A human immunodeficiency syndrome caused by mutations in CARMIL2

Human T-cell function is dependent on T-cell antigen receptor (TCR) and co-signalling as evidenced by immunodeficiencies affecting TCR-dependent signalling pathways. Here, we show four human patients with EBV+ disseminated smooth muscle tumours that carry two homozygous loss-of-function mutations in the CARMIL2 (RLTPR) gene encoding the capping protein regulator and myosin 1 linker 2. These patients lack regulatory T cells without evidence of organ-specific autoimmunity, and have defective CD28 co-signalling associated with impaired T-cell activation, differentiation and function, as well as perturbed cytoskeletal organization associated with T-cell polarity and migration disorders. Human CARMIL2-deficiency is therefore an autosomal recessive primary immunodeficiency disorder associated with defective CD28-mediated TCR co-signalling and impaired cytoskeletal dynamics.

S tudying primary immunodeficiency disorders (PID) has provided fundamental insights into principles governing human tolerance and immunity 1 . Co-signalling disorders such as CD40LG, SAP, OX40 or CD27-deficiency have highlighted unique properties of human host defence 2 . These disorders impair immune cell interactions and are associated with selective susceptibility to infectious agents [3][4][5] , but also shed light on the critical role of anti-tumour immune surveillance. In particular, insufficient T-cell immunity against human gamma herpesvirus-infected cells can lead to malignancy 6 . Whereas SAP-deficiency and CD27-deficiency predispose specifically to Epstein-Barr virus (EBV) infection and EBV-induced lymphoproliferative disorders, OX40-deficiency is associated with Kaposi's sarcoma induced by Kaposi's sarcoma herpesvirus 3,7 .
T cells are central in adaptive immunity, and T-cell signalling and co-signalling govern immune defence, immune homoeostasis and immune surveillance. On antigen encounter, the T-cell antigen receptor (TCR) becomes engaged and in concert with context-sensitive co-receptors, such as CD28, ICOS and OX40, induces T-cell activation, differentiation and function 8,9 . Although TCR-dependent signalling is one of the most studied signalling cascades, knowledge of CD28 co-signalling is incomplete 8,9 . In a murine N-ethyl-N-nitrosourea-mutagenesis screen, the lymphoid lineage-specific actin-uncapping protein Rltpr (RGD motif, leucine-rich repeats, tropomodulin domain and prolinerich containing; murine homologue to human CARMIL2) was identified to be essential for CD28 co-signalling and regulatory T (T reg ) cell development 10 . Mice expressing mutated Rltpr were not able to transduce CD28 co-signalling to its effector protein kinase c theta (Pkcy). Consequently, Rltpr-mutant mice were specifically and completely devoid of T reg and thus functionally phenocopied Cd28 À / À mice 10 . Furthermore, truncated Rltpr expression was associated with increased antibody-dependent CD28 internalization probably reflecting deregulated cytoskeletal dynamics 10 .
Human CARMIL2 (RLTPR) was first described as a gene downregulated in cutaneous lesions of patients with psoriasis vulgaris 11 . The CARMIL family includes three human paralogs: CARMIL1 (LRRC16A), CARMIL2 (LRRC16C) and CARMIL3 (LRRC16B). All proteins have a common domain architecture, that is, one N-terminal pleckstrin homology (PH) domain, several leucine-rich repeats (LRR), one capping protein binding region (CBR) domain and one C-terminal proline-rich domain (PRD) 11,12 . The CBR domain reduces the affinity of CP for actin filament barbed ends and makes them accessible for further actin polymerisation 13 . In contrast to CARMIL1 and CARMIL3, CARMIL2 orchestrates cell polarity by modulating microtubules and intermediate filaments 12 .
In this study, we describe homozygous loss-of-function CARMIL2 mutations in four human patients from two different families presenting with the highly unusual phenotype of disseminated EBV þ smooth muscle tumours (SMT). Our phenotypic and functional studies demonstrate defective CD28 co-signalling with impaired T-cell activation, differentiation and function, and defective cytoskeletal organization associated with T-cell polarity and migration disorders.

