Structural and functional diversity in the activity and regulation of DAPK-related protein kinases



Within the large group of calcium/calmodulin-dependent protein kinases (CAMKs) of the human kinome, there is a distinct branch of highly related kinases that includes three families: death-associated protein-related kinases, myosin light-chain-related kinases and triple functional domain protein-related kinases. In this review, we refer to these collectively as DMT kinases. There are several functional features that span the three families, such as a broad involvement in apoptotic processes, cytoskeletal association and cellular plasticity. Other CAMKs contain a highly conserved HRD motif, which is a prerequisite for kinase regulation through activation-loop phosphorylation, but in all 16 members of the DMT branch, this is replaced by an HF/LD motif. This DMT kinase signature motif substitutes phosphorylation-dependent active-site interactions with a local hydrophobic core that maintains an active kinase conformation. Only about half of the DMT kinases have an additional autoregulatory domain, C-terminal to the kinase domain that binds calcium/calmodulin in order to regulate kinase activity. Protein substrates have been identified for some of the DMT kinases, but little is known about the mechanism of recognition. Substrate conformation could be an equally important parameter in substrate recognition as specific preferences in sequence position. Taking the data together, this kinase branch encapsulates a treasure trove of features that renders it distinct from many other protein kinases and calls for future research activities in this field.


autoregulatory domain




calcium/calmodulin-dependent protein kinase


catalytic domain


death-associated protein kinase

DMT kinases

DAPK, MLCK, TRIO family kinases


eukaryotic protein kinase


myosin light-chain kinase


protein kinase A catalytic α-subunit

R spine

regulatory spine


C-terminal of Ras of complex proteins (ROC) domains


triple functional domain protein


Site-specific phosphorylation of target proteins is one of the most fundamental means of regulating a huge variety of cellular processes. More than 600 protein kinases have been identified in the human genome, and these are associated with a broad spectrum of target substrates. The systematic categorization of the human ‘kinome’ – primarily by sequence comparison of their catalytic domains – has been a seminal milestone in moving towards a complete and integrated understanding of their function, the regulation of their function and their structure [1].

One of the largest ensembles within the kinome is the calcium/calmodulin-dependent protein kinase (CAMK) group with a total of 113 members, indicating that many of these kinases are regulated by Ca2+-bound calmodulin (CaM). Those proteins share a common sequence segment of ~ 50 residues, referred to as the autoregulatory domain (ARD), which follows the catalytic domain (CD) that regulates kinase activity in a Ca2+/CaM-dependent manner. Although the ARD displays recognizable sequence patterns, such as a conserved positively charged residue cluster preceding the Ca2+/CaM-binding segment, it generally has a higher sequence diversity than the catalytic kinase domain, rendering the prediction of Ca2+/CaM binding unreliable. Experimental investigation of Ca2+/CaM-binding properties in individual members of the CAMK family has advanced our understanding of the underlying molecular mechanisms involved. However, these findings have also revealed several instances in which Ca2+/CaM binding could not be confirmed [2-5], leading to questions about alternative mechanisms of kinase activity regulation. At the structural level, mechanistic knowledge about the ARD conformational space remains scarce because, in the absence of Ca2+/CaM, most of the ARD tends to be invisible in available crystal structures. It is worth noting that there are also protein kinases outside the CAMK group, such as ribosomal protein S6 kinase and mitogen- and stress-activated protein kinase (MSK) [6, 7], which comprise a C-terminal ARD, indicating the limitations of the established protein kinase classification scheme [1].

Only two kinase structures have been determined in the presence of Ca2+/CaM and both of these show the respective binding segment to be helical, in agreement with structures of synthetic kinase peptides bound to Ca2+/CaM [8, 9]. Although it is evident from these structures and complementary biochemical data that Ca2+/CaM binding generally leads to removal of the ARD from within or next to the kinase active site, a detailed description of the molecular parameters required for Ca2+/CaM-dependent regulation is still lacking. It remains to be determined whether this type of regulation simply leads to active-site unblocking or whether additional mechanisms are involved – for instance, by inducing conformational changes within the kinase CD. In addition, our knowledge remains fragmentary on how nature has evolved to combine different mechanisms of regulation, in which Ca2+/CaM binding may or may not play a role. Key alternative mechanisms of regulation are by phosphorylation, binding of other protein ligands, changes in oligomerization state, or – as recently suggested – by external mechanical force [10-12].

Most previous overviews on molecular mechanisms of CAMK catalytic activity and regulation have focused on the canonical CaMKs I–IV [13-17], whereas considerably less attention has been given to other kinases of the CAMK group. In this review, we use death-associated protein kinase 1 (DAPK1) as a template and have selected those CAMK family members that are closely related to this kinase. In established kinome phylogenetic trees such as the one from the consortium ( [1], DAPK1 is a member of a distinct CAMK branch that includes three families: the DAPKs, myosin light-chain-related kinases (MLCKs) and triple functional domain protein-related (TRIO) kinases (Figs 1-4 and Table 1). There are 14 proteins in this branch and we refer to them subsequently as DAPK, MLCK, TRIO (DMT)-related kinases (Table 1). As two members of the TRIO family, obscurin and SPEG, contain two kinase domains, the total number of kinases within these sequences is actually 16. The main objective of this review is to extract molecular mechanistic principles from available data, mostly obtained by biochemical and structural biology approaches. We ultimately show that they are probably more divergent in their mechanisms of activity regulation than would be expected from the high level of sequence similarity in their catalytic kinase domains.

