Dysregulated immunity in PID patients with low GARP expression on Tregs due to mutations in LRRC32

Immune dysregulation diseases are characterized by heterogeneous clinical manifestations and may have severe disease courses. The identification of the genetic causes of these diseases therefore has critical clinical implications. We performed whole-exome sequencing of patients with immune dysregulation disorders and identified two patients with previously undescribed mutations in LRRC32, which encodes glycoprotein A repetitions predominant (GARP). These patients were characterized by markedly reduced numbers and frequencies of regulatory T cells (Tregs). Tregs with mutated LRRC32 exhibited strongly diminished cell-surface GARP expression and reduced suppressor function. In a model of conditional Garp deficiency in mice, we confirmed increased susceptibility to inflammatory diseases once GARP expression on Tregs was decreased. Garp deficiency led to an unstable Treg phenotype due to diminished Foxp3 protein acetylation and stability. Our study reinforces the understanding of the immunological mechanisms of immune dysregulation and expands the knowledge on the immunological function of GARP as an important regulator of Treg stability.


INTRODUCTION
Primary immunodeficiency (PID) is a heterogeneous group of disorders characterized by impaired immune development and function due to single-gene mutations. The clinical manifestations of PID are variable and include severe infections, autoimmunity and malignancies. 1 PID conditions that are associated with autoimmune diseases due to defects in the regulation of selftolerance are referred to as immune dysregulation diseases. 2 Currently, several mutations leading to immune dysregulation, such as FOXP3 mutations causing immune dysregulation, polyendocrinopathy, enteropathy, or X-linked syndrome 3 and cytotoxic T-lymphocyte-associated protein 4 (CTLA4) mutations resulting in CTLA4 haploinsufficiency, 4,5 have been identified. Due to the heterogeneity of these diseases, it is highly likely that mutations in other genes may also cause immune dysregulation. The identification of these mutations has critical clinical implications.
Control of immunological self-tolerance is performed by regulatory T cells that coexpress CD4 and CD25 (Tregs). 6,7 The function of Tregs is critically dependent on the Treg-specific transcription factor Foxp3. [8][9][10] In addition to Foxp3 expression, the expression of glycoprotein A repetitions predominant (GARP), also known as leucine-rich repeat containing 32 (LRRC32) [11][12][13] and encoded by LRRC32 in humans, characterizes Tregs. GARP is an 80 kDa cell-surface protein that contains 20 leucine-rich repeats. 14 It associates with inactive latent transforming growth factor β1 (TGFβ1), [11][12][13] which consists of a TGFβ1 homodimer bound to latency-associated peptide (LAP). GARP releases active TGFβ1 upon the interaction of LAP with the integrin αVβ8 and thereby regulates the bioavailability of TGFβ1. 15,16 Active TGFβ1 released by GARP facilitates the development of additional Tregs or T helper 17 (Th17) cells in a paracrine manner and mediates the immunosuppressive capacity of Tregs. 13,17 In this study, we report two PID patients with previously undescribed LRRC32 mutations suffering from severe immune dysregulation and exhibiting Treg defects. By using conditional Garp-deficient mice, we confirmed increased susceptibility to inflammatory diseases in the absence of GARP and deciphered the underlying molecular mechanism.

Ethics
All patients and healthy individuals provided written informed consent. The study was approved by the Ethics Committee of the Universities of Munich and Freiburg.