Results
Case reports. Patients P1.1 and P1.2 were born from a consanguineous Yemenite couple. At the age of 2 years, P1.2 presented with severe failure to thrive, chronic diarrhoea, recurrent skin and upper airway infections (Table 1). At age 4, P1.2 was diagnosed with an EBV þ latency type III SMT in the liver (Fig. 1, Supplementary Fig. 1 and Supplementary Table 1).
Tumour staging revealed additional 30 EBV þ SMT in the colon, terminal ileum and medulla oblongata ( Fig. 1 and Table 1). Liver and brain EBV þ SMT were excised and the patient was treated by an individualized chemotherapy regimen with oral cyclophosphamide. However, disease progressed with additional lesions appearing in the brain, liver and spleen. Considering a therapyrefractory course, the family moved back to Yemen where the patient deceased.
At the age of 7 years, P1.1 presented with failure to thrive, recurrent upper airway infections, eczematous dermatitis and skin warts. Colonoscopy and histopathology revealed several EBV þ SMT (Table 1 and Supplementary Fig. 1). Further tumour staging and treatment were hindered by the family's decision to return to Yemen. Five years later, P1.1 is reported to have progressive disease.
Family 1 has six additional children and sibling 2 (S1.2) reportedly had died at age 19 in Yemen from a similar disease while the other siblings are healthy (Fig. 2a).
P2.1 and P2.2 were born from a consanguineous Brazilian couple and have been reported as cases of familial infantile myofibromatosis (Fig. 2e) 14 . At the age of 1 year, P2.1 developed chronic diarrhoea and, from age 4 on, repetitive infections with Giardia spp., pneumonias and skin warts (Table 1). At age 8, failure to thrive was documented and consecutively multiple gastrointestinal, pulmonary and a solid liver EBV þ SMT were detected (Table 1). P.2.1 received antineoplastic treatment with methotrexate and vinblastine, but disease progressed and he deceased at age 14.
P2.2 presented with recurrent chronic eczematous dermatitis from the age of 1 year on. At age 5 and 6, he had pneumonia and chickenpox, respectively. At age 6, failure to thrive and multiple liver and gastrointestinal EBV þ SMT were documented. At age 11, he had a second episode of chickenpox (Table 1). Despite antineoplastic treatment with methotrexate and vinblastine, the disseminated soft tissue tumours were progressing and he deceased at age 12.
In summary, the patients presented with chronic diarrhoea and failure to thrive, EBV þ SMT, recurrent viral and parasitic infections and dermatitis ( Fig. 1, Supplementary Fig. 1 and Table 1).
Immune phenotype and T-cell EBV-response. As the patients' clinical course and EBV þ SMT pointed towards defective T-cell immunity, we performed immune phenotyping (Supplementary  Tables 2 and 3) 15 . Total peripheral T, B and NK cell counts mostly were within normal limits (Supplementary Tables 2b  and 3b), but there was a profound reduction of T reg (Fig. 3a). P2.1 and P2.2 had increased recent thymic emigrants (RTE) (Supplementary Table 3d) and all patients had a significant increase in naive helper T cells (CD4 T N ) and insufficient gain of central memory helper T cells (CD4 T CM ) and central memory cytotoxic T cells (CD8 T CM ) ( Fig. 3b and Supplementary  Table 3e). CD27 þ IgM À IgD À switched memory B cells were at the lower limit or reduced (Supplementary Tables 2c and 3c). IgG 1 and/or IgG 4 were reduced in P1.2 or P1.1, respectively, T-cell dependent-specific antibodies were reduced in all and EBV-seroconversion was incomplete in P1. 1 To analyse the patients' T-cell response to EBV, we co-cultured peripheral blood T cells from P2.1 and P2.2, and two EBV þ healthy donors (HD) as controls, with autologous irradiated EBV þ lymphoblastoid cell lines (LCL) over 10 passages. We found that expansion rates of LCL stimulated T-cell lines (TCL) from P2.1 and P2.2 were significantly lower than from HD ( Supplementary Fig. 2a). TCL from P2.1 and P2.2 did not secrete IFN-g on stimulation with autologous LCL but secreted IFN-g on stimulation with phytohemagglutintin (PHA) ( Supplementary  Fig. 2b). IFN-g secretion of P2.2 TCL after stimulation with human leukocyte antigen (HLA) mismatched LCL suggested that recognition was not major histocompatibility complex (MHC)restricted ( Supplementary Fig. 2b). TCL from P2.1 and P2.2 were unable to secrete IL-4, even after stimulation with PHA ( Supplementary Fig. 2c).