Table 1. List of DMT kinases. The PDB codes, used for figure generation, are indicated and, where applicable, were selected based on the highest structure resolution. Further structures of these kinases are found in the Protein Data Bank (
Short nameUniProtGene nameFull nameSynonymsSequence lengthPDB code CDPDB code CD+ARD
DAPK1 P53355 DAPK1 Death-associated protein kinase 1 1430 2W4J 2W4K, 2X0G
DAPK2 Q9UIK4 DAPK2 Death-associated protein kinase 2DRP-1370 1ZWS 2A2A
DAPK3 O43293 DAPK3 Death-associated protein kinase 3ZIPK, Dlk454 1YRP  
DRAK1 Q9UEE5 DRAK1 DAP kinase-related apoptosis-inducing protein kinase 1STK17A414  
DRAK2 O94768 DRAK2 DAP kinase-related apoptosis-inducing protein kinase 2STK17B372 3LM0  
MLCK1 Q15746 MYLK1 Myosin light chain kinase 1, smooth musclesmMLCK1914  
MLCK2 Q9H1R3 MYLK2 Myosin light chain kinase 2, skeletal muscleskMLCK596  
MLCK3 Q32MK0 MYLK3 Myosin light chain kinase 3, cardiac musclecaMLCK819  
MLCK4 Q86YV6 MYLK4 Myosin light chain kinase family member 4SgK085388 2X4F  
Titin Q8WZ42 TTN TitinConnectin34 350  1TKI
TRIO O75962 TRIO Triple functional domain proteinPTPRF-interacting protein3097  
Kalirin O60229 KARLN KalirinDUET, DUO, HAPIP, TRAD2985  
SPEG Q15772 SPEG Striated muscle preferentially expressed protein kinaseAPEG-13267  
Obscurin Q5VST9 OBSCN ObscurinObscurin-MLCK7968  
Figure 1.

Domain topology in DMT kinases. The members of the DMT kinase branch are grouped into the three families: DAPKs, MLCKs and TRIO. Color codes: green, CDs (indicated as ‘KIN’); orange, experimentally confirmed ARDs with an ability to bind Ca2+/CaM; yellow, ARDs predicted to bind Ca2+/CaM; dark blue, ARD of titin kinase; gray, leucine zipper regions; lime, signaling domains in the TRIO family. AK, ankyrin repeats; ROCO, C-terminal of Ras of complex proteins (ROC); DD, death domain; IG, immunoglobulin fold; FN, fibronectin type III; PH, pleckstrin homology; DH, paired double homology; S, Src homology 3; SP, spectrin repeat. The dimension of each domain is approximately proportional to its known or predicted length, as indicated by a sequence ruler at the foot of the figure. Where indicated with a ‘≈’, sequences were truncated to fit the scheme.

Figure 2.

Sequence relations in DMT kinases. (A) Alignment of all CD DMT sequences, grouped into the three families: DAPKs, MLCKs and TRIO. The C-terminal kinases of SPEG and obscurin are shown in a separate block, because they comprise several sequence features that are distinct from the remaining kinases of the TRIO family. Those sequences, for which structures of the CD are known (Table 2), have been aligned based on a structural superposition using pdbefold [83], and are shown with blue characters. The remaining sequences have been aligned with clustal w [84] and manually merged with the structure-based alignment. The starting residue of each sequence is numbered, according to those in the UNIPROT identifiers (Table 1). Sequence numbers and secondary structural elements relate to those observed in DAPK1 (PDB entry 2W4J) and the secondary structural labels follow the conventions of the literature [70] (cf. Fig. 4). Residues with specific functional assignments are labeled on top of the alignment as: N, nucleotide binding; C, C spine [70]; R, R spine [70] (Fig. 7). Residues that are either involved or predicted to be involved in an additional specific interaction with helix C, which is consistently found in DMT CD kinase structures, are highlighted in light green (Fig. 5); those anchoring the R spine (cf. Fig. 7) are in light orange. The central phenylalanine/leucine of the DMT-specific HF/LD motif is highlighted in blue, and those residues that either have been shown or are predicted to interact with this residue are highlighted in light blue (Fig. 6). Known phosphorylation sites are highlighted in light pink. Invariant residues (*), highly conserved residues (:) and conserved residues (.) are indicated in the consensus line and their positions are highlighted by different gray shadings in those sequences, in which conservation has been observed. Uninterrupted blocks of aligned residues in the CD structures superimposed, referred to as ‘subdomains’ in the text, are shown below the alignment in numbered boxes with rainbow colors (Fig. 4). For comparison, the aligned PKA C-α sequence is shown below. Previously defined sequence-based PKA C-α subdomains [69] are indicated. The positions of the basic loop and the activation segment are indicated. (B) Alignment of DMT kinase Ca2+/CaM-binding ARDs, starting from the last conserved CD residue: first block, kinases with experimentally confirmed Ca2+/CaM-binding and – in part – known ARD structure; second block, kinases with experimentally confirmed Ca2+/CaM-binding but unknown ARD structure. The ARD of titin kinase is shown for comparison, because it has been aligned using a structural superposition with DAPK1, indicating that there is no detectable sequence conservation. Secondary structural labels are highlighted on the top (cf. Fig. 8). Known ARD phosphorylation sites are highlighted in pink. To demonstrate the involvement of specific ARD residue positions in Ca2+/CaM binding, the surface area occupied by Ca2+/CaM binding is shown schematically, using the coordinates of the structure of the DAPK1–Ca2+/CaM complex [8]. Each box represents 25 Å2 of occupied surface area. Sequence positions crucial for Ca2+/CaM binding are numbered. All remaining colors and label definitions are as in (A). One-character abbreviations are used for amino acids (A, C, D, etc.), for reasons of clarity.

Figure 3.

Sequence relations in the DMT branch of CAMK kinases. Guide Tree depicting the degree of divergence between the CDs of DMT kinases, showing the clustering in the DAPK, MLCK and TRIO families. The tree was built using the neighbor joining method [85]. The calculated distance values are displayed in parentheses. DMT kinases with confirmed Ca2+/CaM-regulated ARDs are in orange; kinases with predicted Ca2+/CaM-regulated ARDs are in yellow; titin kinase with an ARD without the ability to bind Ca2+/CaM is in dark blue; the remaining DMT kinases without evidence or prediction of an ARD are not highlighted.

Figure 4.

