Cell type- and stimulation-dependent transcriptional programs regulated by Atg16L1 and its Crohn’s disease risk-variant T300A

Abstract

Genome-wide association studies (GWAS) have identified common genetic variants impacting human diseases; however, there are indications that the functional consequences of genetic polymorphisms can be distinct depending on cell type-specific contexts, which produce divergent phenotypic outcomes. Thus, the functional impact of genetic variation and the underlying mechanisms of disease risk are modified by cell type-specific effects of genotype on pathological phenotypes. In this study, we extend these concepts to interrogate the interdependence of cell type- and stimulation-specific programs influenced by the core autophagy gene Atg16L1 and its T300A coding polymorphism identified by GWAS as associated with increased risk of Crohn’s disease. We applied a stimulation-based perturbational profiling approach to define Atg16L1 T300A phenotypes in dendritic cells (DCs) and T lymphocytes. Accordingly, we identified stimulus-specific transcriptional signatures revealing T300A-dependent functional phenotypes that mechanistically link inflammatory cytokines, interferon response genes, steroid biosynthesis, and lipid metabolism in DCs and iron homeostasis and lysosomal biogenesis in T lymphocytes. Collectively, these studies highlight the combined effects of Atg16L1 genetic variation and stimulatory context on immune function.

Introduction

Emerging insights into immune pathologies identify key cell types controlling cell stress pathways and inflammation (1–7). In particular, the genetic association of Atg16L1 with increased risk of inflammatory bowel diseases (IBD) implicated autophagy in the dysregulation of immune homeostasis (4, 8–11). Autophagy is a cellular disposal system that directs cytoplasmic cargo into lysosomes for proteolytic degradation. In addition to recycling biomass such as organelles, autophagy targets intracellular pathogens and regulates inflammatory cytokine production (10–14).

The Atg16L1 gene is broadly expressed across cell types, and the functional effects of the T300A allele conferring increased risk of Crohn’s disease are incompletely understood. At the molecular level, the T300A substitution introduces a caspase cleavage site on Atg16L1 (1, 3). Cell stress associated with caspase activation potentiates autophagy defects in the T300A genetic background (1, 3). In response to infection and starvation, Ulk1-mediated phosphorylation of wild type Atg16L1 enhances antibacterial autophagy and, importantly, promotes cleavage of Atg16L1 T300A (15). These findings are consistent with the observation that Atg16L1 T300A renders mice susceptible to infection with Yersinia enterocolitica, and caspase inhibition ameliorates intestinal pathology in this model (3). Thus, cell extrinsic stress pathways prime genetic susceptibility to intestinal pathology. Additionally, T300A disrupts protein-protein interactions mediated by the Atg16L1 WD40-repeat domain (16), which provides a docking site for interaction with TMEM59 and subsequent recruitment of the autophagy machinery to intracellular vesicles (17). Binding of the anti-inflammatory protein A20 to the WD40-repeat domain regulates intestinal epithelial cell death, but this physical interaction is not affected by the presence of T300A (18). As new innate defense mechanisms are described, the importance of autophagy becomes increasingly clear. Atg proteins including Atg16L1 were recently shown to mediate the release of exomes that neutralize pore-forming toxins produced by pathogens such as methicillin-resistant Staphylococcus aureus (MRSA) (19). It remains unclear how Atg16L1 T300A affects disparate cell types in differing environmental contexts and how this cumulatively impacts disease pathogenesis.

Generation of Atg16L1 T300A knock-in mice has facilitated discoveries of cell type-specific programs that control immune homeostasis. In epithelial cells, Atg16L1 T300A is associated with defective antibacterial autophagy, Paneth and goblet cell granule morphology, and secretory function (1, 3, 6, 20, 21). Additionally, Atg16L1 T300A knock-in mice exhibit a variety of pathological phenotypes throughout the immune system. Macrophages derived from T300A mice produce elevated levels of IL-1β in response to pathogen-associated molecular pattern (PAMP) stimulation (1, 22). These results are consistent with previous reports of elevated IL-1β in Atg16L1 knockout mice (23). Autophagy has been implicated in antigen presentation by dendritic cells (DCs) and priming CD4 T cells (24–26). Moreover, autophagy controls T cell homeostasis and attrition (27–29), and Atg16L1 is specifically required for Treg maintenance of peripheral tolerance (30, 31). Taken together, these data identified a range of host pathways that impinge on selective autophagy and implicated a number of functional connections by which Atg16L1 T300A may contribute to diverse pathological phenotypes. It is increasingly apparent that the extent to which Atg16L1 T300A might influence specific pathways triggered in response to different stimuli in one cell type cannot be directly inferred from another cell type, as this may not be representative or generalizable across different cell types but also stimulation or perturbation states. Here, we define T300A-specific transcriptional profiles in T cells, dendritic cells, and mast cells to demonstrate that the Atg16L1 T300A allele conspires with environmental triggers to induce several cell type- and stimulation-specific phenotypes associated with inflammatory pathology.

Methods

Cells and stimulation

Dendritic cells:

Splenic dendritic cells were isolated from C57BL/6J (WT), CD11c-Cre⁺ x Atg16L1^flox/flox (cKO), and Atg16L1^T300A/T300A mutant (T300A) (3) mice with mouse CD11c MicroBeads UltraPure (Miltenyi #130-108-338). For each genotype, the cells were prepared independently from 3 mice as biological replicates. The cells were then cultured overnight in RPMI1640 medium containing 10% FBS, MEM Non-Essential Amino Acids, GlutaMAX, 1 mM Sodium Pyruvate, 55 µM 2-Mercaptoethanol, Penicillin and Streptomycin (all from Thermo). On the next day, the cells were stimulated with 1 µg/mL of LPS for 0, 4 and 12 h. These time points were selected based on published cytokine responses (32).

