Adaptor proteins: Flexible and dynamic modulators of immune cell signalling

To maintain homeostasis, all cells respond to environmental cues via a multitude of surface receptors. In order to act appropriately in their environment, cells are dependent on the transduction of the incoming signal through tightly regulated and interconnected signalling pathways to the cell nucleus. In particular, cells implicated in the immune system greatly depend on such systems to respond in a flexible and dynamic manner to environmental challenges. One major group of intracellular proteins that are involved in these signalling pathways are adaptor proteins. Although adaptor proteins are essential for normal immune cell operation, the functional role of this group of signalling proteins remains to be fully appreciated. So far, research on adaptor proteins has revealed their unique potential in building transient complexes in a reversible, dynamic and inducible manner. In this review, we explore the roles of adaptor proteins – in space and time of intracellular signalling – and their associations with human disease. Examples of adaptor proteins expressed in hematopoietic cells highlight their crucial role in the immune system. Lastly, we present challenges faced in elucidating roles of adaptor proteins, as illustrated by the T cell–specific adaptor (TSAd) protein encoded by the SH2D2A gene.


| INTRODUCTION
All cells depend on tightly regulated signalling pathways to appropriately respond to their environment. The basic model assumes that extracellular information is received by cell surface and/or intracellular receptors. These signals are subsequently relayed through the cytoplasm. Ultimately, the information may reach the cell nucleus, where it is translated into gene activation/ inactivation. 1 Consequently, changes in gene expression should result in the adequate cellular response to external stimuli.
Immune cells are particularly dependent on tightly controlled information transfer from the cell exterior. 2 These cells actively seek and respond to a myriad of challenges in an unpredictable environment. In addition, they must constantly be ready to adapt to ever-changing conditions. Immune cells must respond appropriately, as the whole organism's safety depends on their activity. Activated receptors trigger a complex network of signalling events leading to the modulation of protein activity or gene transcription. This occurs through transient alteration in the signalling proteins involved, by conformational change or post-translational modification, typically resulting in the formation of multi-protein complexes. A group of proteins, which are integral parts of these signalling pathways, may have been outshone by their enzymatically active binding partners. These proteins do not perform enzymatic functions. Since their only activity is binding, they have been designated adaptor proteins. In order to perform their functions, these molecules may contain conserved interaction domains and/or unique binding motifs. This allows adaptor molecules to bring together proteins and adapt signalling pathways.
In this review, we discuss how adaptor proteins can form transient molecular complexes in a reversible, dynamic and inducible manner using their structured domains and disordered regions. Further, we explore the intriguing functions and roles of adaptor proteins, with emphasis on intracellular signalling in immune cells. To point out their relevance at the level of the organism, we examine their role in cancer as well as in genetic and infectious diseases. Finally, we present challenges faced in elucidating roles of adaptor proteins, as illustrated by the T cell-specific adaptor (TSAd) protein encoded by the SH2D2A gene.

| Definition of adaptor proteins
Defining adaptor proteins is challenging and multiple efforts have been made to distinguish it from molecules with anchoring, scaffolding or docking functions. 3,4 These terms are used interchangeably, and there is no consensus in the field as to what should be considered an adaptor as distinguished from a scaffold, an anchor or a docking protein. Therefore, for the sake of simplification and also since their functions are overlapping, in the context of this review, we have chosen to refer to all proteins with adaptor, scaffolding, anchoring or docking functions as adaptor proteins.
The simplest definition of an adaptor protein is an intracellular protein, which facilitates signal transduction through interactions with other proteins. However, this definition includes practically all signalling proteins. Therefore, adaptor proteins can be defined through exclusion. To narrow down the scope of the review, we have made a list of exclusion criteria to filter out non-adaptor proteins.
Firstly, proteins with any extracellular interactions, including ligand-transmembrane receptor interactions, are excluded.
Secondly, as adaptor proteins do not possess enzymatic activity, enzymes involved in cellular signalling are excluded. For instance, a kinase containing two separate binding domains in addition to the enzymatic domain, is not an adaptor, while a pseudokinase, which has lost its enzymatic activity, can be considered an adaptor. 5 Thirdly, DNA-or RNA-binding proteins, that is transcription and translation factors, are excluded. By all the other means, the latter could be considered adaptor molecules due to their multivalency and lack of enzymatic activity.
Fourthly, adaptor proteins, which engage in the formation of stable complexes, are excluded. To appropriately perform their roles, adaptors generally form low affinity interactions of transient or repetitive nature. Proteins forming stable complexes usually only have one well-defined function, in contrast to most adaptors, which are very flexible in both interactions and functions.
The importance of moderate binding affinities for adaptor function is illustrated by artificial SH2 domains binding with high affinity to pTyr, that is superbinder SH2 domains, which may perturb phosphotyrosine signalling pathways. 6 Similarly, Legionella can release superbinder SH2 domains, of even higher affinity but low specificity, 7 into macrophages. Such a phenomenon may be a virulence factor for Legionella, resulting in perturbed signalling in the infected cell.
The requirement of low affinity interactions excludes molecules such as the CD3 chains. These proteins are necessary for signal transduction from the T cell receptor (TCR) 8 and lack extracellular receptor function as well as enzymatic domains. Although CD3 chains as standalone molecules could be referred to as adaptor proteins, they only function as integral parts of receptors, which lack similar domains on their own. Lastly, a group of proteins known as adaptins is excluded. These proteins have a unique role in clathrin-mediated vesicular transport by forming stable and well-defined complexes.
Taken together, an adaptor molecule is an intracellular protein, which lacks enzymatic activity as well as DNA or RNA binding properties and can form dynamic multi-protein complexes by binding to two or more proteins at the same time. Adaptor molecules additionally share a number of specific features, which equip these proteins with abilities necessary for their function. While this review focuses on adaptor proteins as defined above, it is important to be aware that many proteins, under given circumstances, may fulfil adaptor molecule functions in addition to enzymatic or other functions.

