Computational analysis of regulatory regions in human protein kinases

Abstract

Eukaryotic proteins often feature modular domain structures comprising globular domains that are connected by linker regions and intrinsically disordered regions that may contain important functional motifs. The intramolecular interactions of globular domains and nonglobular regions can play critical roles in different aspects of protein function. However, studying these interactions and their regulatory roles can be challenging due to the flexibility of nonglobular regions, the long insertions separating interacting modules, and the transient nature of some interactions. Obtaining the experimental structures of multiple domains and functional regions is more difficult than determining the structures of individual globular domains. High‐quality structural models generated by AlphaFold offer a unique opportunity to study intramolecular interactions in eukaryotic proteins. In this study, we systematically explored intramolecular interactions between human protein kinase domains (KDs) and potential regulatory regions, including globular domains, N‐ and C‐terminal tails, long insertions, and distal nonglobular regions. Our analysis identified intramolecular interactions between human KDs and 35 different types of globular domains, exhibiting a variety of interaction modes that could contribute to orthosteric or allosteric regulation of kinase activity. We also identified prevalent interactions between human KDs and their flanking regions (N‐ and C‐terminal tails). These interactions exhibit group‐specific characteristics and can vary within each specific kinase group. Although long‐range interactions between KDs and nonglobular regions are relatively rare, structural details of these interactions offer new insights into the regulation mechanisms of several kinases, such as HASPIN, MAPK7, MAPK15, and SIK1B.

1. INTRODUCTION

Modular domain structures are common in eukaryotic proteins (Han et al., 2007). The domain architecture of a protein plays a critical role in determining its biological function, regulating its activity, and facilitating interactions with other molecules (Cheng et al., 2014; Orengo et al., ¹⁹⁹⁷). Multiple domains in a protein can act independently or in concert through intramolecular interactions. Modularity of domain structures enables the evolution of novel functions through the combination and variation of domains, contributing to the remarkable diversity of eukaryotic proteins (Bhattacharyya et al., 2006; Patthy, ²⁰⁰³).

The structure of a protein, determined by its amino acid sequence, can be mainly classified into two categories: globular domains and nonglobular regions (Wright & Dyson, 1999). Globular domains are compactly folded into specific shapes and are often responsible for a protein's specific function, such as enzymatic activity. A soluble globular domain typically consists of a hydrophobic core and a hydrophilic surface, enabling it to interact with its environment. In contrast, nonglobular regions are often disordered, lack a defined structure, and can be found at the termini of a protein or in linkers/loops connecting globular domains (Perkins et al., 2010; Tompa & Fersht, ²⁰⁰⁹). These regions are often involved in intermolecular or intramolecular transient interactions (Perkins et al., 2010). Nonglobular regions play important roles in protein function, such as regulating protein–protein interactions or acting as flexible hinges. Various short linear motifs (Davey et al., 2015; Van Roey et al., ²⁰¹⁴) and molecular recognition features (Oldfield et al., 2005) are frequently found in nonglobular and disordered regions. The lack of structure in nonglobular regions allows for greater flexibility and transient interactions, which can be important for the protein to carry out its function. Residues in these regions can undergo posttranslational modifications that provide additional mechanisms for regulating their functions (Bah & Forman‐Kay, 2016).

The short length and variability of sequence motifs in nonglobular regions make their detection challenging using traditional sequence‐based methods. In addition, the transient nature of interactions and frequent dependence on modifications further complicate their structural characterization. However, recent advancements in deep learning, specifically AlphaFold, have transformed the field of structural biology by greatly expanding the availability of high‐quality protein structural models (Jumper et al., 2021; Tunyasuvunakool et al., ²⁰²¹). One advantage of AlphaFold is the ability to predict the structures of large multidomain proteins, which are difficult to obtain as a whole experimentally. This characteristic of AlphaFold structural models offers a unique opportunity to investigate the regulation of a protein's function through the intramolecular interactions of different regions (globular domains and nonglobular regions) (Perrakis & Sixma, 2021).

We aim to probe this aspect by studying the interdomain interactions involving the human protein kinases, which are a diverse group of enzymes that play crucial roles in various cellular processes such as signal transduction, cell proliferation, differentiation, and apoptosis (Manning et al., 2002). Protein kinases catalyze the transfer of phosphate groups from ATP to target proteins, leading to changes in their activity, localization, and interactions. This posttranslational modification is essential for regulating the activity of many proteins involved in crucial cellular functions.