Whole-exome sequencing Genomic DNA was purified from human peripheral blood mononuclear cells (PBMCs) using QIAamp kits (Qiagen, Hilden, Germany) according to the manufacturer's protocol. Wholeexome sequencing (WES) was performed using the custom SureSelect exome sequencing protocol from Agilent (Santa Clara, CA). Exomes were enriched by using SureSelect exome v5 probes. Libraries were sequenced twice (two flow cells) on a HiSeq 2500 v4 with a 2 × 76 bp protocol generating four raw sequence data files (FASTQ) per sample. Data preprocessing was performed according to the GATK best practices and involved the following steps: (1) conversion of FASTQ files into an unmapped BAM file (PICARD tool FastqToSam), (2) addition of tags to the Illumina adapter sequences of the unmapped BAM file (PICARD tool MarkIlluminaAdapters), (3) conversion of the unmapped tagged BAM file into a FASTQ file (PICARD tool SamToFastq), (4) alignment to the reference genome build UCSC hg38 (BWA MEM), (5) identification of duplicated reads (MarkDuplicates PICARD), (6) BAM recalibration, and (7) indel realignment. Variant calling was performed with three different variant callers: GATK Haplotype caller, FreeBayes, and SAMtools. BASH and R scripts were subsequently used to (1) merge the VCF files, (2) identify and unify dinucleotide changes, and (3) format the data sets for importation into an in-house specialized SQL database (Gem-maDB) at the Centre of Chronic Immunodeficiency in Freiburg. Variant annotation was performed using Ensembl's Variant Effect Predictor tool (https://www.ensembl.org/info/docs/tools/vep/ index.html), and allele frequency (AF) data were extracted from the gnomAD exome (v2.1.1) and genome (v3) data sets (https:// gnomad.broadinstitute.org/downloads). Individual frequencies were obtained by transforming the gnomAD AF data. Variant filtering was performed by selecting variants with (1) an individual frequency below 1% in both our internal cohort and the gnomAD (exomes or genomes) populations, which included control cohorts, such as those in the NHLBI-GO Exome Sequencing Project or the 1000 Genomes project, (2) a "high" or "moderate" predicted impact, (3) an alternative AF1 larger than 0.3 and read depth larger than 20, and (4) a zygosity matching an autosomal recessive or X-linked recessive mode of inheritance, since there was no family history of disease and de novo variants could not be identified without parents (variants in genes associated with an autosomal dominant condition were also assessed, to not exclude genes with incomplete penetrance, and the results were limited to only one transcript per variant (that with the highest score)). Resulting candidate variants were assessed individually considering gene function and disease role.
T cell isolation For human T cells, PBMCs were isolated by centrifugation of blood over a Ficoll (Biotest, Dreieich, Germany) layer. Overall, 10 ml of Ficoll was loaded under 20 ml blood diluted with 20 ml phosphate-buffered saline (PBS) and centrifuged at 400 × g for 20 min at room temperature. PBMCs were harvested and washed once with PBS. Total CD4 + T cells, naive CD4 + T cells, CD25 -CD4 + T cells and CD25 + CD4 + T cells were purified from human PBMCs by magnetic cell separation (MACS) using the Human CD4 + T Cell or CD4 + CD25 + Regulatory T Cell Isolation Kit (Miltenyi Biotec).
For murine T cells, spleens were homogenized into a single-cell suspension using gentleMACS™ C Tubes and a gentleMACS™ Dissociator (Miltenyi Biotec). Cell populations were purified from splenocytes either by MACS or by fluorescence-activated cell sorting (FACS) on a MoFlo (Beckman Coulter, Brea, CA). Total CD4 + T cells were purified using the Mouse CD4 + T Cell Isolation Kit (Miltenyi Biotec). CD25 + CD4 + and CD25 − CD4 + T cells were purified using the Mouse CD4 + CD25 + Regulatory T Cell Isolation Kit (Miltenyi Biotec). The purity of the cell populations was evaluated by flow cytometry.
Flow cytometry For extracellular staining, cells were incubated with antibodies for 15 min at 4°C, washed with 2% fetal calf serum (FCS) in PBS (both from Life Technologies Invitrogen) and analyzed on a Cytomics FC 500 (Beckman Coulter).
Collagen-induced arthritis Collagen-induced arthritis (CIA) was induced in 12-week-old mice by intradermal injections of 200 µg chicken collagen (Chondrex) in complete Freund's adjuvant (Sigma-Aldrich) followed by a booster injection of 200 µg collagen in incomplete Freund's adjuvant (Sigma-Aldrich) on day 21. The mice were scored every 2-3 days to assess the redness and swelling of each limb (0-3, with a maximum score of 12 per mouse): 0, normal; 1, slight swelling and/or erythema; 2, pronounced edematous swelling; and 3, joint rigidity. Hind paw thickness was assessed on the day of the first injection (day 0) and at the experimental endpoint (day 38).
Experimental autoimmune encephalomyelitis Experimental autoimmune encephalomyelitis (EAE) was induced by subcutaneous immunization of 12-week-old mice with 200 µg MOG  (R&D Systems) in complete Freund's adjuvant and 400 ng pertussis toxin (both from Sigma-Aldrich). On day 2, an additional 400 ng pertussis toxin was administered by subcutaneous injection. The mice were scored daily for signs of paralysis beginning on day 7 after immunization until the experimental endpoint (day 19). The EAE clinical score was assessed as follows: 0, clinically normal; 1, limp tail; 2, weak hind limbs; 3, partially paralyzed hind limbs; 4, complete hind limb paralysis; and 5, death.