T Cells:

CD4⁺ CD62L⁺ naïve T cells were isolated from C57BL/6J (WT), Lck Cre⁺ x Atg16L1^flox/flox (cKO), and Atg16L1^T300A/T300A mutant (T300A) mice using CD4 negative enrichment kits (Stemcell Technologies, Vancouver, Canada) and CD62L microbeads (Miltenyi Biotec, San Diego, CA). For each genotype, the cells were prepared independently from 3 mice as biological replicates. T cells isolated per genotype each received no stimulation, CD3/CD28 dynabeads (for 4h or 20h) or 1 μM Torin treatment (for 4h). Peak T cell activation, as defined by number of differentially expressed genes, after CD3/CD28 stimulation was observed at 20h; the 4h time point allowed us to perform a time-varying analysis of transcriptional programs in response to CD3/CD28 stimulation. A 4h Torin treatment activated autophagy in T cells.

Mast Cells:

Bone-marrow derived mast cells were isolated from WT, Mcpt5-Cre⁺ x Atg16L1^flox/flox (cKO) and Atg16L1^T300A/T300A mice. For each genotype, cells were isolated from 4 different mice in 2 different batches (3 mice and 1 mice). All cells were treated with 100ng/mL anti-DNP IgE sensitization for 2 hours, followed by either no further stimulation (no stim), 10ng/ml DNP-BSA antigen for 30 minutes or 10ng/ml DNP-BSA antigen for 1 hour. These time points were selected based on previously published studies on the kinetics of mast cell degranulation (33–35).

RNA Sequencing

T Cells:

Cell lysis and mRNA isolation was performed with the Dynabeads mRNA Direct Kit (Life Technologies). mRNA from approximately 150,000 cells was reverse transcribed in a polydT primed reaction with template switch oligo (TSO) and Maxima H Minus Reverse Transcriptase (Thermo Scientific). NEBNext Ultra II Non-Directional RNA Second Strand Synthesis Module (New England Biolabs) was used to generate double stranded cDNA. The cDNA was cleaned and purified with Agencourt AMPure XP beads (Beckman Coulter). Each sample was then tagmented using the Nextera XT DNA Library Prep Kit (Illumina) and the Nextera XT index kit (Illumina). Post reaction purification was performed with Agencourt AMPure XP beads. Samples were then pooled, prepared, and loaded onto a MiSeq (Illumina) and NextSeq (Illumina) per the manufacturer’s instructions.

Mast Cells:

Cell lysis was performed with TCL buffer (Qiagen) containing 1% 2-Mercaptoethanol. Full length cDNA libraries were prepared with lysate from approximately 2,000 cells per sample, using template switching and whole transcriptome amplification in a modified version of the SmartSeq2 protocol described (36). Post SmartSeq2, double strand cDNA was cleaned and purified with Agencourt AMPure XP beads (Beckman Coulter). Each sample was then tagmented using the Nextera XT DNA Library Prep Kit (Illumina) and the Nextera XT index kit (Illumina). Post reaction purification was performed with Agencourt AMPure XP beads. The samples were pooled and run on a 2% E-Gel EX Agarose Gel (Thermo Scientific) and gel extracted with zymoclean gel DNA recovery column (Zymo Research Corporation). Samples were prepared and loaded onto a MiSeq (Illumina) per the manufacturer’s instructions.

Dendritic cells:

Cell lysis was performed with Lysis/Binding Buffer for Dynabeads (Thermo # A33562), and mRNA isolation was performed with the Dynabeads mRNA Direct Kit (Life Technologies). Full length cDNA libraries were prepared with mRNA from approximately 26,250 cells, using template switching and whole transcriptome amplification in a modified version of the SmartSeq2 protocol (36). Post SmartSeq2, double strand cDNA was cleaned and purified with Agencourt AMPure XP beads (Beckman Coulter). Each sample was then tagmented using the Nextera XT DNA Library Prep Kit (Illumina) and the Nextera XT index kit (Illumina). Post reaction purification was performed with Agencourt AMPure XP beads. The samples were pooled and run on a 2% E-Gel EX Agarose Gel (Thermo Scientific) and gel extracted with zymoclean gel DNA recovery column (Zymo Research Corporation). Samples were prepared and loaded onto a MiSeq (Illumina) and NextSeq (Illumina) per the manufacturer’s instructions.

RNASeq analysis pipeline and data preprocessing

Reads were aligned to the mm10 transcriptome using Tophat2 + Bowtie (37). Transcripts were quantified using htseq-count (38), and QC was performed using RseqC (39). TMM normalization as implemented in the edgeR package (40) was used for between-sample normalization. TMM-normalized counts were used as input for further analysis. Genes with zero counts across the board were excluded from downstream analysis.

Differential expression analysis and pathway enrichment

The data were modeled with a generalized linear model of the form y ~ Genotype + Stimulation + Genotype:Stimulation for the dendritic cells and T cells individually. A likelihood ratio test was used to test for differential expression between conditions. An FDR-adjusted p value cutoff of 0.05 was used to select genes for pathway enrichment. Pathway enrichment was performed using the goana and kegga functions in edgeR (41) and the clusterProfiler package (42). KEGG (43), Gene Ontology (44) and Reactome (45) databases were queried to find enriched categories.

CAMERA gene set enrichment analysis of transcription factor targets

Gene sets for targets of known transcription factors were tested for differential expression under the interaction model y ~ Genotype x Stimulation. The CAMERA test (46) as implemented in edgeR was used to set differences in the expression of these gene sets. The CAMERA test accounts for inter-gene correlation, which makes it particularly suitable for testing differential expression of gene sets that have a common biological phenomenon (such as a shared transcription factor in this case) driving their gene expression. Figure 3 shows transcription factor enrichment based on target gene expression in T cells. Each point in that plot is a gene set, e.g. the point labeled Hsf1 is the set of all genes regulated by Hsf1. The plot thus represents the transcription factor target gene sets double-differentially expressed under the interaction model described above, i.e. genes that show a higher or lower change in expression upon stimulation between two different genotypes. Similarly, Supplementary Figure 3 shows transcription factors whose targets are enriched for in the double differential comparisons under LPS stimulation for splenic DCs.

Figure 3. Network-dysregulation and transcription factor-enrichment analyses identify hidden targets of Atg16L1 genetic perturbation in T cells.