| Features of adaptor proteins
Adaptor proteins assist in positioning other molecules to maintain proper signalling within a pathway. In order to do so, adaptors need special features-that is binding domains, binding motifs and structural flexibility. Binding domains are usually conserved regions with defined structure and specificity. In contrast, binding motifs can be flexible short linear motifs defined by the protein sequence domains are an essential part of protein tyrosine phosphorylation pathways in most immune cells responses. The protein tyrosine phosphorylation cascade involves tyrosine kinases and phosphatases, which add or remove a phosphate group of hydroxyl-containing amino acid, respectively. The phosphotyrosine motifs generated by kinases can be recognition sites for the evolutionary conserved SH2 domains. The SH2 domain endows a protein with the ability to 'read' the activity of tyrosine kinases by docking onto particular phosphotyrosine motifs on other proteins. 9 The human genome encodes 111 proteins harbouring a total of 121 SH2 domains. 10 Of these, 28 are non-receptor protein tyrosine kinases. A large proportion of these is expressed in immune-related cells and tissues as compared to other tissues ( Figure 1A). As discussed later in this review, 49 of the SH2 domain-containing proteins are considered adaptor proteins. Several of these SH2 domain-containing adaptor molecules display preferential expression in tissues enriched in immune cells ( Figure 1B). The preferential expression of both tyrosine kinases and SH2 domain-containing adaptor proteins in tissues enriched in immune cells highlights the importance of phosphotyrosine mediated signalling in cells of the immune system. However, not all adaptor proteins involved in intracellular signalling harbour SH2 domains. A keyword search in the Pfam database (https://pfam.xfam.org/) retrieves 443 entries for 'interaction domain' and 1886 entries for 'binding domain'. Many of them can be found in adaptor molecules. An overview of representative binding domains, which may also be present in adaptor proteins, is given in Table 1.
The wide array of binding domains allows adaptor proteins to play roles in a plethora of intracellular pathways, depending on which domains they harbour. While SH2 domains participate in tyrosine phosphorylation cascades, 14-3-3 domains can be involved in serine-threonine phosphorylation cascades and Toll/interleukin-1 receptor (TIR) domains are implicated in Toll-like receptor (TLR) signalling. Other examples of interaction domains include SH3 domains binding proline-rich motifs, pTyr binding (PTB) domains recognizing defined phosphopeptide motifs and pleckstrin homology (PH) domains binding specific polyphosphoinositides 3 (Table 1).
In addition to these structured domains, adaptor proteins can also carry out their function through the unstructured or disordered regions that they possess. Binding motifs and structural flexibility are both consequences of intrinsically disordered regions, which are very often a significant part of an adaptor molecule's sequence (48%-63% on average, depending on the prediction algorithm 11 ).
Intrinsically disordered regions are a part of a protein that lacks a stable 3-D structure as defined by currently available scientific methods. 12 Two decades ago, it was found that disordered regions of most proteins evolve faster than the structured regions. 13 However, a more recent study has observed that within the disordered regions there are linear motifs, recognized by other domains (SH2, SH3 or Ser/Thr kinases), that are surprisingly well conserved. 14 Additionally, the number of proteins containing unstructured regions of more than 50 amino acids in length is much higher in eukaryotes than in bacteria or archaea. 15,16 This suggests that there is an evolutionary advantage in developing proteins with higher flexibility. The flexibility provided by intrinsically disordered regions allows complex organisms to produce multifunctional proteins, such as adaptor proteins, whose role could differ within a signalling pathway or between various cells (so-called moonlighting 17 ). The obvious importance of intrinsically disordered regions makes it surprising that many studies disregard their significance for protein function. 18 There are a substantial number of features, which characterize disordered regions and define their unique functions in intracellular signalling. 19 Just to name a few, intrinsically disordered regions can drive membrane curvature 20 ; form phase-separated compartments 21,22 ; contain adaptable short linear motifs, which can overlap 14 or be post-translationally modified 23,24 ; serve as flexible linkers between domains 25 ; increase the hydrodynamic radius of the protein [26][27][28] ; dynamically change protein structure 24,[29][30][31] ; perform moonlighting 32 ; and promote oligomerization. [33][34][35] Most of these features of disordered regions are highly relevant for the function of adaptor proteins. 36 Short linear motifs (SLiMs) 37 (or molecular recognition features (MoRFs) 14 ) are possibly the most important feature from the examples mentioned above. While the specificity of protein domains and hence their binding partners can to a certain degree be mapped, 38 this has been much more difficult to achieve with SLiMs. It was therefore not before applying sophisticated search algorithms 39 that scientists were able to understand the commonality of SLiMs in the proteome. However, even the best software cannot predict binding to novel targets, as all programmes are trained on existing knowledge. 39 Short linear motifs present in disordered regions are more accessible to binding domains as they are continually exposed. 14 If a binding motif is hidden within a structured domain, the binding conformation will be much more restricted, as in the classic model of lock-and-key protein-protein interactions. Unstructured regions may bind to proteins with high specificity, but with low affinity, which makes them ideal signalling hubs-both specific and reversible. 40 High specificity comes from the ability of disordered regions to complement the binding interface, while low affinity comes from the transient nature of the interaction. However, this perspective was recently challenged, 41 as it was speculated that disordered structures, due to their flexibility, actually combines low specificity with a high affinity. Even if the binding affinity seems low, the disordered structure can adapt itself to the target surface by increasing its own recognition surface and therefore, increasing the interaction stability and its affinity. 42 Not all adaptor molecules consist of large portions of intrinsically disordered regions with SLiMs. There are certain families of adaptor proteins that consist nearly entirely of a single binding domain. For example, members of the signalling lymphocytic activation molecule (SLAM)-associating protein (SAP) family, which includes SAP (SH2D1A) and EAT2 (SH2D1B), contain only one SH2 domain, and the 14-3-3 proteins (YWHAB/E/G/H/Q/Z and SFN) contain only the 14-3-3 domain. However, even in these cases, flexibility may be a key property of the adaptor. Firstly, SAP family proteins contain intrinsically disordered tails, which represent 30% of the total protein sequence. Similarly, 14-3-3 domain contain a hidden disordered region propensity, which most likely provides the protein with means to interact with numerous interaction partners and perform moonlighting functions. 43 Secondly, a single isolated domain is a quintessence of flexibility as no internal structural constraints stop the protein from forming interactions. SAP itself is considered to be a detachable SH2 domain of FYN kinase as it binds to the FYN SH3 domain and extends the enzyme's pool of interaction partners. 44 Another unique property of the SAP SH2 domain is that it is not restricted to conventional pTyr-containing motifs. Although with a lower affinity, the non-canonical SH2 domain can bind unphosphorylated peptides providing a source of flexibility to its binding partners. 45 Post-translational modifications are an important part of any signalling pathway due to their (generally) inducible and reversible nature, and their predominant occurrence in disordered regions, and consequently in SLiMs. 23,46,47 A certain degree of sequence flexibility may be necessary for post-translational modifications to occur. 47 Post-translational moieties may change the structure of a protein sequence, including the binding properties and functionality of SLiMs. 37 The most commonly described modifications are additions of small chemical groups to specific amino acids, such as phosphate (to serines, threonines or tyrosines), methyl (to arginines or lysines) or acetyl (to lysines). Among these, tyrosine phosphorylation is of particular interest to immunologists.
One of the major adaptor proteins in T cells, linker for activation of T cells (LAT), contains multiple phosphotyrosine sites that are necessary for a molecule's ability to bind to different targets 48 and thus regulate T cell activation. Another important adaptor protein, MYD88, involved in TLR signalling, becomes inhibited when acetylated on Lys265. Consequently, autocrine production of growth-promoting IL-6 in B cell lymphoma is suppressed. 49 There are also other, more complex modifications, such as lipidations (palmitoylation and myristoylation), ubiquitination and SUMOylation. Palmitoylation of LAT is necessary for the translocation of the protein to the plasma membrane. LAT lacking palmitoylated cysteines does not perform its functions, that is absence of these cysteines disrupts T cell development and inhibits TCR activation. 50 Ubiquitination is usually, but not exclusively, connected to protein degradation, which can also affect cellular signalling. Ubiquitination of both LAT and SLP-76 (LCP2)-another adaptor protein involved in T cell activation-tags these proteins for degradation. In both cases, the lack of the ubiquitination tag increases TCR signalling. 51,52 SUMOylation seems to be more variable in function. It can promote interaction between the adaptor protein GRB2 and the SOS1 guanine nucleotide exchange factor, which leads to increased ERK activity, and subsequently increased cell motility. 53 On the other hand, SUMOylation of the adaptor TAK1-binding protein 2 (TAB2), which is involved in IL-1 signalling, may decrease the adaptor's activity, as shown by the AP-1 luciferase reporter assay. 54

CONTROL SIGNALLING CASCADES IN SPACE AND TIME
The unique combination of conserved protein-binding domains and intrinsically disordered structures containing SLiMs allows adaptor molecules to control intracellular signalling cascades in space and time. Adaptor proteins can control the localization of signalling molecules, which defines the pathway's space. In parallel, they can control the sequence of enzymatic events, which defines the pathway's timing. These processes are often highly interrelated.
In the following, we will discuss various aspects of the adaptor protein functions. For the sake of simplification, we have separated these into two main, but not mutually exclusive, categories: control of (i) space and (ii) time of cellular signalling ( Figure 2).