In summary, immune phenotypic and functional studies were indicative of a T-cell deficiency with particular absence of T reg , insufficient gain of T-cell memory and deficient T-cell response to EBV.
Genetics. We hypothesized that the T-cell deficiency followed an autosomal recessive Mendelian inheritance pattern with full penetrance (Fig. 2a,e). For family 1, we performed SNP-based homozygosity mapping including the father (F1), mother (M1),  Contour plots of CD4 and CD8 naive (T N , CD45R0 À CD27 þ ), central memory (T CM , CD45R0 þ CD27 þ ), effector memory (T EM , CD45R0 þ CD27 À ) and effector (T EFF , CD45R0 À CD27 À ) T-cells (corresponding percentages are indicated in each square) and summary of CD4 and CD8 T-cell subtype percentages for six HD and four patients. Small horizontal lines indicate the median. Each symbol represents an individual donor. Data are representative for four independent experiments with n ¼ 2. Significance levels are calculated with Welch's t-test and indicated in the summary graphs (NS ¼ non significant). the disease causing gene 18,19 . We confirmed the autosomal recessive CARMIL2 c.489insG nonsense variant (p.E163fs*4) by Sanger sequencing and found segregation with the disease phenotype (Fig. 2a,c,h). CARMIL2 immunoblotting in P1.1 and P.1.2 LCL showed absent protein expression (Fig. 2d).
For family 2, we analysed the autosomal recessive CARMIL2 c.871 þ 1G4T splice-site variant that had been reported as a WES finding 20 . We confirmed CARMIL2 c.871 þ 1G4T by Sanger sequencing and found segregation with the disease phenotype (Fig. 2e,f). In silico analysis predicted the splice-site variant to cause skipping of exon 11 resulting in a frame shift and a premature stop codon (p.D260fs*70) (Fig. 2h). Sanger sequencing of P2.1 CARMIL2 cDNA detected a splice product lacking exon 11 and CARMIL2 immunoblotting in P2.1 and P2.2 LCL showed absent protein expression (Supplementary Fig. 3 and Fig. 2g).
Thus, in both families homozygous loss-of-protein-expression CARMIL2 mutations segregated with the disease phenotype.
CD28-dependent T-cell function and NK and T-cell cytotoxicity. Since murine Rltpr is essential for CD28 co-signalling, we first determined CD28 expression in CARMIL2-deficient patient CD4 and CD8 T cells and found normal surface levels ( Supplementary Fig. 4a) 10 . To assess CD28-dependent T-cell activation, we stimulated CARMIL2-deficient PBMC with anti-CD3, anti-CD3/CD28 or PMA/ionomycin (P/I) and measured surface expression of the early activation antigen CD69, the highaffinity interleukin-2 receptor alpha chain CD25 and the T reg master transcription factor FOXP3. Patient CD4 and CD8 T cells showed normal TCR-dependent induction of CD69 and increased CD69 expression after CD28 co-stimulation (Fig. 4a). In contrast, patient CD4 and CD8 T cells lacked upregulation of CD25 in response to anti-CD3/CD28 (Fig. 4a). Furthermore CD25 and FOXP3 co-expression was absent in resting patient CD4 T cells and could not be induced after anti-CD3 and anti-CD3/CD28 stimulation (Fig. 4b).
To assess CD28-dependent T-cell proliferation and activationinduced cell death (AICD), we applied the same stimulation conditions and analysed CFSE-dilution in combination with CD25 surface expression and 7-AAD cell viability staining. In HD and patient CD4 and CD8 T cells, proliferation was modestly induced after anti-CD3, but only HD T cells showed a robust proliferative response and concomitant CD25 surface expression after CD28 co-stimulation (Fig. 5a). After anti-CD3/CD28 stimulation a small fraction of HD CD4 and CD8 T cells underwent AICD, whereas patient cells did not ( Supplementary  Fig. 5). In all experiments, HD and patient T cells responded equally towards P/I that are bypassing proximal TCR and CD28 co-signalling.