Protein kinases are essential components of eukaryotic cell signaling pathways. They activate or inhibit specific signaling pathways by phosphorylating specific proteins. For example, the kinase AKT is a key regulator of the PI3K/AKT/mTOR signaling pathway, which controls cell growth, survival, and metabolism. Dysregulation of this pathway has been linked to various types of cancer (Osaki et al., 2004). Another example is the kinase MAPK, which regulates the MAPK/ERK signaling pathway involved in cell proliferation and differentiation. Dysregulation of this pathway has been implicated in several human diseases, including cancer, cardiovascular disease, and neurodegenerative disorders (Manning et al., 2002; Wang, ²⁰⁰⁷). Kinases also play a crucial role in the regulation of the cell cycle. Kinases such as cyclin‐dependent kinases (CDKs) and checkpoint kinases (CHKs) control various stages of the cell cycle, including DNA replication, mitosis, and cytokinesis (Bettencourt‐Dias et al., 2004). Dysregulation of these kinases can lead to abnormal cell division and genomic instability, which can contribute to cancer development. Targeting kinases involved in cancer with specific inhibitors has become a promising approach for the treatment of cancer (Roskoski, 2022).

Kinase activity is regulated by a variety of mechanisms, including posttranslational modifications, protein–protein interactions, and changes in subcellular localization (Nolen et al., 2004; Pellicena & Kuriyan, ²⁰⁰⁶; Wang & Cole, ²⁰¹⁴; Wortzel & Seger, ²⁰¹¹). These mechanisms allow for the tight control of kinase activity in response to specific cellular signals and environmental cues. Phosphorylation of specific residues in kinases, especially those in the activation loop, can activate kinase activity. For example, the kinase AKT is activated by phosphorylation of two residues within its activation loop (Alessi et al., 1996), while the kinase ERK is activated by dual phosphorylation of threonine and tyrosine residues within its activation loop (Ramos, 2008). Kinases interact with various regulatory proteins, including scaffolding proteins, adaptors, and activators. These interactions can either activate or inhibit kinase activity by modulating substrate specificity, subcellular localization, or conformational changes. For example, the kinase MEK is activated by its interaction with the scaffolding protein KSR, which enhances its phosphorylation of the downstream kinase ERK (Peyssonnaux & Eychène, 2001). The kinase CHK2 is activated by its interaction with the DNA damage response protein BRCA1, which leads to its phosphorylation and activation (Stolz et al., 2010). Changes in subcellular localization also regulate kinase activity. Many kinases, such as MAP kinases, shuttle between different cellular compartments in response to specific signals, allowing for the temporal and spatial regulation of kinase activity (Harding et al., 2005).

Approximately 500 protein kinases have been identified in the human proteome and classified into several groups (Manning et al., 2002; Modi & Dunbrack, ²⁰¹⁹). Many human protein kinases are modular enzymes consisting of both conserved globular domains and nonglobular regions with various types of functional motifs. Although the kinase catalytic domain is typically 250–270 residues long, about half of the human protein kinases are over 650 residues, and nearly a quarter of human kinases are over 1000 residues. Globular domains and functional regions are critical for regulating the structural integrity, substrate recognition, and catalytic activity of human kinases.

The regulatory domains or regions in kinases are diverse, and they modulate kinase activity through various mechanisms. These domains often regulate kinase activity by altering the conformation of the kinase domain (KD) or by interacting with other regulatory proteins. SH2 and SH3 domains, for instance, mediate protein–protein interactions and signal transduction in Src family kinases (SFKs) (Koch et al., 1991). Other domains, such as the pleckstrin homology (PH) domain (Rebecchi & Scarlata, 1998) and kinase‐associated domain 1 (KA1) (Tassan & Le Goff, 2004), can bind to specific lipids and control kinase activity by recruiting kinases to specific subcellular locations.

Autoinhibition is a common mechanism used by protein kinases to regulate their activity (Bayliss et al., 2015). In this mechanism, the kinase adopts a closed, inactive conformation that prevents ATP or protein substrates from accessing the catalytic site. Autoinhibition is often achieved through intramolecular interactions between regulatory regions and catalytic domains of kinases. A regulatory region, which could be a globular domain or a nonglobular sequence motif, may directly bind to the active site of the kinase, as in the case of orthosteric inhibition (Nussinov et al., 2023). Alternatively, allosteric inhibition occurs when a regulatory region binds to a location (allosteric site) not directly involved in catalysis, and such intramolecular interactions lock the KD in an inactive conformation (Mingione et al., 2023). Here, we systematically probed potential regulation regions in human protein kinases by examining the intramolecular interactions with KDs. Studies of regulatory regions could offer a detailed understanding of autoinhibition mechanisms and may provide opportunities for developing selective kinase inhibitors (Hanold et al., 2017).