| Molecular hub
All adaptor proteins serve, to various extents, as scaffolds for the recruitment of other molecules involved in a given signalling pathway (Figure 2A). One of the best-described examples is LAT. 55 Upon triggering of the TCR, tyrosine phosphorylation of CD3 by LCK recruits and allows activation of another tyrosine kinase ZAP70. LAT is a major phosphorylation target of ZAP70. Membrane-bound LAT phosphorylated on multiple tyrosines becomes a molecular hub, which can recruit other proteins and diverge the signalling pathway ( [48], reviewed in [8). Phosphorylation of LAT on Tyr191 and Tyr226 recruits Vav guanine nucleotide exchange factor 1 (VAV1), which starts GTPase signalling, and subsequently, activation of mitogen-activated protein (MAP) kinase pathway and cytoskeleton reorganization. LAT pTyr171 associates with phosphoinositide 3-kinase (PI3K), which mediates the generation of phosphatidylinositol-3,4,5-triphosphate, and subsequently, activation of protein kinase B (AKT)-induced cell proliferation and survival. Another protein recruited by LAT (to pTyr132) is phospholipase C gamma (PLCγ), which catalyses the generation of inositol 1,4,5-trisphosphate and diacylglycerol from F I G U R E 2 Schematic overview of hypothetical models presenting functions of adaptor proteins. Specific examples for each model are described in detail in the main text. A-D, Adaptor proteins help to define the space of signalling pathways. A, They can nucleate cellular signalling by serving as a binding platform. B, Adaptors can form intracellular microdomains through clustering, oligomerization and phase separation. C, They can serve as mediators between specific proteins and D, increase the probability of interactions between molecules within a signalling pathway. E-H, Adaptor proteins help to define the timing of signalling pathways. E, With the specificity of their binding domains and motifs, they can determine the order of events within a signalling cascade and F, multiply the strength of a signal in order to establish the threshold necessary for the cell's decision-making process. G, Through enzyme repurposing, adaptors can connect signalling pathways initiated by different receptors or make enzymes perform different functions in various cell types. H, Additionally, through direct interaction with proteins, they can stabilize the conformation of other molecules necessary for signal propagation phosphatidylinositol 4,5-bisphosphate. These molecules trigger the intracellular release of calcium ions and activate the Ras-family of small GTPases and protein kinase C (PKC). All these pathways initiate gene transcription necessary for T cell activation. 48 However, transmembrane adaptor proteins, 56 such as LAT, are not the only adaptors known for their vast number of interaction partners. The growth factor receptor-bound protein 2 (GRB2) family is also involved in phosphotyrosine signalling cascades. 57 GRB2, the most studied protein from this family, is ubiquitously expressed in the body ( Figure 1B), allowing GRB2 involvement in numerous growth factor receptor tyrosine kinase pathways: epidermal (EGFR), vascular endothelial (VEGFR), hepatocyte (HGFR), platelet-derived (PDGFR) and fibroblast (FGFR). GRB2 interacts with various proteins, including LAT in the TCR pathway. However, the most well-established interaction of GRB2 is with SOS1, a protein that directly activates the MAP kinase pathway, resulting in differential expression of genes involved in cell survival and proliferation. Not surprisingly, GRB2 knockout mice are embryonic lethal. Moreover, GRB2 is often implicated in cancer transformation (see Adaptor proteins in human disease-Predisposition to cancer). 58

| Signalling segregation
The ability of adaptor proteins to create signalling clusters ( Figure 2B) is again the best described in the case of LAT in T cell signalling. 59 Triggered TCR signalling induces spatiotemporal segregation of molecules into concentric patterns. This phenomenon is called the immunological synapse. It is well characterized but lacks a unified model. 60 The immunological synapse facilitates the direct secretion of cytokines and cytotoxic granules, but it is also orchestrating intracellular signalling. 61 For T cells, it is generally accepted that the TCR locates to the centre of the immunological synapse and is surrounded by co-receptors (CD4/8, CD28, CD2, CTLA-4, PD-1). Adhesion receptors (LFA-1) and larger molecules (CD45) locate to the periphery of the synapse. The nature of the molecular exclusion is still debated. While most of the proposed segregation models focus on the surface receptors, segregation of signalling molecules exists also intracellularly. This could be secondary to the aggregation of TCRs, which may serve as an ignition point for intracellular signalling. However, alternative possibilities exist. It was recently found that LAT pre-assembles signalling clusters, which are actively transported via actin towards the centre of an immunological synapse. 59,62 Moreover, such clusters resemble liquid-liquid phase-separated microdomains. 62,63 A phaseseparated microdomain is a membrane-less cellular compartment, which concentrates certain molecules (reviewed in Ref. [64]). GRB2, NCK1, SLP-76, N-WASP (WASL), GADS (GRAP2), SLP-65 (BLNK) and CIN85 (SH3KBP1) are among adaptor proteins known to be involved in such phenomena, not only in TCR signalling, 21 but also in B cell activation, 65 cell adhesion and receptor tyrosine kinase signalling. 64 Segregation of signalling molecules generally facilitates signalling by providing the basis for specific protein recruitment, signal multiplication and increased probability of interactions/enzymatic reactions. However, data describing microdomains have to be taken with a reasonable caution, as it is still an emerging field, prone to bias and technical artefacts. 66 Another important notion is that many adaptors facilitate membrane anchoring of protein clusters, which further supports signalling segregation (see Increasing probability of interaction below). For this purpose, they can contain transmembrane domains, 56 lipidation sites 67 or phosphoinositide-binding domains and motifs 68 (such as pleckstrin homology domains 3 ). Even though the presence of a transmembrane domain is sufficient for protein localization to the cell membrane, additional modifications may be required for specific locations. For example, LAT has to be palmitoylated on Cys26, which serves as a sorting signal, in order to be correctly translocated to the cell membrane. 50 Membrane anchoring can also be tightly regulated. TIR domain-containing adaptor protein (TIRAP) involved in TLR signal transduction contains a phosphoinositide-binding motif for recruitment to the cell membrane. The motif can be phosphorylated on Thr28, which not only disrupts the membrane binding but also promotes TIRAP ubiquitination and subsequent degradation. 69 This mechanism is proposed to prevent prolonged TLR2 and TLR4 signalling. 70 Thus, adaptor proteins can segregate molecules within signalling cascades, through relocation to or creation of cellular compartments. This will facilitate interactions of specific proteins with their activated receptors.