To assess T-cell effector function, we measured cytokine secretion (IL-2, IL-4, IL-5, IL-10, IL-12, IL-13, IL-17A, GM-CSF, IFN-g, TNF-a) of stimulated PBMC. On stimulation with anti-CD3, we observed an increase of cytokine concentrations with the exception of IL-10 and IL-17A in patients' PBMC. However, when stimulated with anti-CD3/CD28, a marked increase in cytokine concentrations was only seen in HD PBMC. Stimulation with P/I induced comparable cytokine amounts in HD and patient cell culture supernatants (Fig. 5b).
NK and T-cell cytotoxicity play crucial roles in tumour surveillance and control of EBV 21,22 . We therefore determined degranulation of and NKG2D expression on NK and CD8 T cells. NK-cell degranulation in response to co-incubation with the erythroleukemic cell line K562 was impaired and accompanied by a decreased expression of NKG2D. In resting CD8 T cells, NKG2D expression was reduced as well. NK cell and CD8 T-cell pre-culturing with IL-2 abrogated the observed differences in degranulation and NKG2D expression (Fig. 6).
In summary, CARMIL2-deficient CD4 and CD8 T cells show impaired activation, proliferation and effector function in response to CD28 co-stimulation. Degranulation and NKG2D expression were impaired in steady state NK and CD8 T cells, but could be rescued with IL-2.
CD28 co-signalling. The TCR and CD28 co-receptor target an overlapping but not identical array of signalling intermediates leading to joint functional outcomes such as T-cell activation, proliferation and effector function 8 . The TCR recognizes peptide antigens presented by MHC molecules, thus conferring antigen specificity. CD28 interacts with the invariant co-stimulators CD80 and CD86, thus leading to full naive T-cell activation 23 . The precise positioning of CARMIL2 and its interaction partners in CD28 co-signalling are currently unknown 10 . To analyse CD28 co-signalling, we stimulated T lymphoblasts of HD1, HD2, P1.1, P2.1 and P2.2 with anti-CD3, anti-CD3/CD28 for 2, 5, 10, 15 and 30 min and performed immunoblotting and (phospho-) flow cytometry for essential signalling intermediates ( Fig. 7 and Supplementary Fig. 6). P/I stimulation for 5 min served as a positive control. Phosphorylation of the upstream tyrosine kinase ZAP70 as well as the serine/threonine kinase PKCy that participate in the immunological synapse was comparable in HD and patient cells (Fig. 7) 23 . Similarly, there was no difference in phosphorylation of the downstream mitogen-activated protein kinase ERK1/2 in CARMIL2-deficient and HD T lymphoblasts (Fig. 7). In contrast, in CARMIL2-deficient T lymphoblasts, the downstream canonical NF-kB pathway was not activated in response to CD28 co-signalling as illustrated by aberrant degradation of IkBa and phosphorylation of NF-kB p65 (Fig. 7). Signalling data obtained by (phospho-) flow cytometry showed no difference between CD4 and CD8 HD and patient T cells ( Fig. 7 and Supplementary Fig. 6).
We conclude that CARMIL2-deficiency selectively impaired the activation of the canonical NF-kB pathway in a CD28dependent manner.