2. RESULTS

2.1. Structural elements in human KDs

The catalytic domains of human kinases are referred to as the KDs. The structural alignment of 496 human KDs was used to define the structural cores of the KDs and their structural elements (Modi & Dunbrack, 2019) (see Section 4). We introduced two new structural elements: PL (P‐loop) and CAH (Catalytic loop‐Activation loop Hairpin) (see Section 4) in addition to previously defined KD structural elements. CAH is located in between CL (catalytic loop) and ALN (activation loop, N‐terminal part) and is part of the common docking site (Tanoue et al., 2000; Zeke et al., ²⁰¹⁶) that are often used by MAP kinases for substrate binding. In total, 18 structural elements were defined for the KDs of human kinases, including five β‐strands (B1–B5) and seven α‐helices (HC–HI). They are in this sequential order: B1, PL, B2, B3, HC, B4, B5, HD, HE, CL, CAH, ALN, ALC, HF, FL (F‐loop), HG, HH, and HI. An example of a KD with labeled structural elements is shown in Figure 1a. Six structural elements are mostly relevant to the kinase activity (PL, HC, CL, CAH, ALN, and ALC) as they harbor most of the catalytic and ATP‐binding residues. The activation loop is divided into an N‐terminal element (ALN) and a C‐terminal element (ALC) (Modi & Dunbrack, 2019). The KD N‐terminal lobe (N‐lobe) mainly consists of the β‐strands and HC. The other α‐helices mainly constitute the C‐terminal lobe (C‐lobe). Human KDs are classified into 10 groups (AGC, CAMK, CK1, CMGC, NEK, OTHER, RGC, STE, TKL, and TYR) based on phylogenetic and structural analysis (their numbers shown in Figure 1b) (Modi & Dunbrack, 2019).

FIGURE 1.

(a) An example of the AlphaFold model of KD from human AURKA (Aurora kinase A). The structural elements are labeled. N‐tail and C‐tail are colored gray and light pink, respectively. (b) Distribution of the number of human KDs in KD groups. (c) Distribution of the number of residue contacts to KD structural elements from regions outside KD or from long insertions in KD (long_insertion).

2.2. Identification of globular domains and nonglobular regions interacting with KDs

We defined positions outside a KD in a human protein kinase as being either in globular domains or four types of nonglobular regions based on the distance to the KD: N‐tail, C‐tail, N‐long_range, and C‐long_range (see Section 4). The N‐tails and C‐tails correspond to KD‐flanking nonglobular regions of up to 80 residues, while N‐long_range and C‐long_range positions are those more than 80 residues away from the KD. Some KDs have long insertions between neighboring structural elements. These long insertion regions (defined in the category long_insertion) can also form interactions with KD structural elements.

We used two criteria to define intramolecular contacts between residues in the KD structural elements and residues outside the KD or in long insertions of the KD in AlphaFold models: distance between residue atoms and predicted aligned error (PAE) values reported by AlphaFold (see Section 4). For the 496 human KDs, we identified a total of about 26,000 such residue‐residue contacts (Figure 1c). The N‐tail and C‐tail (within 80 residues of KDs) nonglobular regions contribute the most numbers of contacts. Globular domains and long insertions inside KDs each contribute to about 2000 contacts, about one‐fifth of those from N‐tail or C‐tail. Distal nonglobular regions (N‐long_range and C‐long_range) contribute a small fraction of the total number of contacts (Figure 1c).

2.3. Sequence and structure properties of residues that interact with KDs

We studied several sequences and structural properties of residues that form intramolecular interactions with the KDs, such as secondary structures, pLDDT scores, disorder propensity, and sequence conservation (see Section 4). We compared the fraction of a certain property in residues interacting with functional elements PL, HC, CL, ALN, ALC, and CAH to the background fraction calculated for positions outside KDs (see Section 4). These fractions of the properties are plotted in Figure 2.

FIGURE 2.