(A) To uncover the activity of transcription factors that were not expressed at high levels themselves, the targets of transcription factors in the TRRUST database (95) were tested for enrichment by applying the competitive gene set test that accounts for inter-gene correlation (46). Transcription factors Hsf1, Notch3, Nr1h2, Tcf4 were identified as top hits with functionally relevant known annotations. Term size refers to the number of genes in each associated term in TRRUST. (B) To infer gene modules that are significantly dysregulated by Atg16L1^f/f x Lck Cre⁺ and Atg16L1^T300A/T300A, we applied the DeMAND algorithm (48) and compared all stimulation samples (non-baseline) from either group to WT. The submodule centered at Galectin3 (LGALS3) was found to be most dysregulated in the Atg16L1^f/f x Lck Cre⁺ versus WT comparison, with genes involved in lysosome and trafficking contributing to the phenotype.

Protein network perturbation inference

A change in topology due to genotypic perturbations was tested for in the BioPlex network (47) – an established, experimentally verified protein interaction network. The algorithm used was DeMAND algorithm (48). DeMAND, while originally developed to study chemical perturbations through small molecules, can be applied equally well to study genetic perturbations such as cKOs (Atg16L1) or SNPs (T300A). DeMAND uses a gaussian kernel to estimate the interaction probability density and compares it across two conditions to estimate network dysregulation. Since the algorithm works best with larger group sizes, samples from across all stimulations (i.e., non-baseline samples) were grouped and then compared across genotypes to find stimulation-independent genetic perturbations.

Variance partition analysis

The variancePartition (49) was used to quantify the variance attributable the different factors in our experiment. Briefly, a linear model was computed for the expression of each gene individually, and the fraction of variance attributable to the different factors (cell type, stimulation, and genotype) in our dataset after batch-correction was computed. Each factor was treated as a random effect by virtue of being a categorical variable. Since we were using this analysis to find the relative strength of the different factors, interaction terms were not included in the model at this stage.

Antibodies and compounds

Antibodies used for immunoblot are as follows: Egr1 (Cell Signaling 4154, 1:1000); cMyc (Cell Signaling 5605, 1:1000); Rb1cc1 (Cell Signaling 12436, 1:1000); Wnt10a (Santa Cruz sc-376028, 1:500); Ulk1 (Cell Signaling 8054T, 1:1000); Akt1 (Cell Signaling 2938, 1:1000); CD28 (Cell Signaling 38774S, 1:1000); Lamp2 (Sigma L0668, 1:1000); Cathepsin C (Santa Cruz 74590, 1:500); Clathrin (Cell Signaling 2410, 1:1000).

Immunoblotting

Cells were lysed in standard lysis buffer [50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 1% Nonidet P-40, Halt phosphatase inhibitor single use cocktail (Pierce), and protease inhibitor tablets (Roche)] in 4°C for 30 min. Lysates were centrifuged at 18,000 x g at 4°C for 15 min and the supernatant was collected for protein concentration estimation using a Bradford assay. Samples were prepared using 5X loading buffer (250 mM Tris-HCl pH 6.8, 10% SDS, 30% glycerol, 0.02% bromophenol blue, 5% β-mercaptoethanol) and boiled for 5 min. Samples were electrophoresed in 4–20% MP TGX polyacrylamide gels (Bio-Rad) and transferred onto PVDF using wet transfer at 80 V for 1 h. 5% BSA in TBST was used to block the membrane for 1 h. Blots were incubated overnight at 4°C with antibody prepared in 1% Bovine Serum Albumin. After three washes with TBST, the membrane was incubated with HRP-conjugated secondary antibody for 60 min at room temperature. Following secondary incubation, the blot was washed three times in TBST and incubated with chemiluminescent HRP substrate (Millipore). All Western blots were performed at least three independent times.

Results

Autophagy controls multiple immune functions, and genetic knockout of core autophagy proteins exerts pleiotropic effects on immune homeostasis. However, it remains unclear how genetic variants of Atg16L1 impact immune function in different cell types and how environmental context modifies these phenotypes. Here, we applied a stimulation-based perturbational profiling approach in dendritic cells, T lymphocytes, and mast cells to define Atg16L1 T300A phenotypes relative to Atg16L1 knockout. Accordingly, we sought to identify transcriptional signatures that reveal T300A-dependent functional phenotypes associated with immune cell type and activation status.

LPS stimulation reveals Atg16L1 genotype-dependent responses in dendritic cells

First, we employed an endotoxin-elicited immune response model in mouse Cd11c⁺ splenic DCs to assess alteration of the transcriptional landscape in the context of Atg16L1 genotype. In this model, Toll-like receptor 4 (TLR4) engagement by lipopolysaccharide (LPS) in splenic DCs induces an inflammatory response characterized by robust cytokine production (50). CD11c⁺ splenic DCs were isolated from wild type (WT), Atg16L1^f/f x CD11c Cre⁺ (cKO), and Atg16L1^T300A/T300A knock-in (T300A) mice. Stimulation time points (4h and 12h) were selected based on cytokine responses (32) and allowed us to perform a time-varying analysis of LPS-induced transcriptional programs that are dependent on Atg16L1 and the T300A variant.

Stimulation of WT splenic DCs with LPS elicited a strong transcriptional response, peaking at 4h, with many of the genes enriched in discrete biological processes. Enriched categories associated with immunity, including Toll-like receptor and pattern-recognition genes, cytokines and chemokines, type I and II interferons and interferon-regulated genes, signal transduction and transcription factors in immune regulation (Figure 1A), served as confirmation of the fidelity and integrity of LPS stimulation. More importantly, we found that the absence of Atg16L1 dramatically altered this LPS-induced transcriptional response in splenic DCs, affecting the expression of 1,039 genes at 4h. A vast majority (88%; 916/1,039) of the genes impacted were more substantially altered by Atg16L1 deficiency in the LPS-stimulated state than in the unstimulated baseline state. Thus, LPS stimulation accentuated genotype-associated differences and provided the context dependence in which the loss of Atg16L1 was able to manifest its effect.

Figure 1. LPS stimulation reveals Atg16L1 genotype-dependent responses in splenic dendritic cells.