Histological staining Tissues were fixed in 4% PFA (Roth) for 8-12 h at room temperature and embedded in paraffin wax using standard histological procedures. Paws were decalcified in 10% EDTA (Sigma-Aldrich), pH 7.2, for 2 weeks at 4°C and washed overnight in 70% ethanol prior to embedding. One-micrometer paraffin sections were stained with hematoxylin (Sigma-Aldrich) and eosin (Merck, Darmstadt, Germany) (H&E staining) for morphological assessment. Immunohistochemical staining for CD3 + cell infiltration was performed as previously described. 20 Tartrateresistant acid phosphatase (TRAP) staining was performed using the Acid Phosphatase, Leukocyte (TRAP) Kit (Sigma-Aldrich) following the manufacturer's instructions.
Analysis of protein stability CHX (Sigma-Aldrich) was dissolved in ethanol (40 mg/ml) and diluted in PBS to generate working solutions. CHX (6 µg/ml, Sigma-Aldrich) was used in cell culture experiments to determine the half-life of Foxp3. Murine T cells were stimulated as described above for 3, 6, 9, and 20 h. Human T cells were stimulated as described above for 6, 20, and 48 h. Foxp3+ cell frequencies were assessed by intracellular flow cytometry. The half-life was calculated by linear regression of normalized Foxp3+ cell frequencies.
Dysregulated immunity in PID patients with low GARP expression on Tregs. . . P Lehmkuhl et al.
In vitro suppression assay Freshly isolated human naive or total CD25 − CD4 + T cells were resuspended in PBS at a concentration of 100 × 10 6 /ml and labeled with 10 µM CFSE (Life Technologies Invitrogen) for 8 min at room temperature. After labeling, the cells were resuspended in 10% NHS/RPMI 1640 medium (Life Technologies Invitrogen). A total of 50 × 10 3 CFSE-labeled CD25 − CD4 + T cells were stimulated in a 96-well round-bottom plate with 1 µg/ml soluble anti-CD3 antibody in the presence of 100 × 10 3 CD4 − CD8 − PBMCs and increasing amounts of Tregs for 4 days.
LRRC32 allelic expression was examined by allele-specific PCR analysis using assays-by-design for single-nucleotide polymorphism genotyping (Thermo Fisher Scientific). The assay consisted of two unlabeled PCR primers (forward and reverse), a VIC/minor groove binder (MGB)-labeled probe detecting the major allele sequence and a FAM/MGB-labeled probe for the minor allele sequence. The dye emission was measured after 40 cycles.
Gene array Transcriptome analysis was performed on a whole mouse genome oligo microarray (Mouse GE 4x44K v2) (Agilent Technologies) using mRNA purified from FACS-sorted Garp-deficient and control Tregs from five mice per group. The intensity data were subjected to quantile normalization. Fold changes in normalized data were calculated to identify genes with greater than twofold differential expression; log2 intensities of selected genes were mediancentered and visualized by Mayday (version 2.30) using centering for display. Molecular imaging Molecular images were generated with PyMOL (The PyMOL Molecular Graphics System, Schrödinger, LLC) using the PDB file 6GFF.
Statistical analysis Statistical analysis was performed by using Student's t-test. The Kolmogorov-Smirnov test was used to test for a normal distribution. p values less than 0.05 were considered statistically significant and are indicated as *p < 0.05, **p < 0.01, and ***p < 0.001 (n.s.-not significant).