Sequence and structural properties of residues in contact with KD structural elements. (a) Secondary structure types. Each bar is the fraction of a certain secondary structure type (helix, strand, and coil) among all residues that make contact with the structural element (HC, PL, CL, ALN, CAH). The gray bar (BG) is the background fraction of the secondary structure type in all residues outside the KD. (b) The pLDDT scores fractions. Very high: pLDDT ≥ 90, confident: 90 > pLDDT ≥ 70, low: 70 > pLDDT ≥ 50, very low: pLDDT < 50. (c) Order/disorder prediction fraction by IUPred2A. (d) Sequence conservation fractions. High conservation: AL2CO scores 8–9, medium conservation: AL2CO score 6–7, low conservation: AL2CO score 0–5.

For secondary structure types, we observed a depletion of β‐strands and slight preferences of coils in residues contacting the functional elements of KDs, especially for CAH (Figure 2a). The per‐residue pLDDT scores reflect the accuracy of AlphaFold model predictions. Residues contacting CAH have the highest fraction in the “very high” category of pLDDT scores (pLDDT ≥ 90). On the other hand, residues contacting PL and CL have a relatively low fraction in the “very high” category of pLDDT scores, suggesting that some of these residues may have more structural flexibility. This is consistent with their relatively high disorder fraction (Figure 2c). Residues contacting the KD functional elements tend to be slightly more conserved than the background, especially for those interacting with CL (Figure 2d).

2.4. Characterization of globular domains interacting with KDs

Many human protein kinases possess multiple domains with various domain contents. We identified 35 types of globular domains that make intramolecular interactions with KDs in human protein kinases (Figure 3), suggesting that these domains could regulate KD activity directly. The most abundant KD‐interacting domains are SH2, SH3, KD, UBA, KA1, PH, and the β‐propeller domain. Some of these globular domains are mainly found in a specific group of kinases. For example, the SH2 domain and SH3 domain are mostly found in a subset of kinases in the TYR (tyrosine kinases) group. Shared domain composition and domain architecture usually suggest a close evolutionary relationship between these proteins. Conversely, some domains are found in different kinase groups. For example, PH domains interact with KDs from several kinase groups, such as AKT1/2/3 and PDPK1 in the AGC group, PDKD1/2/3 in the CAMK group, MAP3K6 and MAP3K15 from the STE group, and JAK1/2/3 and TYK2 in the TYR group. Such a distribution suggests independent acquisition of PH domains in these kinases, which could facilitate their association with the membrane compartments by the interactions of PH domains and lipids. Most KD‐interacting globular domains are restricted to one or several closely related kinases with specific functions, such as the DEATH domain in IRAK2/4 kinases, the C2 domain in PRKCD and PRKCE, and the SAM‐like domain in EPHA1 and EPHA6.

FIGURE 3.

Globular domains interacting with KDs. Each bar represents the number of KDs that shows the interaction of a globular domain with KD. The list of KD‐interacting domains found for two or one KDs is shown in boxes.

2.5. The most abundant kinase‐interacting domains—SH2 and SH3

AlphaFold models revealed 27 KDs making intramolecular interactions with SH2 domains and 21 KDs making such interactions with SH3 domains. All these SH2 domains and the majority of the SH3 domains were found in the TYR group of tyrosine kinases. Only two SH3 domains involved in KD interactions were found in the TKL kinase group (MAP3K11 and MAP3K9).

A set of non‐receptor tyrosine kinases possess the Src module (SH3–SH2 kinase) and are referred to as the SFKs (Shah et al., 2018). They played important roles in signaling pathways related to cell growth, differentiation, migration, and survival. The SH3 and SH2 domains in SFKs mainly interact with the KD at locations not containing the active site, including structural elements from the β‐strands of the N‐lobe, the CAH element, and the α‐helices of the C‐lobe (Figure 4a). They regulate kinase activity mainly through allosteric mechanisms (Boggon & Eck, 2004; Ingley, ²⁰⁰⁸). The interdomain interactions in SFKs play autoinhibitory roles by keeping the KD in the inactive conformation.

FIGURE 4.

Structures showing interactions of SH2 and SH3 domains with KDs. (a) TYR_SRC (each protein is labeled by the group name [TYR] followed by the protein name [SRC]). (b) TYR_CSK. (c) TYR_SYK. (d) TYR_FES. (e) TKL_MAP3K9. (f) TYR_TNK1. Locations of SH2 and SH3 domains are labeled. The structural element coloring scheme is the same as in Figure 1a. The N‐ and C‐terminal parts outside the KD are shown in gray and light pink, respectively. The interactions between the KD structural element and residues outside kinase domains are shown in lines that have the same color as the structural element.