(A) Heatmap shows unstimulated (No stim) and LPS-induced transcriptional profiles of Cd11c⁺ splenic DCs isolated from Atg16L1^f/f x Cd11c Cre⁺ (cKO), Atg16L1^T300A/T300A (T300A) and wild type (WT) mice. Expression values (Log₂CPM from biological triplicates) were Z-score-transformed and represented by color intensity transitions shown in the color bar. The expression of these genes was strongly influenced by LPS stimulation and genotype and was enriched in pathways and processes associated with immune responses, lipid and steroid metabolism, Wnt signaling, and mitochondrial and Golgi function. (B-F) Boxplots showing expression profiles of key genes (B, Ifna5; C, Idi1; D, Faah; E, Aoah; F, Ch25h) in splenic DCs for which LPS stimulation (for 4h and 12h) and genotype exert a strong effect.

During LPS stimulation, we found the expression of many immune response genes (enrichment p=10⁻⁸) substantially reduced in the absence of Atg16L1 compared to WT. Inhibited genes include cytokines (IL6, IL1b, IL15, IL18, IL27), chemokines (Ccl3,4,7,12, Cxcl1,2,9,11), Tnf, Tnfsf10, Toll-like receptors (Tlr3,8,9), and the inflammasome component Nlrp3. Strikingly, the strong LPS induction of interferon genes (Ifna2, Ifna5, Ifnb1, Ifna4) was inhibited almost completely in splenic DCs lacking Atg16L1 (Figure 1A and 1B). In contrast, splenic DCs expressing the Atg16L1 T300A variant exhibited diametrically opposing trends when compared to Atg16L1 cKO cells; after LPS stimulation of T300A DCs, many of the altered immune genes (i.e., Ifna5) showed expression enhanced beyond the induction level observed in WT splenic DCs (Figure 1A and B).

In striking contrast to the transcriptional profile of immune response genes, key steroid biosynthesis pathway genes (Fdft1, Cyp51, Hmgcs1, Pmvk, Idi1) were upregulated in LPS-stimulated cKO but not WT or T300A DCs (Figure 1A and 1C), suggesting that Atg16L1 plays important dual roles in promoting inflammation and dampening anti-inflammatory processes in the LPS response by inhibiting genes involved in steroid biosynthesis. Thus, LPS stimulation not only accentuated but also polarized genotype-associated differences, namely enhanced expression of inflammatory cytokines and interferon response genes in T300A DCs.

Compared to cytokine responses, LPS-activated metabolic programs that participate in inflammation, autophagy and immunity are incompletely understood, but new insights into these processes are beginning to emerge (51–53). Cellular metabolites can serve as danger signals that trigger proinflammatory responses following LPS stimulation (54). LPS can shift core metabolism in DCs from oxidative phosphorylation to glycolysis, which enhances fatty acid metabolism needed for endoplasmic reticulum and Golgi expansion (55), demonstrating the importance of lipid metabolism to innate immunity and potentially revealing a role in regulating autophagy. Here, we highlight three lipid metabolism genes (Faah, Aoah and Ch25h) that we identified as transcriptionally induced by LPS and dependent on Atg16L1 or its T300A variant (Figure 1D–F). Both Faah and Aoah encode hydrolases, while Ch25h is a hydroxylase. Faah (fatty acid amide hydrolase) participates in the degradation of endocannabinoids anandamide and palmitoylethanolamine, which are involved in intestinal inflammation (56). Induction of the Faah gene by LPS was diminished in the absence of Atg16L1 and in the presence of the T300A variant (Figure 1D). Aoah (acyloxyacyl hydrolase) selectively inactivates bacterial LPS by specifically hydrolyzing the acyloxyacyl-linked fatty acyl chains in the lipid A moiety (57). Our transcriptomic analysis revealed a feedback loop involving Aoah. At baseline, Aoah was expressed at low levels (Figure 1E), and no significant difference across WT, cKO and T300A was observed. Upon LPS stimulation, expression levels of Aoah were markedly upregulated at 4h and further increased at 12h. This is consistent with the role of Aoah in LPS inactivation, establishing a functional feedback loop that senses LPS levels. The absence of Atg16L1 reduced Aoah expression compared to WT when stimulated with LPS at 4h and 12h. Conversely, the T300A variant exhibited even higher Aoah induction levels compared to WT at both LPS-stimulated time points, reminiscent of the expression profile observed for immune response genes in which genotypic differences were polarized by LPS stimulation.

From our analysis, we found a striking concordance between the expression profiles of Ch25h and immune response genes (Figure 1A and 1F); the dependence of Ch25h expression on Atg16L1 became evident only under LPS stimulation. Similar to immune response genes, the induction of Ch25h expression by LPS was significantly diminished in the absence of Atg16L1 compared to WT. At the later time point of LPS stimulation (12h), Ch25h induction was higher in the T300A variant than in WT. As a cholesterol hydroxylase, Ch25h catalyzes the formation of 25-hydroxycholesterol (25-HC), an intermediate in the biosynthesis of oxysterol, which is a ligand for the lymphoid cell chemotactic receptor Gpr183/EBI2. Intriguingly, 25-HC promotes a robust NLRP3 inflammasome assembly, production of IL-1β (58), and mitochondrial ROS (reactive oxygen species)-mediated pathways, which is consistent with TLR4 engagement (59). Transcriptionally, we observed that these components are coordinately upregulated by LPS and Atg16L1-dependent.

We detected an enrichment of genes associated with the Wnt pathway that shared similar expression profiles with the immune response genes. The expression of these genes was induced by LPS in WT and T300A but inhibited in the absence of Atg16L1. Protein levels of Wnt10a, Egr1, and c-myc confirmed that LPS increased expression of Wnt pathway genes in WT and T300A but not cKO DCs (Supplementary Figure 1). The Wnt pathway is not only essential for embryonic development and homeostasis but has more recently been recognized to be important in exerting immunomodulatory influence in inflammation, infection, and autophagy (60–62). Recently, the intracellular bacterium Ehrlichia was found to induce and exploit the Wnt pathway to evade destruction in the autophagolysosome. The Wnt pathway inhibited lysosomal fusion and the autolysosomal destruction of Ehrlichia (62). There is additional evidence that the Wnt pathway can inhibit autophagy by regulating the activation of the mammalian target of rapamycin (mTOR) pathway (63–65) through Akt-mediated GSK3 phosphorylation (66). These findings, together with our identification of key Wnt pathway genes induced by LPS in splenic DCs in an Atg16L1-dependent manner, highlights the importance of Wnt-mediated regulation in antibacterial autophagy and innate immune defense response.