RESULTS
Identification of PID patients with LRRC32 mutations By routine screening of our patients at the immunodeficiency clinic using WES, we identified two patients bearing previously undescribed mutations in LRRC32 (Fig. 1A, Table 1, Supplementary  Tables 1 and 2). Patient 1 had a total of 135,084 variants with respect to the reference genome (build hg38). Subsequent filtering steps resulted in 30 heterozygous candidate variants in genes previously associated with autosomal dominant conditions and 17 compound heterozygous or homozygous variants in genes not previously associated with disease or associated with autosomal recessive conditions. Of these, two heterozygous variants in the gene LRRC32 were the most promising candidates: a novel nonsense mutation (p.Trp247Ter) and a rare missense variant (p.Arg312Cys) (Fig. 1B). Patient 2 had a total of 108,522 variants with respect to the reference genome. Subsequent filtering steps resulted in 33 heterozygous candidate variants in genes previously associated with autosomal dominant conditions and 29 compound heterozygous or homozygous variants in genes not previously associated with disease or associated with autosomal recessive conditions. Of these, two rare heterozygous missense variants in the gene LRRC32 were the most promising candidates: p.Arg421Gln and p.Arg312Cys (Fig. 1B). Sanger sequencing of these patients proved that these variants were in the same allele in patient 1 and in different alleles in patient 2.
Based on the recently resolved crystal structure of the GARP/ latent TGFβ1 complex, 21 we predicted mutated protein structures (Fig. 1C). Prediction of the p.Trp247Ter protein structure resulted in a truncated protein missing the transmembrane region. The predicted p.Arg312Cys structure produced a full-length protein with the exchange of a hydrophilic, positively charged Arg at position 312 to uncharged Cys. The predicted p.Arg421Gln structure was also a full-length protein with the exchange of positively charged Arg to uncharged Gln. Both the Arg312Cys and Arg421Gln exchanges were found to be located at the convex side of the horseshoe and to not interfere with the binding faces for the latent TGFβ1 complex of GARP. However, the prediction of protein stability based on the free energy change (ΔΔG) 22 for both mutants resulted in destabilizing ΔΔG values much below −0.5 Kcal/mol, −2.93 Kcal/mol for p.Arg312Cys and −2.79 Kcal/mol for p.Arg421Gln, suggesting lower protein stability and therefore lower protein expression. Indeed, analysis of mutated GARP protein expression using corresponding plasmids revealed no expression for the p.Trp247Ter-encoding construct and strongly diminished expression of the p.Arg312Cys and p.Arg421Gln variants (Fig. 1D).
We next analyzed the allelic expression of LRRC32 in Tregs from the patients (Fig. 1E). Allele-specific PCR assays for c.741G>A and c.1262G>A were generated and used for the analysis of LRRC32 mRNA expression. Unfortunately, because of the nucleotide sequence, no assay could be generated for c.934C>T. However, because of duplexing of probes for mutant and wt alleles, we were able to completely analyze allelic LRRC32 expression: as both FAM and VIC emissions were detected in the c.741G>A assay performed with Tregs from patient 1 and in the c.1262G>A assay performed with Tregs from patient 2, we concluded that LRRC32 transcription from both alleles occurred in both patients. Tregs from patient 1 therefore expressed wt GARP on the surface, while Tregs from patient 2 expressed the p.Arg312Cys and p.Arg421Gln GARP variants on the surface.
Clinical and immunologic characteristics of the patients At the time of investigation, patient 1 was a 65-year-old male suffering from diminished IgG levels and subsequent recurrent sepsis and life-threatening intestinal obstructions. He was diagnosed with common variable immunodeficiency (CVID) at age 54. Lymphocyte infiltration was observed in the gut mucosa but was not as profound as that usually seen with enterocolitis associated with CVID. At the time of blood collection, the patient was receiving intravenous immunoglobulin substitution and prophylactic antibiotics. Patient 2 was a 49-year-old female suffering from recurrent upper respiratory tract infections and diarrhea due to chronic autoimmune granulocytopenia with autoantibodies against granulocytes. Although prescribed, at the time of blood collection, she was not taking prophylactic antibiotics.