CSK is a SH3‐ and SH2‐containing kinase that regulates the activities of SFKs (Okada, 2012). CSK is not a member of the SFKs despite similar domain contents. The spatial orientations of SH3, SH2, and KDs are different in CSK (Figure 4b) compared to the SFKs. The CSK SH3 domain occupies a similar location where the SH3 domains of SFKs reside. However, unlike the SH2 domain in SFKs that interacts with the kinase C‐lobe, the CSK SH2 domain occupies an entirely different location and interacts with the N‐lobe in the AlphaFold model, which has also been observed in the experimental structure of CSK (Ogawa et al., 2002). The CSK SH2 domain, the SH2–SH3 linker, and the SH3–KD linker make extensive contacts with HC in the AlphaFold model. Such tight interactions have been linked to the active state of the kinase, and the release of the CSK SH2 domain is associated with the inactive state of the KD (Ogawa et al., 2002).

The non‐receptor tyrosine kinase SYK (Mócsai et al., 2010) contains two SH2 domains and no SH3 domain. One of the SYK SH2 domains interacts with the C‐lobe of the KD, like the SH2 domain in the SFKs. The other SYK SH2 domain does not interact with the KD. Instead, a small three‐helical bundle domain in between the two SH2 domains (called interdomain A or IA; Figure 4c) and the SH2–KD linker (called interdomain B) interact with the kinase N‐lobe, CAH, and C‐lobe (Mócsai et al., 2010). Like SFKs, the interdomain interactions of SYK do not involve the kinase active site. Thus, they contribute to the autoinhibition of the kinase activity through allosteric regulations (Mócsai et al., 2010).

Variations of domain architecture and interaction modes were also observed in other SH2‐ or SH3‐containing kinases. The tyrosine kinase FES has a single SH2 domain that interacts with the kinase N‐lobe (Figure 4d). Such a tight interaction involving HC has been shown to maintain the KD in its active conformation (Filippakopoulos et al., 2008). The TKL group kinase MAP3K9 has a single SH3 domain that interacts with HC, CL, and ALN of the KD (Figure 4e). SH3 domains were also found in three other MAP3 kinases (MAP3K10, MAP3K11, and MAP3K21) closely related to MAP3K9. The SH3–KD interaction in these kinases has been implicated in having an autoinhibitory role and could mediate protein–protein interactions (Zhang & Gallo, 2001). The tyrosine kinase TNK1 has a C‐terminal SH3 domain that interacts with the C‐lobe of the KD (Figure 4f) in the AlphaFold model. The function of this interdomain interaction remains to be determined for this kinase with a poorly understood function (Chan et al., 2021).

2.6. Kinase‐interacting domains with diverse evolutionary origins—PH, β‐propeller, and ARM repeats

AlphaFold models of 14 human protein kinases show intramolecular interactions between the KD and PH domain, a popular domain in the human proteome with the ability to bind phospholipids (Lemmon, 2007). These kinases are from four different groups—AGC, CAMK, STE, and TYR, suggesting that their PH domains were independently acquired. Some of these PH domains were found to have autoinhibitory roles for the kinase activity (Levina et al., 2022), for example, in AKT1 (protein kinase B) and PRKD1 (protein kinase D).

The interaction modes of the PH domains and KDs are quite different in different kinases (Figure 5). For AKT1/2/3 of the AGC group (Figure 5a), the PH domain mainly interacts with ALC and HG of the C‐lobe. For PRKD1 of the CAMK group (Figure 5b) and MAP3K6 (Figure 5c) of the STE group, PH domain interacts with the kinase N‐lobe from different directions: PH domain in PRKD1 interacts with the β‐sheet and PH domain in MAP3K6 mainly interacts with HC.

FIGURE 5.

Structures showing interactions of PH domains with KDs. (a) AGC_AKT1 (each protein is labeled by the group name [AGC] followed by the protein name [AKT1]). (b) CAMK_PRKD1. (c) STE_MAP3K6. (d) TYR_JAK1. (e) AGC_PDPK1. Locations of globular domains, including PH domains, are labeled. The structural element coloring scheme is the same as in Figure 1a. The N‐ and C‐terminal parts outside the KD are shown in gray and light pink, respectively. The interactions between the KD structural element and residues outside kinase domains are shown in lines that have the same color as the structural element.