Taken together, LPS stimulation in splenic DCs strongly alters, accentuates, and polarizes the genotypic impact of Atg16L1 or its T300A variant on the transcriptional landscape. In the absence of LPS stimulation, the transcriptional differences between genotypes in splenic DCs were less marked.

Distinct Atg16L1- and T300A-dependent transcriptional signatures in T cells

By profiling naïve CD4⁺ CD62L⁺ T cells from wild type (WT), Atg16L1^f/f x Lck Cre⁺ (cKO), and Atg16L1^T300A/T300A (T300A) mice, we identified transcriptional signatures that are both genotype- and T cell type-dependent. At baseline, we found genes associated with iron-binding, transport, and metabolism (67) to be upregulated in Atg16L1 knockout cells. The ferritin gene Ftl1, genes involved in iron homeostasis (Slc46a1, Slc4a1, Slc40a1, Slc25a37, Trf, Hmox1) and the entire heme complex (Hbb-b1, Hbb-b2, Hba-a1, Hba-2) were upregulated when Atg16L1 was disrupted (Figure 2A). In fact, we found the heme complex genes to be among the most highly differentially upregulated genes between cKO and WT T cells (Figure 2B). The upregulation of Ftl1 when Atg16L1 was deleted suggests a response to elevated ferritin levels, which is indicative of a disruption in the iron transport and metabolism network. Importantly, this elevated transcriptional response to Atg16L1 genotype differences is cell type-dependent. The expression increase observed for these iron-associated genes in cKO T cells (Atg16L1^f/f x Lck Cre⁺) was not seen in cKO splenic DCs (Atg16L1^f/f x CD11c Cre⁺) relative to Atg16L1 T300A and WT cells (Figure 2B). Altered expression of genes in the iron transport pathway is consistent with impaired selective ferritinophagy.

Figure 2. Atg16L1 genotype shapes transcriptional programs and activation responses in T cells.

(A) Heatmap showing expression profiles of naïve CD4⁺ CD62L⁺ T cells from Atg16L1^f/f x Lck Cre⁺ (cKO) and wild type (WT) mice. Atg16L1 genotype- and stimulation-independent genes were associated with iron binding, transport, and metabolism. Expression values (Log₂CPM) were Z-score-transformed and represented by color intensity transitions shown in the color bar. (B) As a representative gene from the heme/iron cluster shown in (A), Hbb-b1 expression profiles of T cells (top) and splenic DCs (bottom) are contrasted in boxplots. T cells were stimulated for 4h or 20h with CD3/CD28 and for 4h with Torin. DCs were stimulated for 4h or 12h with LPS. The expression of Hbb-b1 in T300A T cells was intermediate between WT and cKO. Hbb-b1 expression was upregulated in T300A splenic DCs compared to both WT and cKO. Expression values are presented as Log₂CPM of the gene transcripts for each cell type. (C) Volcano plot shows that autophagy genes were significantly induced by Torin stimulation in WT T cells, whereas Mtor was downregulated. Red and blue dots denote significantly upregulated and downregulated genes respectively (FDR<0.05 and fold change >1.5). Grey dots indicate insignificant change in gene expression. (D) Volcano plot of T cell activation genes significantly induced by stimulation with CD3/CD28 for 20h in WT T cells. (E and F) Barplots showing pathway enrichment analysis of stimulation- and genotype-dependent transcriptional shifts in responses to (E) Torin and (F) 20h CD3/CD28 stimulation. Response to stimulus was captured with a factorial design model that evaluates stimulation-genotype factor interactions (double differentials). Dumbbell plots show these double differential shifts of genes (Log₂(Fold change between stimulated and unstimulated)) driving the enrichment of the respective key pathways and also exhibiting an Atg16L1-dependent response to Torin or CD3/CD28 stimulation. Each node on the dumbbell represents a response. Each pair of nodes connected in a dumbbell indicates the two genotypes being compared (Atg16L1^f/f x Lck Cre⁺ cKO and WT). The distance between each pair of nodes show the shift in response between WT and cKO.

To define altered stimulation-dependent transcriptional programs in T cells, we profiled expression changes in Torin-induced T cells and CD3/CD28-stimulated T cells across genotypes (Atg16L1^f/f x Lck Cre⁺ cKO, T300A, and WT). Torin, a small molecule mTOR inhibitor, mimics cellular starvation and induced autophagy after 4h in T cells. Given that autophagy is essential for T cell activation, we first evaluated the efficacy of these stimulations by examining differential expression of autophagy-related genes and T cell activation-related genes for Torin and CD3/CD28 stimulation, respectively. Consistent with Torin stimulation in WT T cells, we observed down-regulation of Mtor expression and up-regulation of core autophagy complex components (68) including Ulk1 (Atg1), Atg13, Rb1cc1 (Fip200), Atg13, Atg14, Wipi1/2 (Atg18), Atg2b, Map1lc3a (member of the LC3/Atg8 complex), and genes encoding autophagy adaptors Sqstm1 (p62), Ncoa4, Optn and Tax1bp1 (Figure 2C). After CD3/CD28 stimulation, we observed peak T cell activation as defined by number of genes being differentially expressed at 20h and used this time point for subsequent analyses. Including an intermediate (4h) time point again allowed us to evaluate time-varying transcriptional responses to CD3/CD28 stimulation. As expected, genes associated with T cell activation were induced in CD3/CD28-stimulated WT T cells (Figure 2D). We validated a subset of these changes in T cells, observing increased Ulk1 and Rb1cc1 protein expression following Torin treatment as well as elevated Akt1 and CD28 levels following CD3/CD28 stimulation (Supplementary Figure 2).