Structural organization of the DMT CD. (A) Overview of the CD structure, observed in DMT kinases. The coordinates of the CD domain of DAPK1 (PDB code 2W4J) have been used. The side chains of the central phenylalanine of the HF/LD motif as well as the phenylalanine and aspartic acid of the DFG motif are shown in sticks. Color codes are as in Fig. 2. The basic loop, the activation segment and helix C are highlighted. The secondary structural elements are labeled. (B) The same structure, showing structurally rigid subdomains highlighted in colors used and labeled as in Fig. 2A.

It is beyond the scope of this review to provide a comprehensive summary of the temporal and spatial patterns of expression and localization of these kinases, which is, of course, ultimately essential to completely corroborate their functional readout in a physiological setting. To further simplify matters, we will restrict the review to those kinases found in the human genome and will not delve into aspects of organismal evolution. Many of the kinases discussed have several synonyms, which are listed in Table 1. Here, we mostly use the respective gene or gene product acronyms that are found in the UNIPROT database, which may not necessarily represent the most commonly used ones in the recent literature but which we hope will be useful for making connections between the different members.

DAPK family

The DAPK family includes five members (synonyms in parentheses): DAPK1, DAPK2 (DRP1), DAPK3 (ZIPK), DRAK1 (STK17A), and DRAK2 (STK17B). In all of these proteins, the kinase domain is situated at or near the N-terminus of the respective sequence. DAPK1, DAPK2 and DAPK3 have a sequence motif in common – KRXXXXSRRGV (residues 46–56 in DAPK1) – that is situated in the upper lobe of the catalytic domain (Figs 2, 4 and 5). Due to the insertion of about six residues in this motif when compared to other DMT kinase sequences, an extended loop is formed. As a result of the two adjacent arginines in positions 8 and 9 of the loop (residues 53 and 54 in DAPK1), this segment has been named the ‘basic loop’ [18], and these residues are highly exposed and oriented towards the active-site area. There is, however, no evidence that residues from this loop directly affect substrate binding [19], but they seem to be involved in interactions with other binding partners such as DAPK3 [20]. The remaining two kinases DRAK1 and DRAK2 have a shorter and unrelated insert with a KRRKGQD sequence motif (residues 94–100 in DRAK1) (Figs 2 and 5). The relationship between DAPK1/DAPK2/DAPK3 and DRAK1/DRAK2 is, therefore, defined by a relatively high level of sequence identity between other parts of their respective catalytic domains (Table 2). A common functional denominator seems to be their involvement in autophagic signaling pathways, although the level of experimental evidence varies for each of the individual members [21].

Table 2. Sequence and structure relations between DMT kinases. Pair-wise sequence identity percentages of the CDs for all DMT kinases are shown in the upper right. For those CDs with known structure the root-mean squares deviations for all matching residues, as determined by pymol (, are shown as well
DRAK1   664241413836293637373838
DRAK21.351.081.27 4141413835303537353535
MLCK3     74655443293538383640
MLCK40.890.890.88 1.14 645441293535383438
MLCK2       5438283636363635
MLCK1        40263236353736
Titin1.671.441.64 1.60 1.54  272930343434
Obscurin 2          4132343330
SPEG 2           35373632
Obscurin 1            463940
SPEG 1             3533
Kalirin              58
Figure 5.

Basic loop and helix C arrangement in DMT kinases. Zoom into the structural environment of helix C (green) and β-strand 4 (violet) of three DMT kinases: (A) DAPK1 (PDB code 2W4J), (B) titin (PDB code 1TKI) [55], (C) MLCK4 (PDB code 2X4F). A DMT-specific interaction between a positively charged residue from helix C and a highly conserved glutamate at the C-terminus of β-strand 4 is shown and labeled. The central glutamate/lysine pair in the background is part of the kinase active site. Hydrophobic leucines or methiones, part of the so-called R spine, linking the N-terminus of β-strand 4 to the C-helix are highlighted. The basic loop in DAPK1 is highlighted in purple. Two highly exposed arginines from this loop, which are invariant in DAPK1, DAPK2 and DAPK3 (cf. Fig. 2A), are shown in stick representation. Residue numbering is according to those in the UNIPROT identifiers (Table 1). One-character abbreviations are used for amino acids (A, C, D, etc.), for reasons of clarity.

In terms of their overall domain topology, members of the DAPK family can be categorized according to the presence of three sets of sequences (Figs 1 and 2). The first is found in DAPK1 and DAPK2, which both have an ARD next to their catalytic kinase domain (CD) that has the ability to bind Ca2+/CaM. A conserved autophosphorylation site within the ARD of these two kinases (Ser308 in DAPK1, Ser318 in DAPK2) determines CaM binding, which is impaired when this site is phosphorylated [22, 23]. Loss of CaM binding leads to deactivation of kinase activity [8]. DAPK1 has a long sequence tail of additional domains, C-terminal to the N-terminal CD/ARD module, which includes eight ankyrin-binding repeats, a ROCO domain, and a C-terminal death domain [21, 24]. The presence of the ankyrin repeats leads to its prominent association with actin stress fibers [25]. The ROCO domain contributes to the regulation of DAPK1 kinase activity via GTP binding and subsequent Ser308 autophosphorylation [26]. The DAPK1 death domain interacts with several protein partners and has been shown to be involved in regulating the functional readout of DAPK1 [27-30]. A number of DAPK1 substrates have been identified and characterized, such as myosin regulatory light chain 12B and Beclin-1 [25, 31, 32]. However, these substrates do not show an unambiguous sequence preference pattern [33, 34], suggesting that additional parameters, including possible conformational preferences, may play a role in substrate recognition. In contrast to DAPK1, the only additional module in DAPK2 is a C-terminal dimerization element [35]. Dimerization is promoted by Ser318 dephosphorylation, which stimulates DAPK2 activity [23].