PH domains were also found to interact with the pseudokinase domains in four JAK family kinases, JAK1/2/3 and TYK2 of the TYR group. For these proteins, a PH domain and a SH2 domain interact with the C‐lobe of the pseudokinase domain (Figure 5d). These interactions could be important to regulate the activity of JAK1/2/3 and TYK2 through the pseudokinase‐kinase domain interactions (Toms et al., 2013). We also found a PH domain C‐terminal to the KD interacting with the C‐lobe of PDPK1KD (Figure 5e).

The human proteome also has other kinases with PH domains. For example, a PH domain resides N‐terminal to the SH3–SH2 kinase module (SRC module) of several tyrosine kinases such as TEC and BTK and C‐terminal to the KD in some AGC group kinases such as ROCK1 and GRK2. However, interdomain PH–KD interactions were not identified in AlphaFold models for these proteins, suggesting that the PH domain in them may not have a direct role in regulating the kinase activity. Alternatively, some PH–KD interactions may not be detected by our approach as the interactions may be weak and transient.

Like the PH domain, intramolecular interactions between β‐propeller domains and the KDs have been detected in several kinase groups, including NEK (NEK8 and NEK9), STE (MAP4K1 and MAP4K5), and TKL (LRRK1 and LRRK2), suggesting independent origins of β‐propeller domains in these proteins. LRRK1 has a modular domain structure and has the greatest number of globular domains that interact with the KD, including the leucine‐rich repeat (LRR) domain, the ROC domain (GTP binding), two helix‐turn‐helix (HTH) domains, a TBP‐like domain, and a C‐terminal β‐propeller domain (Figure 6a). The domain orientations in the AlphaFold model of LRRK1 are highly consistent with the experimental structure of its close homolog, LRRK2 (Myasnikov et al., 2021). The ROC domain, HTH domains, and the TBP‐like domain in LRRK1 together adopt a compact structure that forms a large interface with the KD involving the N‐lobe β‐strands, HC, CL, ALN, ALC, and C‐lobe α‐helices. The C‐terminal part of the LRR repeats also has interactions with the C‐lobe. The activity of LRRK1 is likely regulated by these interactions through autoinhibitory mechanisms. The β‐propeller domain and a C‐terminal α‐helix after it interact with the LRRK1 KD at C‐lobe α‐helices and CAH (Figure 6a).

FIGURE 6.

Structures showing interactions of β‐propeller domains and ARM repeats with KDs. (a) TKL_LRRK1 (each protein is labeled by the group name [TKL] followed by the protein name [LRRK1]). (b) STE_MAP4K5. (c) NEK_NEK9. (d) OTHER_SCYL1. (e) NEK_NEK10. Locations of globular domains are labeled. The structural element coloring scheme is the same as in Figure 1a. The N‐ and C‐terminal parts outside the KD are shown in gray and light pink, respectively. The interactions between the KD structural element and residues outside kinase domains are shown in lines that have the same color as the structural element.

Orientation of the β‐propeller domain relative to the KD varies in different kinases (Figure 6). The MAP4K5 β‐propeller domain interacts with the KD ALC (Figure 6b), suggesting that such an interaction may inhibit the kinase activity by blocking the substrate‐binding site. The β‐propeller domain of NEK9 mainly interacts with KD N‐lobe and CAH (Figure 6c). In addition, the first and last β‐strands of the NEK9 β‐propeller domain are elongated and form an extended β‐sheet with the kinase N‐lobe β‐strands (Figure 6c).

Several helical repeat domains were found to interact with KDs. ARM repeats were found to interact with the KDs of SCYL1/2/3 (OTHER group) (Figure 6d), NEK10 (NEK group) (Figure 6e), and PIK3R4 (OTHER group). Ankyrin repeats were found in TEX14 (OTHER group) and ILK (TKL group). While NEK10 is an active kinase, the other proteins containing these repeats (SCYL1/2/3, PIK3R4, TEX14, and ILK) are all pseudokinases with an inactive KD, suggesting that these proteins with helical repeats could serve as scaffolds for protein–protein interactions.