| Specific protein recruitment
Adaptor proteins may be critical for the recruitment of specific enzymes ( Figure 2C). They contain domains related to the recruitment of proteins involved in particular signalling pathways. 3 Such interactions are often localization dependent. For example, Src kinase-associated phosphoprotein 1 (SKAP1) is indispensable to recruit and form a complex between Ras-related Protein Rap-1A (RAP1A) and Ras Association Domain Family Member 5 (RASSF5) in the plasma membrane. 71 The RAP1A-RASSF5 complex is necessary to transduce the signal from the LFA-1 receptor in T cells. Similarly, the adaptor TIRAP mentioned above recruits another adaptor protein MYD88 to the plasma membrane. This initiates the signalling of TLRs through IRAK4 kinase. 72 As such, it should not come as a surprise that many enzymes are reported to bind to multiple adaptor proteins. For example, LCK may bind to LAT, 73 SH2B3, 74 HSH2D, 75 TSAd, 76 LIME1 77,78 and SHC1 79 in T cells. This phenomenon allows the repurposing of a single kinase to various signalling cascades by modifying its location or by increasing its pool of substrates.
However, the flexibility of adaptor proteins is not only necessary for repurposing enzymes in a single cell, but also in various tissues. As previously mentioned, moonlighting refers to proteins with multiple functions (in different signalling pathways). Again, SLiMs and disordered regions are features that play a significant role in moonlighting. 32 There are several examples of adaptor molecules that have important functions in distinct tissues. For example, SH2B3 is necessary for negative regulation of myeloid cell proliferation, 80 T cell signalling 74 and endothelial progenitor cell-driven neovascularization. 81 SLP-76 is involved in signalling cascades of both platelet aggregation and TCR. 82 Due to its ubiquitous expression, it is not a surprise that 14-3-3 proteins are involved in a plethora of cellular functions, including a chaperone function. 83 While adaptors can retain a specific protein at a specific location, they may also deplete the pool of certain proteins (so-called prozone effect 84 ) to, for instance, downregulate a signalling cascade. Low concentration of adaptors may promote signalling by the formation of molecular complexes or priming of enzymes. However, at higher concentration, adaptors may form incomplete, non-functional complexes or deplete the pool of enzymes or substrates, and as a consequence inhibit signalling. Such a competitive mechanism is commonly observed for members of the SOCS, SLAP, GRB7 and SH2B adaptor protein families, which can target components of tyrosine kinase pathways. 85 One example is the Srclike adaptor (SLA), which can compete with SRC kinase for the interaction with PDGFR. 86

| Increasing probability of interaction
The hydrodynamic radius is the effective radius of the protein, which is the protein itself and all of the solvent molecules attracted by it. Proteins with large hydrodynamic radii make the largest contribution to membrane crowding. 87 Both theoretical considerations 27 and experimental data 26 show that intrinsically disordered proteins have larger hydrodynamic radii than globular proteins. As many adaptor proteins contain a significant portion of intrinsically disordered regions, they are more predisposed to crowd the cellular space. Consequently, their capture radius is larger and a smaller amount of adaptor proteins is necessary to reach probabilities of interactions similar to structured proteins 27 (so-called fly-casting mechanism 28 ).
Signalling modules are typically triggered in the plasma membrane, which reduces the interaction space into two dimensions. 88 Membrane anchoring and clustering at the cell membrane can induce favourable conformation and orientation of interacting proteins. For example, the presence of SLP-76 in membrane-bound microclusters is required for Tyr775 and Tyr783 phosphorylation of PLCγ by IL-2inducible T cell kinase (ITK), which is an essential event for T cell activation 89 ( Figure 2D). SLP-76 is not only mediating the interaction necessary for enzymatic activity, but it is also stabilizing the catalytically active form of ITK (see Stabilizing the conformation of signalling nodes below).
Adaptor proteins often interact with other adaptors 90 and promote oligomerization, [33][34][35]63 which further increases their interaction radius. Such molecular clusters can be pre-existing, enhancing the probability of the interaction with activated receptors/enzymes. 91 Moreover, due to intrinsically disordered regions, adaptor proteins can form fuzzy complexes. 30 This occurs when the adaptors dynamically change their conformation within the complex and actively adjust to new/modified binding partners. However, one of the most important phenomena that enhances the probability of interactions is spatial confinement, which allows repetitive re-binding. 92

| Transducing the signal
A major function of adaptor proteins is to participate in cellular signalling ( Figure 2E). Most signalling cascades lead to activation and translocation of transcription factors to the nucleus. The resulting altered gene expression may display a graded (analogue) response, exemplified in T cells by expression of CD69 or CD25, which correlates with the amount of specific transcription factor translocated to the nucleus. 93 However, other genes display a binary (digital) response to transcription factors, for example, IL-2 and IFN-γ that are either expressed or not. 94 Binary responses are a major part of the cellular decision-making process (see Signal multiplication below). When contributing to signalling cascades, adaptor proteins may not only provide graded signalling but also enable digital responses (with discrete, binary outcomes), sustained, transient or even oscillatory signalling. 95 The most important role of adaptor molecules in cellular signalling is to ensure a correct order of events (specificity of signalling). 82,[96][97][98] They achieve this through their binding domains of relatively high specificity 3 and SLiMs. The flexibility 37 and the presence of overlapping SLiMs 14 greatly enhance the functionality of the protein within the signalling pathway by providing it with fuzzy dynamics of interactions, or by allowing it to participate in distinct signalling pathways. 37 SLiMs can also define the signalling speed. The phosphorylation rate of Tyr132 in LAT is governed by the preceding Gly131. Substituting Gly131 for Asp or Glu can accelerate TCR signalling, which suggests that Gly131 provides a necessary proofreading step. 99 In addition, the number of ways, in which signalling pathways can be steered through post-translational molecular switches present in SLiMs, is very large. The best-known mechanisms include binary switches, specificity switches, cumulative switches, avidity switches and sequential switches (all reviewed in Ref. [100]). All of these mechanisms can establish and control the order of events within the signalling pathway.
Binary switches can be as simple as the phosphorylation-mediated binding of multiple proteins to LAT, 48 or phosphorylation-mediated inhibition of the ADAP SH3 domain binding to the phosphorylated interaction motif on SKAP1. 101 Both examples are involved in proper regulation of T cell activation. However, binary switches can be also more complicated and can involve allosteric control. When phosphorylated on Tyr221, one isoform of CRK, an adaptor that links tyrosine kinases to small G proteins, may be blocked from interactions with other proteins, 102 as this phosphosite is a ligand for CRK SH2 domain itself.
Specificity switches can affect the range of binding partners conferred by a given motif. Adaptor protein DISC1, which is involved in neurogenesis, switches between two different binding partners depending on the phosphorylation status of its Ser710. 103 If Ser710 is phosphorylated, DISC1 recruits BBS1 to the centrosome and stimulates cell migration. If Ser710 is not phosphorylated, DISC1 binds GSKβ, which activates the Wnt signalling pathway and subsequently, stimulates proliferation. Similarly, the location of some adaptor proteins defines their pool of interaction partners. The location within the cell can be controlled by post-translational molecular switches in the form of lipidation, for example, to correctly localize the adaptor to the plasma membrane. 50 Cumulative switches work as positive or negative rheostats. For example, ADAP contains at least three phosphotyrosines recognized by SLP-76 SH2 domains, which are critical for proper microcluster formation. 104 However, any two of these phosphosites can partially reconstitute protein oligomerization. This suggests that ADAP phosphorylation behaves like a positive rheostat of signalling microclusters formation in the T cells. The avidity switch can be exemplified with the interaction of adaptor protein TKS4 (SH3PXD2B) with SRC kinase. SRC binds with both its SH2 and SH3 domains to specific sites on TKS4. 105 The double interaction gives synergistic enhancement of binding strength, which can lead to prolonged signalling from the EGF receptor. Sequential switches depend on the subsequential order of post-translational modifications. One example of a sequential switch was described above, where ubiquitination and degradation of TIRAP depend on the preceding phosphorylation of Thr28. 69 The sequence of events is a crucial point of all signalling cascades and any mistake could bring disastrous effects, even to the whole organism. Adaptor proteins may prevent unspecific or inadequate molecular events from happening through functional misfolding. 106 The disordered regions enable a protein to escape unspecific (non-native) interactions by dynamically sequestering recognition motifs through structural changes of the surrounding sequence.