The second set of DAPK family sequences has a C-terminal leucine zipper motif in common that has been characterized as a dimerization module in various proteins [36] (Fig. 1). This motif was initially found in DAPK3, which lacks an ARD with Ca2+/CaM-binding ability. Interestingly, a similar motif has recently been found in a second isoform of DAPK2 (DAPK2β), denoted as DRP-1β [4]. This isoform also lacks a Ca2+/CaM-binding ARD, in contrast to the original DAPK2α sequence, perhaps indicating that functional regulation by Ca2+/CaM binding and leucine zipper-mediated dimerization could be mutually exclusive. Lastly in DAPK3, three serine/threonine phosphorylation sites within the CD – Thr180, Thr225 and Thr265 – have been shown to play a role in activity regulation [37]. These sites are invariant in all members of the DAPK family (Fig. 2), suggesting that these kinases may all be additionally regulated by serine/threonine phosphorylation, but a more systematic experimental proof is still required.

Interestingly, additional functional and physical crosstalk has been reported for DAPK1 and DAPK3 [20], showing direct kinase domain-mediated interactions and indicating DAPK3 to be a DAPK1 substrate but not vice versa. DAPK1/DAPK3 interactions, together with phosphorylation of the ribosomal protein L13a, have been shown to establish a regulatory module in inflammatory gene expression [38]. One prominent mechanism for kinase domain-dimerization, at the level of homodimerization and possibly also heterodimerization, is the exchange of activation segments within a dimeric interface [39]. Recent structural data indicate that alternative mechanisms of dimerization may also be conceivable [40]. In contrast to the first three members of the DAPK family, less is known about DRAK1 and DRAK2. Both of these kinases do not possess any established additional activity-regulating modules such as a CaM-binding segment or a leucine zipper motif. DRAK1 has been shown to contribute to pro-death activity, whereas DRAK2 has been indicated to have a role in T-cell regulation [41, 42]. Interestingly, DRAK2 activity is regulated by Ca2+-dependent phosphorylation by protein kinase D [43].

MLCK family

The second DMT family derives its name from myosin light-chain kinase (MLCK) and comprises smooth muscle MLCK (MLCK1), skeletal muscle MLCK (MLCK2) and cardiac muscle MLCK (MLCK3) [5, 44-46]. There is a fourth MLCK-like gene product, named MLCK4 (SgK085), which has not been functionally characterized to date. In contrast to members of the DAPK family, which generally show multiple substrate specificities, MLCKs specifically phosphorylate the regulatory light chain of different muscle and nonmuscle myosins [47, 48]. Tissue-specific myosin substrate preferences can be correlated with distinct expression patterns of the respective kinases, indicating that there are specific kinase/substrate relationships.

All three characterized MLCKs contain a CD/ARD module, reminiscent of those found in DAPK1 and DAPK2 (Figs 1 and 2). However, whereas DAPK1 and DAPK2 have an N-terminal CD/ARD, this module is located at the very C-terminus of the MLCK2 and MLCK3 sequences, and in MLCK1 an additional Ig domain, termed IgT, is present between the CD/ARD module and the sequence C-terminus. This IgT domain binds to the heavy meromyosin subfragment of smooth muscle myosin and thus helps to tether the MLCK1 CD/ARD module to its target-binding site [49, 50]. IgT-mediated myosin binding by MLCK1 competes with a small MLCK1 isoform known as telokin that comprises the IgT domain only. The activities of two of these MLCKs – MLCK1 and MLCK2 – are regulated through Ca2+/CaM binding, as suggested by the presence of the CD/ARD module, although the functional implications differ. Whereas regulatory light-chain phosphorylation of smooth muscle myosin by MLCK1 is essential in the contraction cycle, in skeletal muscle the principle cause of contraction is by Ca2+-bound troponin C, suggesting that MLCK2-mediated regulatory light-chain phosphorylation has a regulatory role rather than being an essential requirement [46, 48].

The precise role of CaM regulation in MLCK3 is less well-characterized. In contrast to MLCK1 and -2, substrate turnover for phosphorylation by MLCK3 is much lower and seems to play a role in regulatory light-chain baseline phosphorylation [51]. MLCK3 contains a CD/ARD module, comprising the key residues required for CaM binding (Fig. 2B), but currently available data do not provide a conclusive picture on the role of Ca2+/CaM in MLCK3 regulation [5, 51].

In the absence of available experimental data on the fourth MLCK, MLCK4, a possible role for CaM in its kinase activity remains a matter of speculation. The MLCK4 sequence comprises a C-terminal tail of slightly more than 20 residues (residues 365–388). Although the tail contains a positively charged residue stretch (KKKNR, residues 374–378), which is a typical signature motif N-terminal to a CaM-binding segment, the remaining tail is too short to allow for the type of CaM binding found in other kinases with a proven CD/ARD module. A key signature residue, represented by a tryptophan within this motif (Trp305 in DAPK1) (Fig. 2B), is also missing.

An outlier – titin kinase – represents the fifth member of the MLCK family. Titin is the largest protein encoded by the human genome with > 30 000 residues and occurs in various isoforms of variable sequence length [52]. Full-length titin spans across half of the longitudinal filament of muscle sarcomeres, from the Z-disk to the central M-band over distances exceeding 1 μm [10, 53]. Data on titin kinase substrates are still scarce. Telethonin (also known as TCAP), a dynamic protein specifically found in the Z-disk of muscle sarcomeres and characterized as an N-terminal titin filament assembly-mediating ligand [54], has been identified as an in vitro substrate [55] but an effective role in a more physiological setting is still pending. Additional substrate interactions between titin kinase and NBR1 and p62/SQSTM1 have been reported, which link titin kinase activity to muscle autophagy [56]. However, this mode of action could only be shown in vitro for modified titin kinase constructs, either by ablating the C-terminal ARD module or by mimicking Tyr32341 phosphorylation (Fig. 2A), a residue located on the P + 1 loop of the CD [55].