Cytoskeleton dynamics and T-cell migration. The cytoskeleton plays an important role at almost all stages of the immune response. Not surprisingly, there is a growing list of PID associated with mostly actin-related cytoskeletal defects 24 . Through its CBR domain, CARMIL2 is supposed to play an important role in actin dynamics 25 . To determine the role of CARMIL2 in cytoskeleton dynamics and migration, we analysed total amounts of filamentous (F)-actin and actin distribution in the leading edge of HD and patient T lymphoblasts migrating on ICAM1 in 2D. Quantification of total F-actin was comparable between HD and patient T lymphoblasts in steady state, after PMA-induced polymerization and latrunculin B-induced depolymerization ( Supplementary Fig. 7a,b). Surface expression of total and active LFA1-that interacts with ICAM1-and subcellular LFA1 distribution were comparable in HD and patient T lymphoblasts as well ( Supplementary Fig. 7c,d). However, CARMIL2-deficient T lymphoblasts migrating on ICAM1 in 2D, showed a dispersed F-actin distribution at the leading edge as compared with a homogeneous distribution in CARMIL2-proficient cells ( Fig. 8a and Supplementary Fig. 7e). In addition to the effect on F-actin distribution, localization analysis of a-tubulin showed a markedly disrupted microtubule network (Fig. 8b, Supplementary Fig. 7f and Supplementary Movies 1-4). Furthermore, we detected a global reduction in post-translationally modified stable acetyland stable detyrosinated glutamyl-a-tubulin monomers in CARMIL2-deficient T lymphoblasts ( Fig. 8c and Supplementary  Fig. 7g). Microtubule dynamics are crucial for migratory cell polarity and microtubule depolymerization is associated with increased spontaneous motility in immune cells 26 . Accordingly, patient T lymphoblasts showed dispersed polarity (Fig. 8d) and increased spontaneous migratory speed both in a 3D collagen gel and in 2D on ICAM1 (Fig. 8e,f). Pharmacological microtubule disruption with nocodazole (NDZ) increased migratory speed of HD cells in 2D but had only mild effects on CARMIL2-deficient cells suggesting that they already displayed an intrinsic microtubule disruption (Fig. 8g). Although migrating with increased speed that resulted in higher accumulated migratory distance ( Supplementary Fig. 7h), CARMIL2-deficient T-cell migration was less orientated with a markedly decreased directness (Fig. 8h) and reduced Euclidean distance ( Supplementary Fig. 7i,j). Reduced straightness of patient T-cell migration was further shown by impaired CXCR4-mediated and CXCL12-guided chemotaxis (Fig. 8i). The reduced chemotactic response could not be attributed to aberrant expression of cell surface CXCR4 ( Supplementary Fig. 7k). Again, NDZ treatment reduced CXCL12-guided chemotaxis in HD but not patient T lymphoblasts (Fig. 8j).
In summary, CARMIL2-deficient T cells showed impaired F-actin distribution at the leading edge and disturbed microtubule network that were associated with increased LFA1-mediated   migratory speed in 2D and LFA1-independent migratory speed in a 3D collagen network. Accumulated migratory distance was increased, but Euclidean distance and migration directness as well as chemokine-guided migration were decreased.

Discussion
We here describe a novel human PID caused by autosomal recessive mutations in CARMIL2 that abrogated protein expression. In the context of preserved canonical TCR signalling and CD28 surface expression, CARMIL2-deficient T cells showed impaired CD28-mediated co-signalling. Whereas upstream phosphorylation of ZAP70 and PKCy and downstream phosphorylation of ERK1/2 were inducible in CARMIL2-deficient T cells, the activation of the canonical NF-kB pathway was impaired. Human CARMIL2-deficiency is reminiscent of a murine model of Rltprdeficiency discovered in a N-ethyl-N-nitrosourea-mutagenesis screen 10 . Rltpr-deficient mice also display defective CD28mediated co-signalling, associated with defective microclustering of Pkcy at the immunological synapse 10 . We could document regular phosphorylation of PKCy, but were unable to assess microclustering. It is possible that CARMIL2 contributes to the stabilization of activated PKCy-microclusters at the immunological synapse and thereby couples CD28 co-signalling to the canonical NF-kB pathway 10,27 . However, precise mechanistic details on the role of CARMIL2 for CD28-mediated co-signalling and potentially other signalling pathways need to be addressed in future studies. On the cellular level, CARMIL2-deficiency did not interfere with thymic development of conventional CD4 and CD8 T cells. This is in line with preserved TCR signalling 2 . In contrast, CARMIL2-deficiency is characterized by impaired naive T-cell activation, proliferation, effector function and insufficient gain of T-cell memory, indicative of a globally compromised peripheral T-cell immunity.