| Signal multiplication
While the sequence of events in the signalling cascade is of high importance, the strength (multiplication) of the signal is equally relevant for any signalling pathway ( Figure 2F). Computational modelling indicates that adaptor molecules can regulate signal duration. 107 As described above, adaptor proteins are excellent molecular hubs 55,57,108,109 that may strengthen and stabilize the signal by recruiting multiple proteins. However, a more potent way of increasing the signal is through oligomerization, 21,[33][34][35]62 which may lead to the creation of phase-separated microdomains. 21,62 It is a simple, but extremely efficient method of multiplying the function of a molecular hub.
Adaptor proteins can also promote a phenomenon called processive phosphorylation, 110 which facilitates phosphorylation of multiple sites by a kinase in one single encounter with a substrate. For instance, GRB2 may promote the processive phosphorylation of LAT by ZAP70. 111 Another example is LCK, which although not classified as an adaptor protein, may function as an adaptor by bridging ZAP70 to LAT, thus promoting processive phosphorylation of LAT by ZAP70. 73 Signal multiplication via adaptor proteins is also an important part of the cellular decision-making process. Computational modelling shows that adaptor proteins involved in phosphorylation pathways improve their determinism (bistability). 112 Moreover, proper organization of signalling nanoclusters allows digitizing the external stimuli to increase the fidelity of a system. 113 Again, this is exemplified by LAT, which has been pinpointed as one of the elements of digitization in TCR signalling. 114 The phenomenon might depend on a precise phosphorylation rate of specific tyrosines present in the LAT molecule. 99 Finally, computational simulations have shown that adaptor molecules not only may amplify a stimulus but may also attenuate it, depending on the context 115 and adaptor concentration. 84

| Joining signalling pathways
The necessity of activation through multiple distinct receptors is best exemplified by TCR co-signalling 116 (Figure 2G).
The need for fine-tuning of the T cell response via co-receptors is clearly illustrated through the development of the CAR T cell technology. The first generation of CAR T cells poorly imitated the behaviour of naturally activated T cells and was later improved by inclusion in the CAR construct of two T cell co-receptor intracellular domains. 117 This observation supports the notion that coordination of proper immune cell activation, where several signalling pathways need to be activated simultaneously, requires a large number of adaptor molecules. 116 This applies to the adaptors FYB1 and SKAP1, which translate the TCR pathway into an inside-out signal, through integrin-mediated adhesion. 118 Another example is the adaptor GRB2 which is necessary for T cell co-receptor CD28 to elicit IL-2 production, 119 an important part of proper activation of T cells. However, it seems that merging of the TCR and CD28 signalling pathways occurs at the level of transcription initiation, as both pathways depend on different adaptor proteins to activate NF-κB transcription factor, LAT/ ADAP and GRB2/VAV1, respectively. 120 Activation of NK cells also depends on the integration of signals coming from a variety of co-stimulatory receptors. A combination of certain receptors can provide synergetic signalling. For example, NKG2D (KLRK1) and 2B4 (CD244) signalling pathways converge at the phosphorylation of SLP-76. 121 The former predominantly induces the phosphorylation of SLP-76 on Tyr128, while the latter induces phosphorylation of Tyr113. Phosphorylation of both sites is necessary to overcome VAV1 inhibition by CBL-C E3 ubiquitin-protein ligase (CBL-C) and subsequent activation of NK cells. This phenomenon is ligand/receptor-dependent because antibody-mediated activation of FcγRIII (CD16) phosphorylates SLP-76 on both tyrosines simultaneously.

| Stabilizing the conformation of signalling nodes
Adaptor proteins can control enzymatic activity through allosteric control 5 ( Figure 2H). In particular, this pertains to pseudokinases, which can be perceived as adaptor proteins due to their loss of enzymatic activity. 122 For instance, in the ERK pathway, KSR pseudokinase can activate RAF kinase through its binding. 123 Other adaptors can also influence the enzymatic activity of their binding partners, such as phosphorylated PAG1, which increases the activity of tyrosine kinase CSK by 6-fold, significantly affecting TCR activation. 124 Similarly, SLP-76 binding of ITK stabilizes the kinase active state, allowing ITK to perform its functions. 89 Allosteric modulation can also be understood as conformational changes, which promotes differential binding preferences as a result of interaction with an adaptor protein. Such a case was described for the interaction between LAT and GRB2, where binding to one phosphotyrosine is increased upon phosphorylation of another tyrosine. 111 In addition, adaptor proteins provide a boosting platform for enzymatic reactions. This is termed substrate (metabolic) channelling and refers to a rapid transfer of a substrate between two subsequent enzymes. 125 A similar phenomenon was reported between protein kinase C and its adaptor AKAP7, but it was instead referred to as the scaffold state switching model. 126 The authors additionally confirmed the adaptor's ability to insulate the enzyme from both substrate competition and ATP competition, but not activation competition. Lastly, an important aspect of the interplay between adaptor molecules and the corresponding enzymes is the protection of moieties. An example of this is when an SH2 domain is bound to a phosphorylated tyrosine, thus protecting it from phosphatases. 127 In conclusion, adaptor proteins play a crucial role in fine-tuning signalling pathways. As such, it would be expected that in the absence of a given adaptor molecule, physiological processes will be perturbed resulting in human disease. Considering the importance of cellular signalling mechanisms for the flexibility and robustness of immune reactions, it is likely that genes encoding adaptor molecules may contribute to various disease phenotypes. To obtain a rough overview of diseases where adaptor proteins have been implied, we performed a search for associated disease phenotypes on DisGeNET. 128,129

HUMAN DISEASE
Genetic diseases are often associated with mutations or deletions in genes encoding proteins harbouring enzymatic activity. However, adaptor proteins, which interact with these enzymatically active proteins, may have important roles in a wide array of biological and physiological processes, given the possible functions they may accommodate, as discussed above.
In the following, we will explore the extent of how adaptor proteins have been implicated in genetic diseases. To this end, we took advantage of the DisGeNET repository, a collection of genes involved in human diseases, which are deposited in several publically available databases. This comprehensive platform integrates data from expert-curated repositories with text-mined data, GWAS catalogues as well as animal disease models. As such, the data range from large scale genome-wide studies to analysis of monogenetic diseases.
As adaptor molecules are heterogeneous and do not have one common feature that is applicable across all proteins, it is difficult to search for adaptor proteins in a given database. Therefore, to get insight into associated human diseases, we decided to focus on adaptor proteins that contain either SH2 and/or SH3 domains as the target genes. To assess the scope of genes implicated in disease, we performed a similar search for the SH2 domain-containing kinases. This search revealed on average twice as many diseases associated with SH2 domain-containing kinases than the number of diseases associated with SH2 domain-containing adaptor proteins. This may be explained by the more apparent phenotypes observed in kinase mutants as well as by the higher research attention they receive. Despite this, the search in DisGeNET revealed genes encoding adaptor molecules that are implicated in human genetic diseases. An overview of T A B L E 2 Overview of SH2-and SH3-encoding genes associated with human disease a selected genes associated with human diseases is shown in Table 2.