Although titin kinase was initially thought to be regulated by Ca2+/CaM because of the presence of a CD/ARD module apparently reminiscent of those found in DAPK1 and DAPK2 and the MLCKs (Fig. 2B), several structural and biochemical studies clarified that the ARD of titin kinase differs from other ARDs [10]. Remarkably, the autoinhibited structure of titin kinase shows the complete ARD as a folded module bound into the kinase active site and anchored at its C-terminus into a small β sheet of the kinase lower lobe [55]. Such a structure, despite major efforts in this direction for DAPK1 (de Diego, Temmerman & Wilmanns, pers commun), has never been seen for any other kinase with a CD/ARD module. In an attempt to rationalize potentially alternative mechanisms of titin kinase regulation, Gautel's group have explored physiologically meaningful features in the sarcomeric M-band environment, where titin kinase is located due to its position close to the C-terminus of titin. In this sarcomeric microenvironment, pulling forces in the order of 30 pN are generated by myosin filament movements, as a result of the action of an array of myosin motor domains. Atomic force spectroscopy experiments confirmed that ATP binding to titin kinase could be enhanced under low external forces [57], suggesting a novel mechanism of kinase activation for which titin kinase could serve as a paradigm. Indeed, stretch-induced phosphorylation by MLCKs has been reported [47, 58], but this molecular mechanism has not been verified in the absence of appropriate structural and biochemical data. Remarkably, titin kinase and MLCK1, in contrast to the other MLCKs, share an extensive topology of additional structural domains, most of which are either structured into an Ig-like or a fibronectin III-type fold, C-terminal and N-terminal to the respective kinase domain, respectively. Available data indicate that many of these domains are involved in interactions with similar domains either from other muscle or cytoskeletal proteins and/or allow the formation of large assemblies, partly by promoting well-established filament structures. If a mechanically driven regulation of kinase activity can be further confirmed by additional in vivo data, a fascinating question on the possible role of these neighboring domains in kinase activity regulation would emerge.

The TRIO family

The four members of the TRIO family – Trio, kalirin (Duo), SPEG and obscurin – represent the most diverse and complex family among the three discussed in this review. In terms of function, the four proteins can be divided into two rather disconnected pairs. Trio and kalirin play roles in neuronal development, specifically in the growth cones of immature neurons, and act as guanine-exchange factors [59, 60]. Of the five experimentally validated kalirin isoforms, only the full-length (isoform-1 or kalirin-12) and isoform-4 (also referred to as Duet or Trad) contain a C-terminal kinase domain [59].

Obscurin and SPEG are specifically expressed in cardiac and skeletal muscle cells, and thus are more closely related to members of the MLCK family [61, 62]. The two proteins share substantial sequence similarity and matching domain structure, suggesting a common evolutionary origin. Whereas SPEG has not yet been characterized in detail, research on obscurin has advanced considerably. This protein occurs in several isoforms and, depending on the variant expressed, it either localizes to the central M-band or peripheral Z-disk of muscle sarcomeres [63]. Several binding sites for other sarcomeric protein components, including titin and myomesin, have been identified and characterized. Recent imaging studies indicate that obscurin may wrap around sarcomeric myofibrils and connect to other cell organelles such as the sarcoplasmatic reticulum but the molecular details of this arrangement are still unknown [63-66].

All four proteins comprise very long sequences exceeding 2000 residues, with a protein kinase domain near the C-terminus (Fig. 1). Interestingly, no substrate has been identified for any of these proteins. All four C-terminal kinase domains are followed by a sequence signature motif that may be indicative of Ca2+/CaM binding (Fig. 2B), but none have been experimentally verified to date. The potential ARD modules of SPEG and obscurin harbor a tryptophan in the ‘1’ CaM-recognition position [14], which is invariant in all ARD modules that bind to Ca2+/CaM in the DMT branch and is thus a strong indicator for such a function. By contrast, in Trio and kalirin, the respective position is replaced by a histidine, and it remains to be seen whether such a side chain can be tolerated in this position. Another remarkable feature is the occurrence of a second kinase domain in obscurin and SPEG, C-terminal to the common central kinase domain, but which lacks an additional adjacent ARD module.

Moreover, three members of this family – Trio, kalirin and obscurin but not SPEG – contain a conserved arrangement of signaling domains, including a Src homology 3 domain, a paired double homology domain, which is equivalent to a Rho-guanine nucleotide exchange factor module, followed by a pleckstrin homology domain. This arrangement is repeated twofold in Trio and kalirin but not in obscurin. In obscurin, the Rho-guanine nucleotide exchange factor module has been shown to bind to RanBP9, which may localize the C-terminus close to the Z-disk of sarcomeres [67] and activate small GTPases [68].

Structure-based insight into DMT protein kinase active-site conformation

At present, there is a rather imbalanced repository of DMT protein kinase high-resolution crystal structures. By far the most thoroughly characterized DMT kinase is DAPK1 with a total of 32 entries in the Protein Data Bank. Structures of this kinase cover the CD only, CD+ARD in the absence of further ligands, and CD+ARD in the presence of Ca2+/CaM, but structural details of the remaining non-kinase domain structure of DAPK1 (Fig. 1) are still unknown. Several DAPK1 CD structures have been determined in the presence of nucleotide ligands and small molecule inhibitors (Table 2). Similarly, several structures of the related DAPK2, both CD and CD+ARD, are available, but neither a complex with Ca2+/CaM nor a full-length DAPK2 structure including the C-terminal dimerization module has been determined to date. Other structures from members of the DAPK family are those of the CDs of DAPK3 and DRAK2.

Structures of members of the MLCK family are limited to that of MLCK4, which still lacks functional characterization, and the ‘outlier’ extended kinase domain of titin. No structural information exists for any member of the TRIO family as yet. As far as possible, we combine sequence and structural data to draw conclusions in terms of common and diverging structural/functional principles in DMT kinases. We specifically focus on extracting structural information that will confirm DMT kinases as functional eukaryotic protein kinases (EPKs) and on kinase-specific mechanisms of activity regulation.