Both patients were lymphopenic with particularly reductions in CD4 + T cells (Fig. 2A). Moreover, Treg frequencies were strongly Fig. 1 Identification of patients with mutant LRRC32 variants. A The filtering strategy for exome sequencing data with indication of variant numbers at each step is demonstrated. freq., individual frequency in our internal cohort and in gnomAD exomes and genomes; hi/mo, predicted high or moderate impact of a mutation on gene function/structure based on the type of variant (e.g., a "stop_gained" variant is predicted as "high", whereas a missense variant is predicted as "moderate"); AF1 frequency of reads with the alternative allele, DP read depth (number of reads covering a variant), MOI mode of inheritance, AD autosomal dominant, AR autosomal recessive, XLR X-linked recessive. B Schematic presentation of LRRC32 alleles in patient 1 and patient 2. Affected nucleotides and amino acids and their positions are indicated and denoted by green and red for the wt and mutant variants, respectively. C The predicted protein structure of each mutant GARP proteins in complex with TGFβ1 and LAP is depicted. GARP is indicated in green, TGFβ1 is indicated in orange, and LAP is indicated in gray. Mutated amino acids are indicated in red and noted with arrows. The dashed line indicates part of GARP with a nonresolved crystal structure. N-ter Nterminus, C-ter C-terminus, er extracellular, ic intracellular. D Expression of mutated GARP proteins on the surface of transfected HEK293 cells after staining for GARP. Representative staining out of three independent experiments with similar results is shown. E Analysis of allelic LRRC32 expression in patients. Total mRNA was purified from Tregs and analyzed by allele-specific PCRs for c.741G>A and c.1262G>A in triplicate. Mean values are indicated. Results from patients are indicated in red. Individual results from seven healthy controls are indicated in gray. The probes for the wt and mutant alleles were labeled with VIC and FAM, respectively, and duplexed in one reaction. Shaded areas indicate background fluorescence diminished in both patients (Fig. 2B). Surface GARP protein expression is upregulated in response to T cell receptor (TCR) stimulation. 23 We therefore measured GARP expression on Tregs activated for 2 days (Fig. 2C). As a control, we measured the surface LAP expression. Tregs from both patients demonstrated markedly low GARP levels on the surface. Correspondingly, LAP expression was also diminished. Interestingly, however, LRRC32 mRNA expression was increased in freshly isolated cells (day 0) or at the upper limit of the normal range on day 2 (Fig. 2D). Next, we analyzed the suppressor function of Tregs from these patients (Fig. 2E, F). We used either naive or total CD25 − CD4 + T cells as the effector T cell (Teff) population. Independent of the effector population used, proportionally more Teffs entered proliferation in the presence of Tregs from patients than in the presence of those from healthy donors, indicating that Tregs with lower GARP expression exhibited a reduced suppressor function. We therefore concluded that mutations in LRRC32 might lead to diminished GARP levels on the surface of Tregs and consequently to a reduction in Treg numbers and moreover to Treg dysfunction.
Garp deficiency in mice leads to an inflammatory phenotype Because of the markedly diminished expression of GARP on the surface of patient Tregs, we analyzed Garp knockout mice (C57BL/ 6.Lrrc32 fl/fl ;Cd4-Cre) to gain deeper insights into molecular mechanisms beyond Treg dysfunction. Tregs from these animals did not express GARP on the surface (Fig. 3A). Interestingly, in contrast to the patients, the frequencies of Foxp3 + Tregs were not diminished in these mice (Fig. 3B). However, knockout mouse Tregs expressed reduced levels of Foxp3. Consistent with the patient findings, Garp-deficient mice demonstrated reduced T cell counts and lymphocyte infiltration in various tissues (Fig. 3C, D).
We next investigated the susceptibility of these animals to autoimmune disorders. Garp-deficient mice, in contrast to control wt animals, were highly susceptible to CIA (Fig. 4A). They developed severe aggravated arthritis (Fig. 4B) with markedly increased paw thickness (Fig. 4C). The severe inflammation in Garp-deficient mice was confirmed by histological analysis, displaying a high degree of lymphocyte infiltration into the joint cleft, increased osteoclast differentiation, and severe cartilage damage (Fig. 4D). Similar to the results for CIA mice, Garp-deficient mice were 100% susceptible to EAE and developed severe paralyzing disease (Fig. 4E, F). Furthermore, Garp-deficient mice developed rapid and devastating colitis in response to DSS administration, characterized by a high clinical score and lymphocyte infiltration into the mucosa (Fig. 4G-J). Thus, in mice, low GARP expression on the surface of Tregs leads to a phenotype characterized by immune dysregulation.
GARP controls Treg phenotype stability We then analyzed the regulatory capacity of Garp-deficient Tregs. Tregs were purified from either Garp-deficient or control mice (Fig. 5A) and transferred together with CD45.1 + CD25 − CD4 + Teffs into Rag1-deficient mice. One group of animals received Teffs only. The expansion of CD45.1 + CD25 − CD4 + Teffs in the spleen was analyzed after 7 days (Fig. 5B). As shown in Fig. 5C, strong Teff expansion was detected in animals without Treg cotransfer. The addition of Tregs from control animals significantly inhibited the proliferation of Teffs, whereas the addition of Garp-deficient Tregs had little effect. Remarkably, the insufficient control of Teff expansion by Garp-deficient Tregs correlated with the markedly lower recovery of Garp-deficient Tregs (Fig. 5D). Both the number and frequency of recovered Garp-deficient Tregs were markedly diminished compared to those of control Tregs.