| Predisposition to cancer
DisGeNET classifies the associated diseases based on its nature into 'Disease Types': Diseases, Groups and Phenotype. We noticed that the search results are mainly of the type Diseases; that is, they are associated with specific diseases (eg Diabetes mellitus) ( Figure 3A). To get a better understanding of the type of diseases that the genes are associated with, we further grouped results in the type Disease based on their annotated MeSH code and/or semantic type (Box 1). This revealed that a majority of the SH2-and SH3-encoding genes are primarily associated with neoplasms ( Figure 3B). This is possibly a skewed result, due to the vast amount of research available in the field of oncology. Several of these studies are GWAS studies that have been a powerful tool in searching for cancer targets. 130 On the other hand, adaptor proteins are frequently implicated in signalling pathways that maintain normal cell growth and proliferation, including SH2B1, 131 BCAR1, 132 (Table 2). Therefore, it is perhaps not surprising that many adaptor molecules are reported to be involved in predisposition to cancer. Of particular immunological interest are adaptor proteins implicated in several leukaemias. The GRB2 family proteins, GRB2 and GADS, have been shown to mediate oncogenic fusion protein BCR-ABL-driven myeloid and lymphoid leukaemia, respectively. The SH2 domain of GRB2 interacts with BCR-ABL, which recruits another adaptor protein GAB2. 136,137 This complex subsequently activates the Ras signalling pathway and drives cellular transformation. Similarly, the SH2 domain of GADS binds BCR-ABL and creates a molecular hub, including SLP-76, which subsequently activates signalling pathways downstream of BCR-ABL. 137 Furthermore, CRK family proteins are suggested as mediators of BCR-ABL-dependent leukemogenesis. CRKL and, to some extent, CRK have been shown to bind to BCR-ABL in transformed granulocytes and myeloid precursor cells (reviewed in Ref. [135]). Together, these events exemplify how adaptors may be exploited in transformed cells to promote advantageous protein complex formation (Figure 2A), specific protein recruitment ( Figure 2C) and signal transduction ( Figure 2E), as discussed earlier.
Several cancer treatments aim at targeting proteins of the JAK-STAT pathway, in particular inhibiting kinases or other enzymatically active proteins that drive cancer transformation. However, some of these treatments are over time met with resistance. In efforts to circumvent this problem, several groups have started to develop inhibitors targeting adaptor proteins instead, for instance by downregulating BCAR1 in the tumour. 138 Similarly, both the SH2 domains of GRB2 and NCK1, which are implicated in several cancers, have been suggested as targets for cancer therapies. 58,139,140

| Monogenic disorders
In contrast to cancer, which is multifactorial, there are also distinct diseases caused by defined mutations in adaptor proteins. Several SH2 domain-containing adaptor proteins are associated with disease phenotypes, as previously reviewed. 141 A prominent example of an adaptor molecule involved in disease is SAP, encoded by the gene SH2D1A. This molecule contains a single SH2 domain 142 and is mainly associated with infectious diseases and diseases of the hematopoietic system ( Figure 3B). In 1998, three groups simultaneously identified loss-of-function mutations in SH2D1A, which were associated with X-linked lymphoproliferative disease (XLP) triggered by infection with Epstein Barr virus. [142][143][144] These mutations range from large deletions to point mutations and are mostly found within the SH2 domain and primarily cause structural mutations. 145 Normally, SAP associates with SLAM and competes with the phosphatase SHP-2 for the same binding site, thereby limiting the recruitment of SHP-2 to the receptor. Furthermore, SAP bridges SLAM with FYN, a kinase that can enhance phosphorylation of SLAM receptors. In individuals harbouring mutations in SH2D1A, SLAM receptor signalling in T cells is consequently abrogated which causes the symptoms observed in XLP. 146 Further, patients who have recovered

Box 1 Identification of adaptor proteins associated with disease in the DisGeNET database
DisGeNET is a comprehensive platform collecting gene-disease associations from several publically available databases. 128,129 To identify adaptor proteins associated with disease in this database, we first identified all human proteins that contain an SH2 and/or an SH3 domain in UniProt 212 (data retrieved on 16 April 2020). Of these, proteins with enzymatic activity such as kinases or guanine exchange factors, including those that have DNA-or RNA-binding activity or are involved in the endocytic pathway, were excluded from further analysis. All remaining genes were considered adaptor molecules, as defined within the scope of this review. Diseases associated with this list of genes were extracted from DisGeNET. Using the classification annotated by DisGeNET, the proteins were then grouped by Phenotype, Group and Disease, based on the nature of the associated disease as defined by DisGeNET. Within the Disease group, we further grouped the genes based on their MeSH codes and semantic type into one of the following groups: Neoplasms, Infections, Haematopoietic system, Nervous, Cardiovascular and Other. Several diseases in DisGeNET are assigned multiple MeSH codes, given the multifactorial nature of the disease. Genes associated with such diseases were placed into the group that was the initial cause of disease and not its consequence. For instance, for T cell lymphomas, which is assigned the following MeSH codes: Hemic and Lymphatic Diseases; Immune System Diseases; Neoplasms, were placed under Neoplasms, as this is the source cause of disease. The MeSH codes in our defined groups are divided as follows: Neoplasms (C04), Infections (C01, C02, C03), Haematopoietic system (C15, C20), Nervous (C10), Cardiovascular (C14) and Other (remaining MeSH codes). from symptomatic XLP or asymptomatic individuals with malfunctioning SAP have been shown to develop B cell lymphomas. 44,[142][143][144] The underlying mechanisms are still unclear; however, recent data suggest that SAP is crucial for CD8 + T cell-mediated immune surveillance of transformed B cells. In the absence of SAP, signalling through the SLAM receptor 2B4 is disrupted in T cells and proper activation of naïve CD8 + T cells is abolished. 147 Together, this illustrates how SAP may be involved in signal attenuation as well as signal multiplication, as discussed earlier ( Figure 2C and 2F).
Other examples of SH2-encoding genes associated with immune dysfunction are BLNK, 148 SH2B3 80,149,150 and SH2D2A. [151][152][153] This could be explained by a relatively higher expression of these adaptor proteins in hematopoietic cells ( Figure 1B). Similarly, mutations in SH3 domain-encoding genes are associated with specific diseases including MAPK8IP1 in diabetes mellitus, 154 NPHP1 in Joubert syndrome 155 and SH3PXD2B in Frank-Ter Haar Syndrome. 156 These latter examples are diseases of specific tissues, possibly owing to their functional importance in given cells. However, it may also reflect the diverse function adaptor proteins play in that particular tissue as well as across tissues, illustrating the concept of moonlighting.
Taken together, the search on DisGeNET for human diseases reveals the vast number and diversity of types of diseases which adaptor proteins are implicated in. Not only are they involved in multifactorial diseases such as cancers, but deletions or point mutations of the adaptor molecules alone can cause severe disease and developmental defects.

| Diseases associated with other adaptor proteins
SH2 and SH3 domain-containing proteins make up only a fraction of all adaptor proteins present in the human genome. And so, the diseases mentioned here illustrate just the tip of the iceberg in terms of the imperative role adaptors play in human health. For example, KSR1 has been implicated in malignancies 157 and increased expression of GAB2 is associated with mammary carcinogenesis. 158 Furthermore, the scaffolding protein SANS has been implicated in Usher syndrome. 159 Additionally, it remains important to bear in mind that disease-causing mutations associated with adaptor protein function may also occur in binding partners, which have lost their ability to bind to the adaptor protein. An example of this is in Noonan syndrome, where a mutation of RAF1 abrogates its interaction with 14-3-3 proteins, and consequently abolishes the autoinhibition of RAF1, otherwise promoted by 14-3-3 binding. 160 This lack of negative regulation results in developmental deformities observed in Noonan syndrome. As such, given their regulatory function, when exploring drug targets it is relevant to identify and consider all potential interaction partners of adaptor proteins.

