios), and ZNF683 (encoding Hobit) were almost exclusively expressed by fetal and neonatal memory T cells and absent in pediatric/adult samples, whereas ID2 and PRDM1 (encoding BLMP1) were enriched at later time points. The distinct transcriptional profile between early-life and late-life memory T cells supports their unique functional roles at the various time points.

As a proof of concept that this developmental atlas can be used to identify potential developmental abnormalities underlying mucosal diseases, we integrated it with samples from patients with NEC, a severe gastrointestinal complication of prematurity. Interestingly, we observed a large expansion of NC slan⁺ monocytes, a monocyte subtype found in healthy late second trimester fetal samples and in neonatal SI tissue, indicating that the ingress of these monocytes into the SI occurs during normal development. However, in premature infants with NEC, the influx of these cells into the SI was largely exaggerated potentially due to the abnormal ex utero environment. Moreover, these cells, along with other antigen presenting cells (APCs) in the SI, drastically down-regulated their MHCII expression, explaining the increase in naive T cells in the NEC-associated samples. Moreover, the NEC environment also resulted in the down-regulation of regulatory factors and increase in inflammatory factors in these cells, perhaps resulting in memory T cells with decreased tolerance and increased inflammatory capacity, as observed in the NEC samples.

In summary, our comprehensive atlas of the developing human SI immune system provides biological insights into our understanding of the complex cellular landscape of the intestinal immune system in utero and postnatally and offers a valuable resource for the community. Moreover, we provide a framework of how the atlas can be used to identify deviations from normal developmental trajectories associated with disease states. This has potential implications for disease and for the engineering of in vitro systems.

METHODS

Human samples

All the single-cell resource data integrated in this manuscript have been published, and the detailed sample information can be found in Supplementary Table 1. Briefly, fetal tissue was collected from patients undergoing elective termination of pregnancy with institutional review board (IRB) approval and informed consent. The neonatal and NEC samples were obtained from surgical resections in infants under IRB exemption. Pediatric samples were obtained from clinically indicated colonoscopies, and adult samples were either obtained from fresh excised mucosal tissue from deceased transplant organ donors after ethical approval and informed consent from the donor families or obtained from intestinal biopsies from patients undergoing clinically indicated colonoscopies, with informed consent and IRB approval. The human small intestine samples used for validation staining were obtained with IRB approval.

Single-cell dissociation and sequencing

The single cells were isolated either from fresh tissue or biopsies digested by Liberase^21,23 or Neural Tissue Dissociation Kit,^22,32 or dispase,³³ or cryopreserved samples digested by collagenase.^29–31 Some dissociated cells were further enriched by magnetic cell separation (MACS) with EPCAM positive selection²³ or CD45 positive and negative selection.^21,29,31 All sample libraries were made using the Chromium 10× Genomics single-cell library kit with 3′ version^{22,23,29–33} or with 5′ version²¹ and sequenced on an Illumina Hi-seq 4000^{21,23,29–31,33} or Illumina Hi-Seq 2500.^22,32

Processing, clustering, and gene differential analysis of scRNA-seq data

Single-cell transcript counts output from Cell Ranger (10× Genomics) for fetal and postnatal samples were obtained from the published available resources (https://www.gutcellatlas.org/, https://doi.org/10.17632/x53tts3zfr.1., https://doi.org/10.17632/x53tts3zfr.1., https://doi.org/10.5281/zenodo.5813397) (Supplementary Table 1). Scanpy (v1.9.2) package³⁵ was applied to the scRNA-seq data for preprocessing and clustering. The cells for each dataset were filtered for >200 genes and <50% mitochondrial reads, and genes were filtered for expression in more than three cells. Scrublet (v.0.2.3)¹⁴⁹ was used to detect doublets, with doublet score produced. Gene expression for each cell was normalized and log-transformed. Afterwards, highly variable genes were identified using the scanpy.pp.highly_variable_genes function with default parameters. In addition, the effects of the percentage of mitochondrial genes, percentage of ribosomal protein genes, and unique molecular identifier (UMI) counts were regressed out using scanpy.pp.regress_out function before scaling the data.