Next, we modeled our data with a full factorial design matrix including interaction terms and evaluated the stimulation-genotype factor interaction (or double differential) terms to find transcriptional programs that change in response to stimulation between cKO, T300A and WT T cells. This allowed us to identify a change in response upon stimulation between the different genotypes. We identified genes associated with lysosomal processes and pathways to be the most highly enriched (Figure 2E). These lysosomal genes exhibited stimulation-dependent response shifts that were altered by genotype. Most of the deregulated lysosome-associated genes showed a downregulation in their response to Torin (i.e., Torin versus baseline) in cKO compared to WT (Figure 2E). These include Lamp2 (69), Cltc (70), and Litaf (71), which have been implicated in autophagosome formation, consistent with the role of Atg16L1 in the formation and maturation of the autophagosome. Other lysosomal genes dysregulated by Atg16L1 loss include Ctsh, Ctsc, Dnase2, Galc, Acp2, Cd68, Pcyox1, and Creg1, which constitute the CLEAR (Coordinated Lysosomal Expression and Regulation) gene network regulated by Tfeb (Transcription Factor EB), the master regulator of lysosome biogenesis and degradation of glycosaminoglycans and sphingolipids (72). Consistent with this observation, we identified glycosphingolipid metabolism genes (St8sia1, Ggta1, B4galt4, B3galnt1, A4galt) that were enriched, nearly all of which exhibited a response to Torin that was dampened in the absence of Atg16L1 (Figure 2E). Additionally, we found cholesterol metabolism, ether lipid metabolism, and arachidonic acid metabolism to be strongly enriched and impacted in an Atg16L1-dependent manner (Figure 2E), revealing an important regulatory influence by Atg16L1 on lipid-associated metabolic pathways and lysosome-associated catabolic processes. We also observed an enrichment of genes functioning in cholesterol metabolism and lysosomal pathways that were downregulated in the absence of Atg16L1 in T cell responses to CD3/CD28 stimulation (Figure 2F) but not as extensive as seen in response to Torin stimulation (Figure 2E). Selecting from the most highly enriched stimulation- and genotype-dependent lysosomal genes, we confirmed decreases in protein expression of Cltc and Lamp2 following Torin treatment as well as Ctsc following CD3/CD28 stimulation in cKO T cells relative to WT (Supplementary Figure 2).

To identify transcriptional mechanisms controlling the deregulated expression states defined by genotype differences, we adopted an approach to infer the potential contribution of transcription factor activity. We applied a competitive gene set test that accounts for inter-gene correlation (46) to assess differences in the expression of transcription factor targets across cKO, T300A and WT T cells. Hsf1 was associated with the differential response between both cKO versus WT and T300A versus WT. In this context, Hsf1 target genes showed a positive fold change under Torin stimulation in the two mutant samples compared to WT (Figure 3A). Hsf1 was previously shown to have an opposing function relative to Nrf2-Keap1 in regulating autophagy (73) and upregulation of its targets suggests that autophagy may be suppressed in Atg16L1 knockout or T300A, which is consistent with previous studies (1). Additionally, activation of Hsf1 upon CD3/CD28 stimulation shows a time dependence, with targets of its paralog Hsf2 being more differentially expressed at the 4hr time point and with Hsf1 activity dominating at the 20hr time point when T cell activation is higher. Under Torin stimulation, Hsf1 target enrichment was found to be higher.

Transcription factors whose targets were downregulated upon Torin stimulation included Tcf4, which has been linked to autophagosome formation (63). A decrease in signal for Tcf4 target genes is consistent with defective autophagosome formation in the cKO context. In Atg16L1 T300A cells, we identified Notch3 as the transcription factor whose targets were most significantly downregulated, consistent with the immunosuppression observed in the mutant cells.

A similar analysis identified differentially regulated transcription factors between cKO, T300A and WT splenic DCs under LPS stimulation. Transcription factors related to the NFκB pathway (Nfkb1, Nfkbia, Rel, Ikbkb) show consistent dysregulation, with targets of inhibitory transcription factors (Ikbkb, Nfkbia) upregulated and those of activating transcription factors (Rel, Nfkb1, Rela) downregulated in cKO and T300A versus WT under the double differential comparisons. This is consistent with a dampened response of immune-related genes in the absence of Atg16L1. In stark contrast to T cells, where we highlighted a positive fold change in targets of Hsf1 (an inhibitor of Nrf3) in both cKO and T300A relative to WT, targets of Nrf3 (Nfe2l2) were enriched in Atg16L1 knockout and T300A DCs (Supplementary Figure 3).

Next, we analyzed the BioPlex protein interaction network (47) with the objective of identifying key nodes in the gene networks of T cells that are disrupted by perturbation of autophagy. We used the DeMAND algorithm for network dysregulation (48), treating the Atg16L1^f/f x Lck Cre⁺ cKO as a genetic perturbation of WT cells. We compared all the non-baseline samples in each genotype group and found that the subnetwork surrounding the Lgals3 (Galectin-3) node was the most dysregulated in cKO T cells compared to WT (Figure 3B). Galectin-3 is a member of the beta-galactoside-binding protein family that has been implicated in inflammation (74), metastasis (75), and apoptosis (76) among other immune-related processes. Notably, Galectin-3 is thought to be regulated by MITF (a paralog of TFEB) in certain cell types and has been reported to play a role in iron trafficking in association with transferrin (77–79). Transferrin itself is the second-most dysregulated node in the T300A versus WT comparison of network disruption, with the most dysregulated node being Zbtb9, a predicted transcriptional regulator with incomplete functional annotation. Other notable genes common to T300A and cKO T cells that were associated with disrupted network modules included the enzyme Rnase3, which regulates antibacterial function (80), and Ifna21, a paralog of Ifna4 that we showed to be regulated at the transcriptional level in a genotype-dependent manner in splenic DCs (Supplementary Figure 3 and Figure 1A).