| Mouse models
Groups interested in identifying the genetic cause of diseases mentioned above begin by searching for mutations by positional cloning (or more recently by whole genome sequencing) in families where more than one member has the disease. If a candidate gene is identified, the next questions are whether mutations also occur in the same gene in non-related patients and whether a disease mechanism can be identified in knockout (KO) animals where the gene of interest is deleted. While there are some exceptions, most of these KO models develop disease phenotypes similar to those observed in patients ( Table 3).
Examples of exceptions are NCK1 and NCK2, which have been implicated in several types of cancer. [161][162][163] In mice where either NCK1 or NCK2 are disrupted, the proteins seem to be functionally redundant and the mice are viable. However, when generating double KOs of these proteins, loss of NCK results in embryonic death. 164 Similarly, CRK seems to harbour a redundant function. 165 CRK KO mice are phenotypically normal and without embryonic abnormalities. This lack of phenotype is in stark contrast to the wide array of functions ascribed to CRK, including regulation of cellular processes from cell proliferation, migration, adhesion to apoptosis. 166,167 Examples where adaptor molecules appear to be redundant highlight several important issues. As mentioned above, the effect of mutations in genes encoding adaptor proteins may only be observed in specific tissues. Thus, creating a ubiquitous KO may not accurately recapitulate the disease phenotype as seen in humans. As such, it is important to consider conditional KOs, with a deletion of the gene in specific tissues. Although adaptor proteins initially seem to have redundant functions, they may also harbour multifunctional characteristics, which could explain the apparent lack of phenotype in KO animals.
The concept of multifunctional proteins was first termed as moonlighting proteins in the late 1990s, 17 and it is still very much relevant today. As discussed earlier, given their flexibility and multi-functionality, adaptors are prodigious in moonlighting. This may explain the diverse range of diseases including several types of cancer that many adaptor proteins are involved in. Keeping this in mind, the function of an adaptor protein may be revealed only under defined conditions, for instance during an infection with a particular microbe (eg SH2D1A [142][143][144] ). However, with an improved standard of living, health care and reduced exposure to triggering pathogens, this may limit the prevalence of disease phenotypes caused by mutations in genes encoding adaptor molecules. Thus, it may be that a pool of genetic variants exists that are disease-causing in the context of specific pathogens and that have yet to be identified. For instance, it is likely that in the current pandemic, the new coronavirus, SARS-CoV-2 may trigger specific COVID-19 disease phenotypes in the context of specific genotypes. This is particularly interesting since entire populations are thought to be susceptible to the virus. And while a large fraction of the population has already been infected, the resulting disease is highly variable both across different ages but also between seemingly healthy individuals of the same age and sex. Since the pathogenesis of SARS-CoV-2-induced disease is to a large extent dependent on a perturbed immune response against the virus, 168,169 examining whether underlying polymorphisms in adaptor protein-encoding genes is associated with COVID-19 susceptibility could be highly relevant.
The SH2D2A gene encodes the T cell-specific adaptor protein (TSAd), an example of an adaptor, whose function is still a matter of ongoing discussion. This gene harbours a number of expression quantitative trait loci (eQTL) (GTEx database); however, it has only been suggested to be associated with a limited number of diseases ( Figure 3B). The functional role of this adaptor, which is preferentially expressed in hematopoietic cells ( Figure 1B), will be discussed in the last section of this review.

WITH AN AMBIGUOUS ROLE IN CELLULAR SIGNALLING
As we described in the previous sections, adaptor proteins fulfil a vast number of functions in cellular signalling. However, their exact physiological role is often hard to define due to the absence of related diseases and mild or lack of phenotype in animal KO models. One example is the adaptor protein TSAd encoded by the SH2D2A gene. TSAd has two defined structural features: an SH2 domain and a long disordered C-terminal proline-rich tail containing four phosphotyrosines ( Figure 4A), which are both highly conserved across species ( Figure 4B). It has been found to interact with LCK, RLK and ITK as well as VEGF receptor 2 (VEGFR2), thus its alternate designations LCK adaptor protein (LAD), 170 RLK/ITK-binding protein (RIBP) 171 and VEGF receptor-associated protein (VRAP). 172 Two decades after it was first identified in 1998, 173 its role in cellular signalling is still only partially understood.
In the previous sections, we have presented a number of reasons highlighting why adaptor proteins may be understudied. Several of these reasons apply to TSAd including: (i) expression across various tissues, (ii) moonlighting functions, (iii) mild phenotype of SH2D2A KO mice, (iv) weak association with human disease, (v) multiple and distinct binding partners and (vi) poorly understood disordered regions, which constitute the majority of the protein.
In the following, we will present current knowledge about TSAd focusing on reasons why TSAd's role in cellular signalling has been challenging to define.

| Regulation of TSAd expression
Five alternative splicing variants of TSAd exist. Especially interesting is variant 5, which omits exon 7 and, as a result, does not contain the proline-rich region and the tyrosines. 174 F I G U R E 4 Structural overview and conservation of TSAd. A, Schematic representation of human TSAd. The unstructured regions (grey), SH2 domain (blue), proline-rich region (PRR) (red) and four conserved C-terminal pTyr sites (Y) are indicated. B, Curated ClustalO alignment of TSAd across ten species and alignment conservation annotation by Jalview. 257 Peptide sequence covering the SH2 domain is indicated with a blue bar, and PRR is indicated with a red bar and Tyr260, Tyr280, Tyr290 and Tyr305 are indicated with red asterisks. Aligned sequences are coloured with varying intensities of blue, from light to dark, indicating % similarity with the consensus sequence, from low to high, respectively. High conservation is indicated by high numerical values and bright yellow bars below the alignment. Black asterisks indicate conserved amino acids. Filled triangles among conservation values indicate sites with conserved physicochemical properties. '+' in consensus sequence indicates that the modal value is shared by more than 1 residue. Due to space limitations, we have excluded 55 N-terminal and 9 C-terminal amino acids in the chicken sequence and 9 C-terminal amino acids in the alligator sequence This is consistent with the observation that alternatively spliced protein variants are enriched in unstructured regions, 175 which is another way to increase intracellular signalling complexity.
As indicated in Figure 1B, TSAd (SH2D2A) is predominantly expressed in lymphocytes (specifically in T cells 173,176 and NK cells 177 ), endothelial 172 and epithelial cells. 178 TSAd expression is regulated by TCR 173,176 and cAMP signalling, 179 where the latter strongly induces TSAd mRNA expression in primary T cells. Additionally, protein kinase A (cAMP-dependent protein kinase) activity is required for TCR-dependent induction of TSAd expression. 180 Another potent activator of TSAd expression is a combination of phorbol myristate acetate and ionomycin (which bypasses immunoreceptor tyrosine-based activation motif phosphorylation), both in T cells and NK cells. 181

| TSAd in human disease
There are 474 eQTLs associated with SH2D2A, most of which are single nucleotide polymorphisms (SNPs) (as listed in the GTEx database). A polymorphism in the promoter region of SH2D2A, resulting in a shorter promoter sequence, has been associated with increased susceptibility to multiple sclerosis, 151 juvenile rheumatoid arthritis, 182 chronic inflammatory demyelinating polyradiculoneuropathy 153 and Sjögren's syndrome. 183 Further, T cells homozygous for shorter variants of the promoter region displayed lower levels of TSAd upon TCR stimulation. 151 A non-synonymous SNP resulting in serine to asparagine substitution at amino acid position 52 in TSAd increased susceptibility to multiple sclerosis 152 and ovarian cancer. 184 Furthermore, asparagine in position 52 of TSAd was associated with increased transcriptional activity and promoted TSAd interaction with LCK as measured by the yeast β-galactosidase reporter assay. 184 While these reports indicate that variation in TSAd expression may influence disease susceptibility, it is unclear how specifically this may affect the risk of disease.