Batch correction of samples was performed with bbknn (v1.5.1)³⁴ on 50 principal components. Dimensionality reduction and Leiden clustering was carried out on the remaining highly variable genes, and the cells were visualized using Uniform Manifold Approximation and Projection (UMAP) plots. The clusters with highest predicted doublet scores were excluded from the analysis. Cell lineages were manually annotated based on known markers genes found in the literature, in combination with algorithmically defined DEGs in each cluster, using scanpy.tl.rank_genes_groups function with the Wilcoxon test. This function was also used to identify DEGs across different age groups in selected cell types (threshold of pvals_adj < 0.05, log FCs > 1 or < −1, scores > 1 or < −1). The R package EnhancedVolcano (v1.14.0) (https://github.com/kevinblighe/EnhancedVolcano) was used to visualize the DEGs in the selected conditions. Hierarchical clustering analysis was performed with function of scanpy.tl.dendrogram. A comparison of cell type proportion between two selected developmental stages was performed using the scProportionTest package (https://github.com/rpolicastro/scProportionTest).

Trajectory inference analysis

We applied PHATE (v1.0.10)⁷⁶ for aligning cells into the developmental trajectory. The normalized datasets were imported into PHATE to instantiate a PHATE estimator object with default parameters. Then, PHATE embedding was generated with in a low dimension, which recapitulated the expected lineage relationships between the clusters, suggesting the progressive differentiation. scVelo was also applied to the RNA velocity analysis using dynamical mode.⁷⁷ The calculated RNA velocity vectors were embedded to force-directed graph in Scanpy package with the scanpy.tl.draw_graph function. Latent time was assessed through the scvelo.tl.latent_time function.

Differential abundance analysis

Differences in cell abundances associated with GA was analyzed using Milo with python implementation milopy (v0.1.1)¹⁵⁰ using the standard workflow. Briefly, a k-nearest neighbor graph was constructed using the Scanpy principal component analysis (PCA) embedding, and the cells were assigned into neighborhoods with function milo.make_nhoods. Finally, the number of cells in each sample within each neighborhood was counted and the differential abundance was modeled using a negative binomial generalized linear model to model the effects, such as GA or tissue regions using functions milo.count_nhoods and milo.DA_nhoods, respectively. The differential abundance estimates for each neighborhood were visualized using violin plot, with each node representing a given neighborhood in each cell type, and the negative log FC values representing the cells were enriched in earlier stages and vice versa.

Cell-cell interaction analysis

Cell-cell interaction from scRNA-seq data was predicted using R package CellChat (v1.6.1) based on the expression of known ligand-receptor pairs in humans.¹⁵¹ A comparison analysis was performed between the fetal and postnatal samples. First, we constructed the individual CellChat object. The normalized gene expression data of cells and assigned cell type from fetal and postnatal samples were used as input to construct the CellChat object. Subsequently, the fetal and postnatal objects were merged by the “mergeCellChat” function for comparison. Finally, the compareInteractions function was used to compare interaction strength of the inferred cell-cell communication networks, “netVisual_heatmap” was used to generate comparison heatmap, and upregulated/down-regulated signaling ligand-receptor pairs in fetal and postnatal samples were identified based on the differential gene expression analysis. Specifically, a differential expression analysis between fetal and postnatal conditions for each cell type was performed; then, the FC of ligands in the sender cells and receptors in the receiver cells was computed and visualized using the chord diagrams at the threshold of log FC > 0.1.