Consistent with an established role of murine CD28 co-signalling for the induction of Foxp3 and the development of natural T reg , human CARMIL2-deficient T cells, that were able to express the TCR-inducible activation marker CD69, did not express CD25 and FOXP3 in response to CD28 co-signalling 28 .    ARTICLE Similar to murine Rltpr-deficiency, human CARMIL2-deficient patients were devoid of T reg in the peripheral blood 10 . At first sight, it may appear surprising, that a lack of T reg in CARMIL2deficiency was not associated with immune dysregulation. However, as discussed before and as shown by our data, CD28 co-signalling is necessary to fully activate naive T cells and to induce proliferation and effector T-cell (T eff ) responses 23 .
Therefore, the complete absence of T reg could be counterbalanced functionally by impaired T eff responses. CARMIL2 has been shown to provide a functional link between vimentin intermediate filaments and membrane-associated actin networks during lamellipodia formation, cell migration and invadopodia-mediated matrix degradation in fibrosarcoma cells 29 . Our studies in CARMIL2-deficient T lymphoblasts       showed impaired actin distribution in the leading edge and a severely disturbed microtubule network associated with increased spontaneous migratory speed but decreased directness and chemokine-directed migration. It is possible that high migratory speed of CARMIL2-deficient T cells results in suboptimal scanning of peptide:MHC-complexes on professional antigenpresenting cells and thus affects antigen-specific T-cell activation 30 . In line with that, we found impaired proliferation and cytokine production in co-cultures of antigen-presenting autologous EBV þ LCL and CARMIL2-deficient TCL. However, keeping in mind the impaired canonical NF-kB pathway discussed before, it is difficult to weight the particular contributions of disturbed antigen-scanning and CD28 co-signalling on T-cell activation. Despite increased migratory speed of CARMIL2-deficient T cells, impaired directness and chemokine-mediated migration might additionally impair T eff responses in CARMIL2-deficient patients. Defective microtubule stability has been associated with increased spontaneous migration in 3D networks and decreased chemokine-guided migration of T cells 31,32 . Thus, CARMIL2 now emerges as a critical regulator of cytoskeletal dynamics in T cells. At this point, we cannot demarcate specific effects of defective T-cell activation and defective T-cell migration accounting for the state of immunodeficiency.
The most prominent clinical finding in all four CARMIL2deficient patients was EBV þ SMT. EBV þ SMT is a very rare tumour entity usually associated with secondary immunodeficiency disorder, for example, in the context of HIV-infection or iatrogenic immunosuppression in organ transplant patients 15 .
In human OX40-deficiency, HHV8 þ Kaposi's sarcoma cells strongly expressed OX40L and impaired OX40 signalling on immune cells was supposed to be a key pathogenic factor 3 . However, we did not detect increased expression of CD80 or CD86 on EBV þ SMT.
In human SAP, CD27 and CD70 co-signalling deficiencies, orthotopic EBV-infection causes hemophagocytic lymphohistiocytosis (HLH) and/or highly proliferating and immunogenic B-cell lymphomas, respectively 4 . In striking contrast, CARMIL2deficiency is not associated with EBV-induced B-cell proliferation, but rather with a slowly proliferating and pauci-immune SMT associated with dystopic EBV-infection 33 . Since CD28 cosignalling is essential to activate naive T cells and to induce T eff responses, EBV þ SMT in the context of CARMIL2-deficiency might evade anti-tumour immunity and indicate a particular importance of CD28 co-signalling in controlling this tumour entity 23 . On the other side, even in the absence of EBV þ lymphoproliferative disorder, we found that CARMIL2-deficient T cells also had decreased capacity to recognize EBV þ autologous LCL. Further studies may unravel why EBV causes leiomyomatosis rather than B-cell proliferation in CARMIL2deficient patients.
Of note, exposure to recombinant IL-2 partially restored markers of NK and cytotoxic T-cell function that are crucial to prevent EBV-triggered lymphoproliferative disorders 34 . This observation might not only be relevant for potential immunomodulatory effects in vivo, it further might indicate that besides CD28 co-signalling additional receptor systems could depend on CARMIL2. From a technical point of view, IL-2 responsiveness also severely hampered our attempts to reconstitute the T-cell phenotype in vitro by lentiviral gene transfer.
In summary, CARMIL2-deficiency is a novel primary autosomal recessive human immunodeficiency disorder highlighting a role for CARMIL2 as orchestrator of CD28 co-signalling and cytoskeletal dynamics necessary for T reg development, adequate T-cell activation, proliferation, differentiation, effector function as well as T-cell polarity and migration.