Cell type and stimulation exert strong influences on genotype differences

We observed not only stimulation-dependent but also cell type-dependent programs regulated by Atg16L1 genotype in T cells and splenic DCs. Subsequently, we sought to quantify the extent to which each of these factors influenced the transcriptional landscape. Toward this end, we expanded the cell types profiled to include bone marrow-derived mast cells obtained from wild type (WT), Atg16L1^f/f x Mast Cre⁺ (cKO), and Atg16L1^T300A/T300A (T300A) mice. Mast cells were stimulated with DNP-BSA for 30 minutes and 1 hour to induce degranulation. These time points were selected based on previously published studies on the kinetics of mast cell degranulation (33–35). As with the previous experiments, splenic DCs were stimulated with LPS at 4h and 12h time points to initiate immune responses. T cells were activated by CD3/CD28 stimulation for 4h and 20h and also stimulated with Torin for 4h to activate autophagy. All RNAseq libraries were resequenced at a comparable depth to minimize the confounding effects from quality control differences.

To ascertain the relative importance of the three factors driving the transcriptional differences observed across the immune cell types and under their respective stimulations, we adopted a tiered dimensionality reduction approach. We first reduced the dimensionality of our dataset using principal component analysis. Principal components (except those most correlated with batch) were then subjected to a second dimensionality reduction step by applying spectral t-SNE (stochastic neighbor embedding). tSNE provided a high-level visual summary of the transcriptomic landscape and revealed that cell type differences dominated over other factors (Figure 4A). In the tSNE, samples cluster most strongly by cell type, followed by stimulation, and then by genotype, indicating that genotype-dependent programs owing to Atg16L1 deficiency or the T300A variant are highly cell type-specific.

Figure 4. Transcriptional responses depend on contexts of cell type, stimulation condition, and genotype.

We examined the extent to which cell type, stimulation, and genotype influence the transcriptional profile of Cd11c⁺ splenic DCs, naïve CD4⁺ CD62L⁺ T cells, and bone marrow-derived mast cells across genotypes as labelled. WT, wild type. cKO, Atg16L1^f/f x Cd11c Cre⁺ for DCs and Atg16L1^f/f x Lck Cre⁺ for T cells. T300A, Atg16L1^T300A/T300A. Cells were unstimulated (NoStim) or subjected to different cell type-relevant stimulations as indicated. An additional summary is available in Supplementary Figure 2. (A) High-level visual summary of the transcriptomic landscape using tandem principal component analysis (PCA) and t-distributed stochastic neighbor embedding (tSNE) suggests that cell type differences dominate over other factors. (B) Variance partition analysis provides a quantitative assessment of the extent to which the variance of each gene can be attributed to cell type, genotype and stimulation state, showing that cell type represents the major contributor to this variance, followed by stimulation and then genotype. This illustrates the importance of context-specific analysis of transcriptomic data and the need to study the effects of Atg16L1 and the T300A variant in each specific cell type separately. (C) Ranking of the 20 top most genotype-dependent genes. Residuals show the rest of the genes that do not significantly vary between groups. Cathepsin E (Ctse) is the only gene that has most of its variance (>60%) explained by genotype. Ctse expression profiles in splenic DCs and T cells are shown in Supplementary Figure 4.

We next conducted variance partition analysis (49) to further quantify the effect of the different factors on the transcriptional profile. A linear model was computed for the expression of each gene individually in the dataset after batch-correction, which allowed us to establish the fraction of variance attributable to genotype, cell type, and stimulation state. Each factor was treated as a random effect by virtue of being a categorical variable. Variance partition analysis also pointed to cell type as the major contributor to variance, followed by stimulation, and then genotype (Figure 4B). Strikingly, we found only one gene (Ctse, Cathepsin E) that was predominantly genotype-dependent, with over 60% of its variance attributable to genotype (Figure 4C). Ctse showed significantly elevated expression in Atg16L1 cKO and T300A splenic DCs and T cells compared to WT (Supplementary Figure 4). For the other top genes exhibiting a substantial genotype dependence, we found that their transcriptional profiles were all markedly influenced by cell type and stimulation factors (Figure 4C). With the exception of Ctse, we have observed that cell type and stimulation states strongly influence the transcriptional landscape and alter the context in which genotype differences can manifest.

Discussion

Our results support the growing recognition that the impact of Atg16L1 genotype differences on autophagy and immunity is strongly dependent on cell type-specific contexts and is further shaped by cellular states triggered by environmental cues. Work from previous studies suggests that environment-gene and gene-gene interactions can modify Atg16L1 genotype-dependent processes involved in antibacterial autophagy (1, 81) and Paneth cell function (82). In splenic DCs, we found that LPS stimulation dramatically altered or polarized the Atg16L1 genotype effect on transcriptional responses to an extent not observed in the unstimulated baseline state. Without LPS stimulation, transcriptional changes owing to genotype differences were unremarkable. Thus, LPS stimulation provided the context dependence or regulatory landscape that allowed genotype differences to manifest. Upon LPS stimulation, immune response genes in DCs were particularly impacted by genotype differences. The extent of downregulation of immune response genes (including cytokines, chemokines, interferon-regulated genes, Toll-like receptors, and inflammasome components) in the absence of Atg16L1 was striking. Steroid biosynthesis genes, however, were upregulated in the absence of Atg16L1. Others have also linked autophagy to innate inflammatory signaling by demonstrating that Atg16L1 is necessary in myeloid (83) and epithelial (84) cells to prevent overproduction of type I interferons. Myeloid cell-specific loss of Atg16L1 increased IFNβ and IL-1β production and enhanced intestinal immunity to Salmonella typhimurium through an interferon receptor-dependent mechanism. Moreover, overproduction of IFNβ and IL-1β was observed in human macrophages carrying Atg16L1 T300A, and an interferon-response gene signature was elevated in IBD patients resistant to anti-TNF therapy (83). Together, these observations indicate that Atg16L1 is required for the pro-inflammatory responses of DCs. Our findings contrast with a previous study in which Atg16L1-deficient macrophages were found to produce higher levels of IL-18 when stimulated with LPS (23). Instead, we observed a very strong suppression of IL-18 expression in Atg16L1-deficient DCs, suggesting that Atg16L1-driven programs are cell type-specific. After LPS stimulation, the T300A variant was associated with an enhanced expression of immune response genes, representing a shift towards an inflammatory state. At baseline, however, there were no significant expression differences detected for immune response and steroid biosynthesis genes between genotypes, reinforcing the notion that genotype- and allele-specific differences are only revealed in appropriate functional contexts.