Correspondingly, in in vitro experiments, we found higher frequencies of early apoptotic cells in CD25 + CD4 + T cells but not in CD25 − CD4 + T cells or CD19 + B cells from Garp-deficient mice (Fig. 5E). Moreover, significantly higher numbers of Garp-deficient Tregs than control Tregs entered late apoptosis in culture (Fig. 5F).  Moreover, Garp-deficient Tregs lost Foxp3 expression much faster than control Tregs (Fig. 5G). When we determined the half-life of Foxp3 by using CHX, we found that the half-life of Foxp3 was 11.9 ± 1.0 h in nonstimulated Garp-deficient Tregs and 14.9 ± 0.56 h in nonstimulated control Tregs (p = 0.02) (Fig. 4H).
Stimulation of Tregs with anti-CD3 and anti-CD28 antibodies did not stabilize Foxp3 in Garp-deficient Tregs but rather accentuated the difference between Garp-deficient and control Tregs. ( Fig. 5I). Thus, as Foxp3 is the key transcription factor of Tregs, these data indicate a critical function for GARP in Foxp3 stability and therefore in Treg stability.
GARP regulates the expression of Hdac9 and the stability of Foxp3 by acetylation To gain insights into how GARP might regulate Treg phenotypic stability, we compared the transcriptomes of Garp-deficient and control Tregs (Fig. 6A-E, Supplementary Table 3, Supplementary  Fig. 1). Consistent with our findings on apoptosis, significantly diminished expression of Bcl2, encoding an antiapoptotic molecule, but increased expression of Bik, encoding a proapoptotic factor, and several caspase genes were observed in Garpdeficient Tregs (Fig. 6A). The expression of multiple genes involved in regulating the cell cycle was also deregulated in Garp-deficient Tregs (Fig. 6B). Moreover, the expression of the Treg-specific genes Cd25, Nrp1, Ctla4, Il-10, and Hdac9 was diminished, while the expression of several effector cytokines was increased in Garp-deficient Tregs (Fig. 6C, D). Most intriguingly, however, the expression of the Treg-characteristic histone deacetylase Hdac9 was increased, whereas the expression of other histone deacetylases and histone acetyltransferases was not significantly different (Fig. 6E). Hdac9 is a histone deacetylase highly expressed in Tregs. It regulates Treg function and stability by deacetylating Foxp3. 24 Augmented expression of Hdac9 in Garp-deficient Tregs might therefore lead to Foxp3 instability due to increased Foxp3 deacetylation. The results of a transcriptome analysis for Cd25, Nrp1, Ctla4, and Il-10 were confirmed at the protein level by flow cytometry (Fig. 6F, G), and those for Hdac9 and several other Hdacs were confirmed by PCR analyses (Fig. 6H). We then analyzed the acetylation level of Foxp3 by using a FRET antibody pair specific for Foxp3 and AcK (Fig. 6I). Remarkably, the frequency of Tregs that expressed acetylated Foxp3 was significantly decreased in Garp-deficient Tregs compared to control Tregs, strongly suggesting that the upregulation of Hdac9 results in diminished Foxp3 acetylation.