Cell signaling genes scoring

The gene lists of proliferation, glycolysis, Myc, E2F, mTOR, inflammatory, p53, and TCR signaling were downloaded from the Gene Set Enrichment Analysis website (https://www.gsea-msigdb.org/gsea/msigdb/human/search.jsp) and, together with MHCII gene list, the cytotoxic gene list that can be found in Supplementary Table 4. The scanpy.tl.score_genes function was used to quantify the signature expression in each cell. The signature score for each cell was defined as the average expression of the selected genes subtracted with the average expression of reference genes. The reference gene set was randomly sampled from the gene pool for each binned expression value. Two-way analysis of variance (ANOVA) with multiple comparisons was performed for statistical analysis of cell signaling scores between the fetal and postnatal samples using GraphPad Prism 9.

Gene regulatory network analysis

The gene regulatory network was inferred using pySCENIC (v0.12.1, a lightning-fast python implementation of the SCENIC pipeline).^49,152 Firstly, gene-gene co-expression relationships between TFs and their potential targets were inferred using the grn command with the GRNBoost2 algorithm. Next, command ctx was performed to identify the regulons (TF + target genes) using the ranking databases (hg38_500bp_up_100bp_down_full_tx_v10_clust.genes_vs_motifs.rankings.feather, hg38_10kbp_up_10kbp_down_full_tx_v10_clust.genes_vs_motifs.rankings.feather) and motif annotation database (motifs-v10nr_clust-nr.hgnc-m0.001-o0.0.tbl). After that, the command aucell was used to calculate the activities of the regulons. Cell type or age specific regulons were identified based on the Z score of the area under the recovery curve values for a given cell types.

Highly correlated gene module identification

We used Hotspot package (v1.1.) to identify highly correlated genes into modules, which computes the gene modules by finding informative genes with high local autocorrelation, evaluating the pairwise correlation between genes, and clustering the results in a gene-gene affinity matrix.¹²² Briefly, we used the “normal” model on the log-normalized counts to create the Hotspot object and construct the k-nearest neighbors graph with 30 neighbors and selected the top 500 genes with the highest autocorrelation Z scores under the threshold of false discovery rate < 0.05 for module identification. We then computed the pairwise local autocorrelations between these genes, and clustered genes into modules using the create_modules function (minimum gene threshold of 20, false discovery rate threshold of 0.05, core_only=True). Finally, the aggregated gene module scores were calculated using function calculate_module_scores.

IF staining

Formalin-fixed, paraffin-embedded small intestinal tissue was sectioned into 4–5-μm thick sections for staining, as described.²⁹ Slides were deparaffinized using xylene and alcohol and placed in 1 × antigen retrieval buffer (R&D Systems, #CTS013) at 95°C for 1 hour. Next, slides were washed in distilled H20 (ddH20) and Dulbecco’s Phosphate Buffered Solution (Gibco). Slides were blocked with 10% horse serum and incubated with primary antibodies (HPGDS, Invitrogen, MA5–24347; CD3, Abcam, ab135372). Slides were then incubated with secondary antibody (anti-rabbit and anti-mouse, Invitrogen). Images were taken at 20 × using Echo Revolve microscope. CD3⁺HPGDS⁺ and CD3⁺ cells were quantified using Cell Counter tool on FIJI ImageJ.

In situ hybridization (RNAscope) combined with IF

The RNAscope Multiplex Fluorescent Reagent Kit v2 with TSA Vivid Dyes assays (Advanced Cell Diagnostics) were conducted using target RNA probes, MMP9 (#311331), IL1B (#310361), IL3RA (#421241-C2), and CCR9 (#515171-C3) based on the manufacturer’s instructions. In brief, paraffin sections were deparaffinized, treated with a citrate buffer, and hybridized sequentially with target probes. After RNA signal development, a blocking step was performed with 10% horse serum, followed by overnight incubation at 4°C with desired primary antibody (S100A4, Invitrogen, MA5–32347 or CD45RA, Invitrogen, MA1–19113) dilutions, and incubated with secondary antibody as previously described. Images were taken at 20 × using Echo Revolve microscope. RNA signals appear as dots.

Data and code availability

The sequencing data used in this study are listed in Supplementary Table 1 where the data resources can be found. Major code used in this study is available on https://github.com/WG20231/SI-atlas.git. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.