Superposition of the known structural coordinates of the DMT kinase CDs shows that their 3D structures are highly preserved, leading to rmsd values of the structurally matching parts of the sequences in the range 0.4–1.6 Å (Table 2). This primarily depends on the level of sequence identity, which ranges from > 80% (DAPK1, DAPK2, DAPK3) to < 40%, especially for sequence comparisons with titin (Table 2 and Fig. 3). A structure-based alignment of the kinase domains with known structures reveals a total of eight structurally conserved subdomains. Whereas the entire lower lobe defines one single rigid subdomain (subdomain VIII in Figs 2 and 4), there is considerably more conformational flexibility in the upper lobe with a total of seven subdomains. This variability seems to represent functional diversity across the members of the DMT kinases, as described in more detail below. The alignment of the remaining DMT kinase CD sequences is unambiguous, except for the very N-terminus (Fig. 2).

Our analysis on DMT kinases is largely in agreement with a previous one performed on all EPKs, illustrated with the catalytic α-subunit of protein kinase A (PKA) [69] (Fig. 2). The four C-terminal subdomains found in the global EPK analysis indeed merge into one in the structure-based analysis of DMT kinases, indicating a lack of conformational variability in the C-terminal lobe, in marked contrast to the N-terminal lobe.

Using the criteria established by the Taylor group [70, 71], we have identified most of the molecular properties required to qualify DMT kinases as EPKs. However, all DMT kinases lack the ability to regulate catalytic activity by phosphorylation within the activation segment, which has been widely described as a key property of EPKs. Strikingly, none of the 16 DMT kinases have an HRD motif C-terminal to helix E that represents a signature for interaction with a phosphorylated serine/threonine in the activation loop. Instead, in all DMT kinases the second sequence position (residue 138 in DAPK1) is conserved either as leucine or phenylalanine, as such replacing the canonical HRD motif by a DMT kinase-specific HF/LD motif (Figs 2 and 6). In all known DMT kinase structures, this residue is embedded in five residues, which are predominantly hydrophobic and collectively create a local hydrophobic core (Fig. 6). In DAPK1, these five residues are Ala136, Ile168, Ile177, Ile188, Leu194 and Ala198 (Fig. 2). The remaining DMT kinases structure is known form an equivalent hydrophobic core with some variation in contributing residues (Fig. 6). The last residue of this cluster, Ala198 in DAPK1, is substituted by a serine or threonine in most other DMT kinase sequences (Fig. 2). There is, however, no evidence that this residue could be a target for specific phosphorylation. This HF/LD motif-mediated hydrophobic cluster is in marked contrast to the same area in canonical EPKs, which is dominated by an HRD motif-mediated arginine–phosphogroup salt bridge when the kinase is in an activated state (Figs 2 and 6). In the absence of activation loop phosphorylation, this area is, at least partly, disordered in the inactive conformation. In all DMT kinase structures, the phenylalanine of the invariant DFG motif (residue 162 in DAPK1) (Fig. 6) has an ‘in’ conformation [70], indicating an activated status. This observation can be confirmed graphically by displaying a set of four residues non-contiguous in sequence, known as the regulatory (R) spine [70], as an uninterrupted core of four side chains in all known DMT kinase structures (Fig. 7).

Figure 6.

HF/LD motif-mediated arrangement of the local hydrophobic core in the activation segment of DMT kinases. Zoom into the structural environment of the HF/LD motif of three DMT kinases: (A) DAPK1 (PDB code 2W4J), (B) titin (PDB code 1TKI) [55], (C) MLCK4 (PDB code 2X4F). (D) For comparison, the equivalent area of PKA C-α (PDB code: 1ATP) [86], which presents a prototype protein kinase with an HRD motif, is shown as well. The phenylalanine/leucine of the HF/LD motif (Phe138 in DAPK1; Phe32297 in titin; Leu226 in MLCK4) and the phenylalanine of the DFG motif (Phe162 in DAPK1; Phe32319 in titin; Phe248 in MLCK4, Phe186 in PKA C-α) are shown by sticks in blue. All other side chains that contribute to the core around the central HF/LD residue are shown as well and highlighted in beige or, those within the activation segment, in cyan. In PKA C-α, the hydrogen bond interaction between the central arginine side chain of the HRD motif (Arg166) and phosphorylated Thr198 is indicated. One-character abbreviations are used for amino acids (A, C, D, etc.), for reasons of clarity.

Figure 7.

Assembled R spine in DMT kinases. (A) DAPK1 (PDB entry 2W4J); (B) titin kinase (PDB entry 1TKI) [55]. The residues contributing to the R spine are shown by a mixed stick/surface presentation in blue and are labeled. The R spine is anchored by its bottom residue (His137 in DAPK1, His32296 in titin), via main chain-mediated hydrogen bonds, to either one or both residues, shown by orange sticks, from helix E (His131 in DAPK1; His32290 in titin) and helix F (Asp199 in DAPK1, Asp32356 in titin). One-character abbreviations are used for amino acids (A, C, D, etc.), for reasons of clarity.

As in canonical EPKs, the R spine is anchored via the HF/LD motif to a conserved aspartate (residue 199 in DAPK1) from the long helix F within the lower kinase lobe. This residue is invariant in all DMT kinase sequences. The subsequent GHI subdomain (Figs 2, 4 and 6), which comprises a bundle of three helices from the lower lobe and is considered another signature for EPKs, is connected via a highly conserved stabilizing salt bridge, formed by Glu187 of the signature APE motif and Arg263 in DAPK1, similar to other EPKs [72]. Because these two residues are invariant in all DMT kinase sequences, it can be safely assumed that the interaction is also conserved. These observations indicate that the HF/LD motif-mediated hydrophobic core in DMT kinases does not have an impact on otherwise conserved structural arrangements in the lower lobe. In terms of future studies, it would be interesting to determine whether this mechanism of activation, present in most EPKs, can be replaced by alternative mechanisms of activity regulation in DMT kinases.