Compared to DCs, we observed distinct transcriptional programs in T cells under the control of Atg16L1 genotype. Specifically, iron metabolism machinery was dysregulated in Atg16L1^f/f x Lck Cre⁺ cKO T cells in a stimulation-independent manner. Upregulation of ferritin and heme complex genes in the absence of Atg16L1 suggests that a deficiency in Atg16L1 might be associated with an accumulation of iron in the cell. Many genes involved in iron transport were also differentially expressed, suggesting that the mechanism underlying this phenotype might be a disruption of intracellular iron handling at the level of impaired ferritinophagy. Transcriptional alterations associated with lysosomal and autophagosome dynamics were more pronounced upon Torin stimulation compared to CD3/CD28 stimulation, indicating a predominant mTOR signaling-dependence. This is also consistent with TFEB regulation, since many of the dysregulated lysosomal genes constitute the TFEB-regulated CLEAR network that is important for lysosomal biogenesis. Deregulation of the CLEAR network is linked to lysosomal diseases (85). Further, we identified additional transcription factors that drive gene expression changes in cKO and T300A T cells through transcription factor-target enrichment approaches. Notably, Hsf1 targets were upregulated upon both stimulations (Torin and CD3/CD28) and both genotypes (cKO and T300A). Hsf1 is known to function in suppression of autophagy and antagonize Nrf2/Keap1. Additionally, targets of Nr1h3 (LXRA), a transcription factor implicated in induction of lethal autophagy, were downregulated after CD3/CD28 stimulation. This observed effect after CD3/CD28 but not Torin stimulation suggests that regulation of Nr1h3 target genes occurs in an mTOR-independent manner. Lastly, we tested for the disruption of the BioPlex protein-protein interaction network and found key nodes Galectin-3 and Transferrin, both of which have roles in iron-trafficking, to be affected by Atg16L1 deficiency.

Insights from genetics, particularly polymorphisms in Atg16L1 and IRGM, helped to define the complex role for autophagy in IBD. Conditional Atg16L1 knockout and T300A knock-in mice revealed cell type-specific phenotypes that could be described relative to wild type as more mild or severe. The cumulative effects of these cell type-specific phenotypes contribute to IBD pathophysiology. In this study, we extend Atg16L1 genotype- and cell type-dependent observations and report on stimulation-specific programs in splenic DCs, T cells, and mast cells. LPS-induced Ifna5 expression in DCs was abrogated by Atg16L1 loss but enhanced beyond wild type levels by the T300A substitution. We identified genotype- and stimulation-dependent metabolism genes in both DCs and T cells that impair mucosal immunity. Our analysis revealed a feedback loop in DCs in which LPS induced expression of Aoah, a lipid metabolism gene with a known role in LPS inactivation. Altered LPS inactivation in Aoah knockout DCs leads to decreased IL-6 production, which results in inhibition of Th17 cell polarization and induction of regulatory T cells (86). We identified a second genotype- and stimulation-dependent lipid metabolism gene, Ch25h, that was shown to function as part of an axis regulating osteoarthritis (87). This mechanism may have a broader role in the gut. Lastly, we observed enrichment of cholesterol metabolism genes in an Atg16L1-dependent manner in Torin-stimulated T cells. Atg16L1 deficiency and Atg16L1 T300A cause cholesterol accumulation in lysosomes that contributes to defective plasma membrane repair, linking membrane integrity to IBD (88). Comparing transcription factor-enrichment analyses of DCs and T cells in response to stimulation, we revealed transcription factors that behave similarly in the context of Atg16L1 deficiency (Nfkbia) as well as those with opposing behaviors (Nfe2l2) in these cell types. Together, these results provide starting points for understanding the crosstalk between autophagy, inflammation, immunometabolism that maintains tissue homeostasis. IBD is a model disease in this context, as risk genes control interconnected functional pathways in cell type-dependent manners; assigning variants pathway risk scores may help define disease subtypes, biomarkers, and targeted therapeutic strategies (89). Indeed, genomic and transcriptional profiles have been combined with computational approaches to position variants affecting the responsiveness of genes to stimuli (responsiveness quantitative trait loci or reQTLs) within molecular circuits (90).

Collectively, our studies in DCs and T cells provide a more quantitative estimate of the influences of cell type, stimulation, and genotype on transcriptional profiles. Our findings from variance partition analysis demonstrated that cell type and stimulation effects predominate over transcriptional alterations owing to genotype differences. Although expression of many genes showed strong genotype dependence, expression of very few genes exhibited independence from cell type and stimulation state. In fact, we found only one gene (Ctse) whose expression was predominantly genotype-specific across cell types and stimulation states examined. Ctse is a non-lysosomal aspartic protease, with a structure that is almost identical to that of Ctsd (91), a lysosomal gene previously reported to be upregulated in IBD (92, 93). Ctse has been implicated in MHC Class II presentation (94), suggesting that it could also play an important role in IBD development through the disruption of antigen presentation in the context of the T300A risk variant. Ctse had most of its variance attributable to genotype and showed elevated expression in Atg16L1 T300A compared to both WT and Atg16L1 knockout cells. Thus, Ctse transcription appears to track inversely with the T300A variant and could serve as a useful candidate marker. Our finding that Ctse represents the sole exception rather than the rule reinforces the concept that cell type and stimulation states establish the context in which genotype differences can manifest. These observations strongly suggest that studies into disease-associated genotypes and variants be performed and interpreted in the context of each specific cell type and stimulation state.

Key Points.

• Stimulation-based profiling defined genotype- and cell-specific Atg16L1 programs.

• Cell type and stimulation strongly influence Atg16L1 T300A genotypic differences.

• Genotype- and stimulation-dependent lipid metabolism genes impair mucosal immunity.