GARP controls autocrine TGFβ1 homeostasis in Tregs Given that GARP functions as a receptor of latent TGFβ1, we hypothesized that the decreases in Foxp3 acetylation and stability in Garp-deficient Tregs resulted from the deregulation of TGFβ1 signaling. Indeed, basal phosphorylation of SMAD2 and SMAD3 was significantly diminished in Garp-deficient Tregs compared to control Tregs (Fig. 7A). However, the response of Garp-deficient Tregs to exogenous TGFβ1 was not altered (Fig. 7B). Thus, we treated Garp-deficient Tregs with TGFβ1 for 20 h and analyzed Hdac9 expression and Foxp3 acetylation. Hdac9 expression in Garp-deficient Tregs was significantly reduced in response to TGFβ1 (Fig. 7C). Moreover, treatment with TGFβ1 elevated the frequency of Tregs expressing acetylated Foxp3 in Garp-deficient Tregs to the frequency seen in control Tregs (Fig. 7D). Furthermore, the half-life of Foxp3 in Garp-deficient Tregs was normalized (23.83 ± 2.26 h in Garp-deficient Tregs vs. 23.59 ± 2.31 h in control Tregs, p = 0.77) (Fig. 7E, right panel) when cultured in the presence of TGFβ1 prior to CHX incubation, whereas the half-life of Foxp3 in Garp-deficient Tregs cultured in the absence of TGFβ1 remained significantly diminished (17.95 ± 1.28 h vs. 14.33 ± 1.36 h, p = 0.016) (Fig. 7E, left panel). We therefore concluded that the lack of GARP expression in Garp-deficient Tregs reduces the TGFβ1 signaling necessary for the formation of a stable Treg phenotype.
Tregs from patients demonstrate signs of TGFβ1 deprivation Finally, we analyzed SMAD2/3 phosphorylation in Tregs from the PID patients with mutated LRRC32. Similar to Garp-deficient mice, the patients showed a diminished basal level of SMAD2/3 phosphorylation despite normal SMAD3 amounts in Tregs compared to healthy controls (Fig. 8A). Furthermore, Tregs from both patients expressed elevated levels of HDAC9 (Fig. 8B). Moreover, analysis of Treg stability using CHX revealed that the half-life of Foxp3 in Tregs from pt 1 was 46.28 h and that in Tregs from pt 2 was 41.64 h compared to 95.46 ± 19.13 h in Tregs from healthy controls. Thus, the Foxp3 half-life in Tregs from patients with mutated LRRC32 was markedly diminished. These observations are consistent with the findings from Garp-deficient mice. They confirmed that diminished GARP expression on the surface of Tregs causes disrupted TGFβ1 homeostasis, leading to Foxp3 instability and Treg dysfunction.

DISCUSSION
In this study, we demonstrate that diminished GARP expression contributes to immune dysregulation. It is apparent that the immune system of both patients with LRRC32 mutations cannot properly resolve an immune response. This deficiency led to chronic inflammation, collapse of peripheral tolerance, and disease manifestation. We observed persistent inflammation of the gut mucosa in patient 1 and the development of antineutrophilic antibodies and complete elimination of neutrophils in patient 2. Remarkably, in our cohort of PID patients, we identified two patients with mutated LRRC32. Whereas c.934C>T and c.1262G>A were previously detected in the general population    ( Table 1), c.741G>A is a novel mutation. It is also remarkable that the presence of the mutations was found to contribute to a strong decrease in or a complete lack of GARP expression. It is therefore believable that a distinct amount of GARP on the surface of Tregs is necessary to ensure the proper biological function of GARP. In this regard, Nasrallah et al. 25 reported that the risk locus 11q13.5 associated with immune-mediated diseases contains an enhancer of LRRC32. The authors showed that in Tregs, this enhancer forms conformational interactions with the promoter of LRRC32 and upregulates GARP expression. Several enhancer risk variants were associated with reduced GARP expression. Similar to our finding, the authors concluded that a distinct level of GARP expression is critical for proper Treg function. This conclusion is further supported by our previous observations of a decreased level of GARP on Tregs from patients with rheumatoid arthritis. 26 Moreover, for atopic dermatitis, a significant excess of low-frequency missense mutations in LRRC32 leading to low GARP expression has been reported. 27 Finally, it has been shown that Garp-deficient Tregs facilitate antitumor immunity but are unable to suppress pathogenic T cell responses in a T cell transfer model of colitis, pointing once again to impaired suppressive Treg function in the absence of GARP. 28 Interestingly, both patients developed immune dysregulation disorders not in childhood but rather at ages resembling the age of onset of many autoinflammatory diseases. In a pathologic adult autoinflammatory situation, it is supposed that an environmental factor triggers disease development in individuals with a genetic susceptibility to the disorder. Similar situations are conceivable for both PID patients. In this regard, naive T cells from both patients appeared to be less proliferative than naive T cells from healthy controls (Fig. 4E). We conclude that this might be an indication of disease chronification, where perpetuated autoinflammation during life impacts the activation of bystander naive T cells. This assumption is supported by the findings from Garpdeficient mice. Garp-deficient mice did not exhibit a spontaneous phenotype but developed vigorous inflammatory disorders in response to a trigger. Our data therefore identify LRRC32 as a genetic factor determining susceptibility to inflammatory disorders and point to its important function in the integrity of the immune system.