Proper orientation of the DFG motif leads to additional sets of interactions in the kinase active site, which are sensitive to the kinase activity status. The aspartate from the DFG motif (residue 161 in DAPK1) (Fig. 6) generally interacts with the Mg2+ ion associated with the nucleotide substrate and is in close proximity to a salt bridge between two residues that are invariant in all EPKs, Lys42 and Glu64 in DAPK1 (Fig. 5). In the absence of this interaction, major movements of helix C, on which Glu62 is located, have been observed, leading to a generally inactive active-site conformation [73]. This interaction is found in all of the structures of the DAPK family (DAPK1, DAPK2, DAPK3, DRAK2), but in the two known structures of the MLCK family (titin and MLCK4) the distance is increased by ~ 2 Å. Whether this indicates inactivation of the active-site conformation remains unknown, as no major diversity in terms of helix C orientation is detectable. Interestingly, we have not observed major conformational changes in the nucleotide-binding pocket in any DMT kinase structure, regardless of whether a nucleotide ligand is bound, leaving any regulation by the nucleotide substrate an open question.

Helix C, in turn, is relatively short in members of the DAPK family and is preceded by the ‘basic loop’, which has been described as a signature motif for members of this kinase family [21]. Despite any lack of sequence similarity in the DAPK1/DAPK2/DAPK3 and the DRAK1/DRAK2 sub-families (Figs 2 and 5), the basic loop is similar in overall conformation and highly exposed in all available DAPK family structures. It remains to be seen whether differences in the residues of this basic loop result in any additional specific functional readouts for the two DAPK sub-families.

By contrast, helix C in the two structures of the MLCK family – titin and MLCK4 – is longer by about one turn, and is preceded by a sharp turn that connects this helix with the next β strand. In many canonical EPKs, helix C is preceded by another short helix B, which is missing in all DMT kinase structures. Interestingly, in most of the 16 DMT kinase sequences, we found another pair of positively and negatively charged residues – Arg58 from helix C and Glu84 in DAPK1 – that form a salt bridge in all available DMT kinase structures (Figs 2 and 5). In titin kinase, in which the residue equivalent to Arg58 in DAPK1 is substituted by a glutamine, the partner of the positively charged residue is shifted by one turn in helix C (Lys32220 in titin), whereas the second, negatively charged residue position is highly conserved in DMT sequences. By contrast, these two residues are generally variable in canonical EPKs. Thus, this additional interaction may serve as a DMT kinase-specific structural feature that locks helix C in a defined orientation and supports an active conformation. We have also noted that further hydrophobic residues of helix C, specifically Leu68 (DAPK1) that belongs to the R spine residue sequence [74], are equally involved in interactions with other residues from the upper kinase lobe, including the highly conserved Leu79 from β-strand 4 (Fig. 5), which is part of the R spine as well. A consequence of a firmly locked helix C and the HF/LD motif that immobilizes the activation segment is a constitutively assembled R spine conformation, observed in all DMT kinase structures irrespectively their activity state (Fig. 7).

C-terminal tail-mediated activity regulation of DMT kinases

Because the DMT kinases belong to the CAMK group, it may be implied that these kinases are regulated by ARD-mediated Ca2+/CaM binding. Using previously identified Ca2+/CaM-binding sequence signatures [14] as well as structural data from a number of synthetic Ca2+/CaM-binding DMT kinase peptides in complex with Ca2+/CaM [75, 76] and the structure of the DAPK1– Ca2+/CaM complex [8], we have unambiguously identified a C-terminal Ca2+/CaM-binding motif that is only present in a minority of DMT kinases (Fig. 2B). All these Ca2+/CaM-binding sequences have an invariant tryptophan in position 1 [14], which is deeply buried within the known Ca2+/CaM complexes (Fig. 8). When this residue is mutated, Ca2+/CaM binding is significantly impaired or even abolished, depending on the type of mutation, both in the context of kinase Ca2+/CaM-binding peptides and complete kinase sequences [8]. In addition, preceding this tryptophan is a highly conserved positively charged residue cluster, which is involved in interactions with both the respective kinase CDs and Ca2+/CaM. Available Ca2+/CaM complexes indicate additional crucial interactions at positions 8 and 13/14. In the DMT kinase sequences with known structure, hydrophobic residues are found at positions 8 and 14, whereas position 13 is conserved as an arginine in the sequences of DAPK1, MLCK1 and MLCK2 (Fig. 2B). The structures available, however, do not reveal a unique specific and conserved interaction between this arginine and Ca2+/CaM. The structure of the DAPK2 peptide–Ca2+/CaM complex (PDB code 1WRZ) additionally shows that alternative amino acids, such as a histidine in DAPK2, can also be accommodated.

Figure 8.

CD–ARD arrangement in DMT kinases. (A) DAPK1–Ca2+/CaM complex (PDB code 2X0G) [8]; (B) titin kinase (PDB code 1TKI [55]; (C) DAPK2 (PDB code 2A2A); (D) for comparison, CaMKIδ (PDB code 2JC6). Colors are as in Fig. 2B, Ca2+/CaM is shown in raspberry. The insert in (A) shows the structural environment of the Ca2+/CaM-binding signature residue Trp305 in DAPK1. One-character abbreviations are used for amino acids (A, C, D, etc.), for reasons of clarity.

Based on these structural observations and sequence alignments, we predict the presence of a Ca2+/CaM-binding motif for MLCK3, as well as the N-terminal first kinase domain in obscurin and SPEG, because they contain a tryptophan in position 1 as a hallmark for Ca2+/CaM-binding. By contrast, there is no indication of a Ca2+/CaM-binding motif following each C-terminal kinase domain in obscurin and SPEG, raising the possibility of diverging functions, or even lack of function, in the twofold repeated kinase domains of the two proteins. Even in each of the predicted Ca2+/CaM-binding motifs, there are significant amino acid deviations in the remaining key binding positions, rendering experimental verification essential. The ARD of titin kinase may serve as a negative control sequence, as it has been shown that its ARD sequence does not bind Ca2+/CaM [10].

Structures of DMT kinases encompassing the C-terminal ARD are currently confined to DAPK1, DAPK2 and titin kinase. Except for the structures of titin kinase and the DAPK1–Ca2+/CaM complex, visibility of the ARD has been limited to the N-terminal part, covering ~ 25 residues of the ARD. The conformation of this sequence segment is highly conserved in all available structures and folds into two consecutive helices that wrap around the bottom of the lower kinase lobe (Fig. 8). This part of the ARD is virtually identical to those of known structures of canonical CAMKs (Fig. 8), indicating that whenever Ca2+/CaM regulation occurs, the complete arrangement including the CD, the N-terminal part of the ARD, and the Ca2+/CaM-binding segment of the ARD is structurally preserved.