Methods
Patients. Patients were referred to the Dr von Hauner Children's Hospital, Ludwig-Maximilians-Universität, Munich, Germany and the Department of Pediatric Gastroenterology, Federal University of Minas Gerais, Belo Horizonte, Brazil. Informed consent was obtained according to current ethical and legal guidelines. The study protocol was approved by the Institutional Review Board at the University Hospital, Ludwig-Maximilians-Universität, Munich, Germany (Project Number 381-11). The study was conducted in accordance with the Declaration of Helsinki.
LCL-responsive T-cell lines. For initiating T-cell lines (passage 1), 1 Â 10 6 peripheral blood mononuclear cells (PBMC) from patients P2.1 and P2.2 and from two EBV-seropositive healthy donors (HD1 and HD2) were co-cultured with gamma-irradiated (80 Gy) autologous EBV þ lymphoblastoid cell lines (LCL) at a 10:1 ratio (PBMC:LCL) in a total volume of 2 ml T-cell medium consisting of AIM-V lymphocyte medium supplemented with 10% pooled human serum, 2 mM L-glutamine, 10 mM HEPES and 1% amphotericin B. To prevent expansion of FCSreactive T cells, stimulator LCL were continuously cultured in RPMI 1640 medium supplemented with 10% human serum, 2 mM L-glutamine, 1% nonessential amino acids, 1 mM sodium pyruvate, 2 mM L-glutamine and 50 mg ml À 1 gentamicin as described 35 . After 48 h, 50 U ml À 1 IL-2 (Novartis) was added to the culture medium. The T-cell lines were restimulated every two weeks in the same way and split when needed. After ten passages, the lines contained predominately (495%) CD3 þ CD8 þ cells and few (o5%) CD3 þ CD4 þ cells. Specificity of T cells was tested by incubating 5 Â 10 4 target cells (autologous or allogeneic HLA-mismatched LCL) with 5 Â 10 4 T cells in a final volume of 200 ml T-cell medium. After 24 h of co-culture, ELISA (R&D Systems) for measuring supernatant cytokines was performed.
Genetic analysis. Genomic DNA of family 1 subjects F1, M1, P1.1, P1.2 and S1.1 (Fig. 2a) was genotyped using the Genome-wide Human SNP array 6.0 on a GEO Platform GPL6801 (Affymetrix) and used to prepare exome libraries using the SureSelect XT Human All Exon V4 þ UTRs kit (Agilent Technologies). Barcoded libraries were sequenced on a SOLiD 5500 XL next-generation sequencing platform (Life Technologies) at the Dr von Hauner Children's Hospital sequencing facility to an average coverage depth of 80 reads. Bioinformatic data processing followed an in house pipeline: Reads were aligned to the human reference genome (version hg19) using LifeScope Analysis Suite 2.5.1 (Life Technologies, Carlsbad, CA, USA) using standard parameters. The lane data for each sample were merged and the PCR duplicates were removed using SAMtools version 1. 18 (ref. 36). SAMtools was also used to call SNPs with the command: 'mpileup -M 300 -I -g -D -C50 -S'. Small insertions and deletions (indels) were identified using DinDel version 1.01 (ref. 37), by first identifying putative indels for each sample Individually, and then re-running the DinDel on the pooled set of samples from family A with the previously identified indels as suggested variants. Variants were annotated and their manifestation at the mRNA and protein level identified according to the Ensembl Database (version 68). In addition, information on variant frequencies was collected from HapMap (HM) 38 , 1000 genomes (TG) 39 , NHLBI Grand Opportunity Exome Sequencing Project (NH) and ExAC 40 .The WES approach for genomic DNA of family 2 subjects F2, M2, P2.1 and P2.2 has been published before 20 . The c.489insG and c.871 þ 1G4T CARMIL2 mutations were confirmed by Sanger sequencing (see Supplemantary Table 5 for primer information). CARMIL2 cDNA was reverse transcribed from P2.1 T lymphoblast total mRNA (RevertAid H Minus Reverse trascriptase; ThermoFisher Scientific) and skipping of exon 11 was confirmed by Sanger sequencing (see Supplemantary Table 5 for primer information).