Our data show that the immunological phenotype observed in the absence of GARP occurred due to a reduced availability of TGFβ1 for Tregs. We therefore report for the first time evidence for an autocrine mechanism of action of GARP-bound TGFβ1. Based on previous investigations, it was concluded that GARP-bound TGFβ1 facilitates the development of other Tregs or Th17 cells in a paracrine manner rather than being utilized by Tregs in an autocrine manner. 13,17 This conclusion was generated based on the observations that Garp-deficient mice have a normal frequency of Tregs and that Garp-deficient Tregs exhibit normal suppressive function in vitro. 29 Moreover, a normal suppressive capacity was observed for Tgfb1-deficient Tregs in vitro. 29,30 In contrast, analysis of the suppressive capacity of Tgfb1-and Garpdeficient Tregs in vivo revealed their inability to inhibit colitis in a transfer model of inflammatory bowel disease. 28,31 Consistently, we observed an apparently normal suppressive capacity for Garpdeficient Tregs in vitro but insufficient control of effector T cell expansion after adoptive transfer in vivo.
Marked alterations in the gene expression profile are the most impressive indication of how important GARP, and consequently GARP-associated TGFβ1, is for Tregs. Among genes with altered expression, Hdac9 is especially notable. HDAC9 is distinctively expressed in Tregs and regulates Foxp3 acetylation. 24 Acetylation of Foxp3 prevents its polyubiquitination and proteasomal degradation, thus maintaining Foxp3 protein stability. 32 In addition, acetylation of Foxp3 is necessary for its transcriptional suppressor function. 33 We propose therefore that in Garp-deficient Tregs, in the presence of abundant HDAC9, Foxp3 becomes less acetylated and therefore less stable. Using FRET analysis, in which the fluorescence signal for Foxp3 could be detected by flow cytometry only if an acetyl group-binding antibody and an anti-Foxp3 antibody bound to their targets in immediate proximity, we observed diminished Foxp3 acetylation in Garp-deficient Tregs. In experiments with CHX, the diminished half-life of Foxp3 in Garpdeficient Tregs was confirmed. Both the diminished level of Foxp3 acetylation and the diminished Foxp3 half-life could be restored by exogenous TGFβ1. Furthermore, the abundant Hdac9 mRNA expression was markedly reduced by exogenous TGFβ1.
Destabilization of Garp-deficient Tregs through reduced Foxp3 acetylation is probably the central event determining the unstable phenotype of Garp-deficient Tregs. The development of Tregs occurs during thymopoiesis and relies on the recognition of selfantigens presented by thymic medullary antigen-presenting cells in the context of MHC class II molecules. [34][35][36] Considering that Tregs express potentially self-reactive TCRs, it is conceivable that reduced Foxp3 expression might have harmful consequences for the host. Indeed, several studies have demonstrated that the downregulation or loss of Foxp3 is associated with a loss of the regulatory phenotype and acquisition of an effector phenotype in Tregs. [37][38][39] In Garp-deficient Tregs, the expression of all cytokine types, including the Th1, Th2, and Th17 types, was increased, indicating a general deregulation of cytokine expression resulting from diminished Foxp3-mediated transcriptional repression. Interestingly, the phenotype of Garp-deficient Tregs resembles the phenotype of Tregs from mice with genetically engineered attenuated Foxp3 expression. 40 Due to decreased Foxp3 expression, Tregs from these animals exhibit a reduced regulatory function and express effector cytokines. Our findings are therefore in line with the hypothesis that a certain level of Foxp3 is necessary for the maturation of Tregs with a stable phenotype. It has been shown that mice prone to lupus nephritis exhibit Treg instability due to diminished Foxp3 expression. 41 Moreover, associations between diminished Foxp3 expression and human immune disorders have been shown. 42,43 Thus, taken together, our data demonstrate that in the absence of GARP, Tregs develop signs of TGFβ1 deprivation characterized by increased expression of Hdac9 and subsequent destabilization of Foxp3, resulting from reduced acetylation. GARP therefore plays a central role in the stabilization of the immune system.