In DMT kinase structures with a Ca2+/CaM-binding motif, the ARD generally becomes invisible from the positively charged residue cluster preceding the Ca2+/CaM-binding sequence onwards. This suggests that the C-terminal part of the ARD, in the absence of Ca2+/CaM, is either unfolded or too flexible to allow visualization in crystals. Autophosphorylation of residues in the Ca2+/CaM-binding segment has been reported for two related kinases, Ser308 in DAPK1 and Ser318 in DAPK2. Extending the available structures by modeling suggests that Ser308 may well fit into the respective kinase active site as a target residue for phosphorylation [8]. The Ca2+/CaM-binding segment of at least these two kinases may therefore also function as a pseudosubstrate and could allow positional preferences for trans substrates to be derived [77]. The pseudosubstrate nature of the DMT Ca2+/CaM-binding segment may explain why these segments seem to be flexible in the available structures.

Strikingly, the structure of the DAPK1–Ca2+/CaM complex reveals that the Ca2+/CaM-binding sequence – in the context of a complete kinase CD – folds into a long helix [8], confirming data from several kinase peptide–Ca2+/CaM complexes. However, in the structure of the DAPK1–Ca2+/CaM complex, Ca2+/CaM only wraps around approximately two-thirds of the ARD Ca2+/CaM-binding segment, in contrast to several kinase peptide–Ca2+/CaM complexes in which virtually complete wrapping has been observed [78]. Whether the structural arrangement in the DAPK1–Ca2+/CaM complex reflects an activated conformation awaits further validation as Ca2+/CaM in part blocks the kinase active site. Interestingly, in another structure of the CaMKII–Ca2+/CaM complex, the CaMKII ARD–Ca2+/CaM is disconnected from the CaMKII CD [9]. In both structures, crystal contacts may have promoted a particular arrangement, indicating that complementary structural biology methods in solution will be essential to further characterize possible conformational plasticity.

In contrast to all DMT kinase structures that include the CD and ARD modules, the structure of titin kinase shows the complete ARD folded into the kinase active site [55]. This unique observation has also been confirmed in structures of the related kinase domain of twitchin from Caenorhabditis elegans [79, 80]. Whereas the conformation of the N-terminal part of the ARD is virtually identical to those observed in DAPK1 and DAPK2 (Fig. 8), key signatures of the Ca2+/CaM-binding motif, such as a positively charged residue cluster followed by a tryptophan, are missing in the titin kinase ARD sequence. In the subsequent C-terminal titin kinase ARD, numerous additional interactions with residues from the CD have been observed. The very C-terminus of the ARD is integrated into an additional small β sheet with the CD and provides a second anchor between the two domains, thus providing a structural rationale for tight ARD binding. For those DMT kinases with a Ca2+/CaM-binding module, their isolated CDs can be purified as stable proteins but, interestingly, the separate titin kinase CD in the absence of the subsequent ARD is less stable as purified protein [55]. This perhaps indicates that the ARD in titin kinase has an additional function in maintaining structural stability beyond its suggested role in functional regulation.

In a previous review, we speculated about a different mechanism of activation for titin kinase that we referred to as a ‘looping out’ of the ARD as opposed to it ‘falling apart’, which we suggested for those kinases possessing an ARD with the ability to bind Ca2+/CaM [81]. The latter mechanism has since been confirmed for CaMKII, which, however, does not belong to the DMT kinase branch discussed here [9]. Although in the structure of the DAPK1–Ca2+/CaM complex the ARD has not completely fallen apart, any kind of ‘looping out’ mechanism, as suggested for titin kinase, can be excluded. Little is known about how titin kinase could be activated at the molecular level. Molecular dynamics studies have provided ideas on possible disassembly mechanisms under external force conditions, but experimental proof is still lacking [82].

Future perspectives

Compared with many kinases of the human kinome in general and canonical CAMKs specifically, little is known about the DMT branch of CAMKs. However, available sequence, structural and functional data point to the lack of activation by phosphorylation within the activation segment as their most common and prominent feature, as well as their high level of sequence similarity. Furthermore, all available DMT kinase structures, irrespective of their functional state, show the conserved DFG motif in an active ‘in’ conformation and do not indicate conformational variability of helix C, as found in many other protein kinases, in which this helix plays a critical role to regulate catalytic activity. As a consequence of these structural findings, both R and C spines are constitutively assembled, suggesting that conformational variations associated found in many other kinases do not apply to DMT kinases. By contrast, ARD-mediated regulation by Ca2+/CaM does not emerge as a common theme across the entire DMT branch. Because it is unlikely that DMT kinases are less regulated than other kinases, a major challenge remains to identify and characterize an extended arsenal of combinatorial mechanisms of activity regulation. It remains of particular interest to see whether mechanical forces within the respective cellular compartments, paradigmatically explored for titin kinase, have a role in other members of these protein kinases. Ultimately, there is a need to explore the role of other and in part distant domains of the DMT kinases, beyond the ARD, as there is an increasing body of evidence for their involvement in activity regulation.

Little is also known about potential substrates of most of the DMT kinases. At least for those with an ARD next to the CD, the ARD may serve as a pseudosubstrate to facilitate novel substrate identification and characterization. Assuming that a pseudosubstrate model is applicable to true trans substrates, an observed tendency to form helical structures in the relevant part of the ARD suggests that some of the DMT kinase substrates are helical, at least in part, which deviates from available observations that generally display kinase substrates in an extended conformation. If this assumption is true, conformational constraints of substrates may play a more prominent role than anticipated, shifting the balance from sequence-driven to structure-driven substrate recognition by protein kinases.


KT is supported by an EMBL Interdisciplinary Postdoctoral (EIPOD) fellowship under Marie Curie Actions (COFUND).