| Phenotypes of TSAd KO mice
The SH2D2A-deficient mice were generated on the 129 genetic background and backcrossed to C57BL/6. 171 The initial report of SH2D2A-deficient mice having a mild autoimmune phenotype 185 has not been confirmed by our group. 186 It is possible that this discrepancy in observations is due to differences in the extent of genetic backcrossing, as many hybrid strains between 129 and C57BL/6 mice develop autoimmunity spontaneously. 187 However, a number of phenotypic characteristics have been revealed when the SH2D2A-deficient mice were challenged in various models. Tumour growth in TSAd KO mice is slower, possibly due to reduced angiogenesis. 188 Vessels in TSAd KO mice do not respond to VEGF with increased vascular permeability, and VEGF stimulation does not disrupt VE-cadherin junctions in endothelial cells lacking TSAd. 189 Additionally, TSAd KO mice are more resistant to myeloma development than TSAd wild-type mice in a myeloma-specific TCR-transgenic model. 186 Upon viral challenge, TSAd-deficient mice displayed reduced clearance of murine cytomegalovirus in the spleen. 181 Murine SH2D2A −/− CD4 + T cells exhibited impaired polarization in the immunological synapse of multiple molecules involved in TCR signalling. 190 Moreover, TSAd KO mice displayed accelerated rejection of heart transplants in an MHC class II-mismatched model, and resistance to graft-prolonging therapy of costimulatory blockade in the fully mismatched model. 191 As such, the TSAd KO phenotype is still not fully explored.

VEGFR signalling
TSAd has been reported to control the opening of adherens junctions of the endothelium and, consequently, vascular permeabilization. 188,189,192 Upon VEGF binding to VEGFR2, the receptor dimerizes and initiates downstream signalling, which engages tyrosine protein kinases. The SH2 domain of TSAd recognizes pTyr951 (Tyr949 in mice) in the cytoplasmic part of VEGFR2. Through this interaction, TSAd bridges VEGFR2 with SRC kinase, which can bind to prolines in the TSAd unstructured region. As a result of membrane translocation, SRC kinase phosphorylates the VE-cadherin's intracellular domain, which leads to the opening of adherens junctions. 189 VEGFR2 Y949F mutant mice have significantly reduced vascular permeabilization, which blocks molecular extravasation, oedema and metastatic cancer. 192

| TSAd in TCR signalling
While TSAd has been studied much more extensively in TCR signalling, there is no consensus, so far, on its role in T cells. However, the interaction of TSAd with LCK has been best characterized. LCK can bind with its SH3 domain to the proline-rich region of TSAd 174 and with its SH2 domain to TSAd phosphotyrosines (Tyr280, Tyr290 and Tyr305). 193 Additionally, the presence of TSAd increases the diffusion of the LCK clusters in the cell membrane. 194 TSAd promotes phosphorylation of ITK Tyr511 195 and LCK Tyr192, 196 but downregulates phosphorylation of LAT, SLP-76, PLCγ, ZAP70 and CD3ζ. 176,197 Our group has also shown a connection between LCK pTyr192 and TSAd pTyr290. Phosphorylation of the LCK SH2 domain on Tyr192 (as mimicked by LCK Y192E mutation) changed the domain's specificity and increased its affinity to TSAd pTyr290 peptides. 196

| Other interaction partners of TSAd
In addition to LCK, SRC and VEGFR, well-established interaction partners for TSAd also include ITK, which binds to the proline-rich region of TSAd via its SH3 domain. 195,198 In mice, TSAd has been shown to become phosphorylated upon PDGFR activation in bronchial epithelial cells. 199 This leads to the association of TSAd with PDGFR and GRB2. Furthermore, TSAd binds to MAP3K2 and co-localizes with it in the immunological synapse. 200 Upon EGFR stimulation, the TSAd-MAP3K2 interaction results in the activation of MAPK7 and JNK, most likely by facilitating their phosphorylation by SRC. 201,202 Upon stimulation with CXCL12 and CCL5 chemokines, TSAd brings together the β-subunit of G protein-coupled chemokine receptor, LCK and ZAP70, which is necessary for the activation of the latter molecule. 203 The SH2 domain of NCK1 can bind to TSAd pTyr280 and pTyr305, while its SH3 domains can bind to the TSAd proline-rich region. 204 This binding facilitates NCK1 interaction with LCK and SLP-76, which potentially promotes actin polymerization in T cells. Finally, TSAd is suggested to interact with DSCAM and DSCAML1 in neurons, recognizing their phosphotyrosine motifs, although the physiological relevance is unknown. 205 The main structural feature of TSAd is its SH2 domain. It is thus of particular interest to identify interaction partners to the TSAd SH2 domain in T cells. As previously mentioned, the TSAd SH2 domain can bind pTyr951 on VEGFR2. 189 Otherwise, it has been reported to interact with SMAD2 and SMAD3, which are involved in the TGF-β receptor (TGFBR) signal transduction pathway. 206 Additionally, the TSAd SH2 domain and the proline-rich region mediates binding to the laminin-binding receptor (RPSA), which promotes T cell migration. 207 TSAd may also bind via its SH2 domain to CD6 and LAT, 208 and Valosin-containing protein (VCP). 209

| TSAd-open questions in immune cell signalling
While TSAd may interact with cytosolic kinases in T cells and NK cells, TSAd's role in immune receptor signalling is still not well defined. The recent report that TSAd is critical for graft rejection has revealed a novel and potent phenotype. 191 However, its significance in the context of signalling in T cells still awaits deciphering.
Most of the TSAd protein sequence (75%) consists of an intrinsically disordered structure. To what extent this sequence has functional relevance is not well characterized. Mice expressing the TSAd SH2 domain in the absence of the disordered regions had T cells with reduced TCR-dependent production of IL-2, proliferation, migration and inflammatory responses. 210 Parts of the disordered region in TSAd are conserved between species 193 (Figure 4B). Predicting binding sites and motifs hidden within disordered regions remains difficult, as it is based on comparison with pre-existing knowledge. 39 Identification of short sequence motifs that are conserved between species will help identify regions of interest within the unstructured regions. This may allow testing of single mutations of conserved amino acids, which may pinpoint their importance or even function. Such mutations can be further tested using genome editing technology such as CRISPR/ Cas9, 211 which mediates deciphering roles of adaptor proteins in signalling pathways in relevant cell lines.

| CONCLUSION
As described in this review, a straightforward explanation of the role of adaptor proteins in cellular signalling is unattainable. This is partly due to the multitude of different functions that these molecules may fulfil. While many adaptor molecules have been characterized in great detail, others are still understudied, and a larger focus on this intriguing group of proteins is required. Adaptors may be perceived as redundant parts of signalling pathways since their absence often does not result in striking phenotypes. However, increasing evidence points towards adaptors being imperative cellular components, by providing intracellular signalling networks with essential flexibility and fine-tuning in a dynamic manner. This is particularly true for cells of the immune system, which have evolved to depend on adaptor proteins in their constant effort to respond adequately to external challenges and, ultimately, keep the organism alive.

ACKNOWLEDGMENTS
Part of the text in this review was also included in PB's doctoral thesis: Borowicz P Regulation of T cell activation by two conserved phosphotyrosines in Lck and its adapter protein TSAd. Oslo: Institute of Basic Medical Sciences, Faculty of Medicine, University of Oslo, 2020. The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health (US), and by NCI, NHGRI, NHLBI, NIDA, NIMH and NINDS.