NK and CD8 T-cell degranulation and NKG2D expression. Degranulation of resting and activated NK cells was measured by surface staining of CD107a without (medium-cultured cells) and 3 h after stimulation with K562 cells at a ratio of 1:1 (ref. 41). The erythroleukemic cell line K562 (ATCC, CCL-243) was used as target cell line. NK cells were cultured in medium containing 600 U ml À 1 IL-2 (Novartis) for 48 h to assess degranulation of activated NK cells. CTL (cytotoxic T lymphoblasts) degranulation was evaluated in T lymphoblasts 48 h after stimulation with 1.25 mg ml À 1 phytohemagglutinin-L (PHA-L, Sigma) and 200 U ml À 1 IL-2 (Novartis). CTL degranulation was calculated by the difference in median fluorescence intensity of CD107a of CTLs stimulated with CD3/CD28 coated microbeads (ThermoFisher Scientific) at a ratio of 1:10 for 3 h and mediumcultured cells.
Quantification of F-actin. To quantify the amount of F-actin, T lymphoblasts were left untreated or were incubated with 100 nM phorbol-myristate-acetate (PMA, Sigma-Aldrich) for 5 min or with 0.5 mM latrunculin B (Calbiochem) for 10 min at 37°C followed by fixation with 4% paraformaldehyde. Fixed cells were permeabilized with 0.01% Triton X-100 (Sigma-Aldrich), stained with phalloidin-Alexa594 (1:500, Invitrogen) and F-actin amount was quantified by flow cytometry analysis using a BD FACSCalibur and FlowJo software (Tree Star).
Migration assays. Two-dimensional migration of T lymphoblasts on ICAM1 were analysed on m-slides (Ibidi) coated with 0.15 mg recombinant human ICAM1/ Fc (R&D Systems) per channel over night at 4°C. Slides were blocked with 1% BSA/PBS for 1 h before use. A total of 5 Â 10 4 T lymphoblasts per channel were allowed to migrate in a steam-saturated incubation chamber at 37°C, 7% CO 2 , and images were taken every 8 s for 1 h with a Plan-Apochromat 40 Â /1.3 NA oil objective using an Axiovert 200 microscope (all Zeiss Microscopy). The migratory speed, accumulated and Euclidean distance and directness were quantified using ImageJ software (National Institute of Health) and the chemotaxis and migration tool V2.0 from Ibidi. Cells migrating o80 mm in 30 min were not included as they were mostly stationary. To analyse the three-dimensional migration of T lymphoblasts, a collagen matrix was prepared using 1 mg ml À 1 collagen type I from rat tail according to the manufacturer's instructions (Ibidi). Activated T lymphoblasts were embedded into the collagen matrix at a concentration of 1 Â 10 5 cells within a m-slide. Analysis and quantification of cell migration was performed as described above by taking pictures every 8 s for 30 min using an EC Plan-Neofluar 20 Â /0.50 NA. Migration assays in response to CXCL12 were performed using 6.5 mm transwells with 5 mm pore size (Corning). 5 Â 10 5 T lymphoblasts were added on the top of each filter. Medium containing 2.5% serum and 25 mM Hepes was placed at the bottom and supplemented with 100 ng ml À 1 CXCL12 (R&D Systems) when indicated. Assays were run for 2 h at 37°C and 7% CO 2 . Migrated lymphocytes were collected and quantified using CountBright Absolute Counting Beads (Molecular Probes) and a BD FACSCalibur.
Statistical analysis. Data were analysed with the software environment R (version 3.2.3) or Prism 5 (GraphPad) for statistical computing and graphics. Normal Gaussian distribution of data sets was tested using Shapiro-Wilk normality test. Significances between two groups were calculated with the two-sided unequal variances Welch's t-test or two-sided unpaired Student's t-test. Two-way ANOVA was used for comparisons of more than two groups. P values of o0.05 were considered significant.
Data availability. CARMIL2 variant data that support the findings of this study have been deposited in ClinVar with the primary accession codes SCV000322745 and SCV000322755. Other data that support the findings of this study are available from the corresponding author upon request.