2.7. Other globular domains that interact with the KDs

A number of other globular domains interact with the KDs. Eight human protein kinases possess two KDs that interact with each other: RPS6KA1/2/3/6 (Magnuson et al., 2012) with an N‐terminal KD of the AGC group and a C‐terminal KD from the CAMK group and the four JAK family kinases (JAK1/2/3 and TYK2) with both KDs belonging to the tyrosine kinase group. The N‐terminal KD of JAK1/2/3 and TYK2 are pseudokinase domains with degenerated active sites. These pseudokinase domains are important for autoinhibition of the C‐terminal active KDs by allosteric interactions (Toms et al., 2013). Compounds that bind pseudokinase domains have been developed to inhibit JAK kinase activity by allosteric mechanisms (Lu et al., 2020).

A set of CAMK group kinases possess a UBA domain and/or KA1 domain at its C‐terminal end. For example, the MARK1 UBA domain with a three‐helical bundle fold interacts with the N‐lobe of the KD in the AlphaFold model, consistent with its experimental structure (Marx et al., 2006). KA1 domain interacts with CL, PL, and CAH in MARK1 (Figure 7a). Like PH domain, some KD‐interacting domains are lipid‐binding domains, such as PX, KA1, and C1. They could function to recruit the kinases to various membrane locations or serve as autoinhibition modules (Chu et al., 2020; Emptage et al., ²⁰¹⁷; Leonard & Hurley, ²⁰¹¹). Most of the KD‐interacting domains with low occurrence frequencies are restricted to a specific kinase or several closely related kinases. For example, the WNK autoinhibitory domain (WIN_autoinhib, Figure 7b) (Moon et al., 2013) and CRIB domain (Tu & Wigler, 1999) (Figure 7c) are found exclusively in WNK1/2/3/4 and PAK1/2/3, respectively. The IKKB_dimerization domain and the Ubiquitin‐like domain only occur in four closely related inhibitors of NFκB kinases (CHUK, IKBKB, IKBKE, and TBK1) (Xu et al., 2011) (Figure 7d).

FIGURE 7.

Structures showing interactions of globular domains with KDs. (a) CAMK_MARK1 (each protein is labeled by the group name [CAMK] followed by the protein name [MARK1]). (b) OTHER_WNK1. (c) STE_PAK1. (d) OTHER_IKBKB. Locations of globular domains are labeled. The structural element coloring scheme is the same as in Figure 1a. The N‐ and C‐terminal parts outside the KD are shown in gray and light pink, respectively. The interactions between the KD structural element and residues outside kinase domains are shown in lines that have the same color as the structural element.

2.8. KD interactions from KD‐flanking regions

The kinase flanking regions (N‐ and C‐tails) are the regions surrounding the KD that do not directly participate in kinase activity but could play important regulatory roles. The functions of protein kinase flanking regions include stabilization of the KD, substrate recognition, protein localization, regulation of kinase activity, and protein–protein interactions. N‐tails and C‐tails often surround the KD and can adopt various conformations, including irregular loops, α‐helices, and β‐strands. The length, conformation, and interaction sites of these regulatory regions can vary significantly among different kinases. Due to their spatial proximity, the N‐tails and C‐tails mainly interact with the KD N‐lobes and C‐lobes, respectively. However, some of these tails may travel across the KD to interact with the opposite lobe.

Some N‐tails and C‐tails exhibit group‐specific characteristics. A commonly observed structural feature of the AGC group kinases lies in their C‐tails that form extensive interactions with KDs at various locations and play essential roles in the regulation of kinase activity (Kannan et al., 2007). A typical AGC C‐tail starts from the C‐lobe and interacts sequentially with HD and HE of the C‐lobe, CAH, the β‐strands of the N‐lobe, PL, and finally HC of the N‐lobe (Figure 8a), forming a complex network of interactions. Although this C‐tail pattern is observed in most AGC group kinases, some variations were observed. For example, the C‐tail of the AGC group kinase RSKR is shorter and does not interact with its HC, while its N‐tail interacts with its HC instead. PDPK1 and PDPK2P of the AGC group also lack many of the C‐tail to N‐lobe interactions.

FIGURE 8.

Structures showing interactions of nonglobular regions with KDs. (a) AGC_PRKCA (each protein is labeled by the group name [AGC] followed by the protein name [PRKCA]). (b) CAMK_TTN. (c) CMGC_SRPK1. (d) AGC_MASTL. The structural element coloring scheme is the same as in Figure 1a. The N‐ and C‐terminal parts outside the KD are shown in gray and light pink, respectively. The interactions between the KD structural element and residues outside kinase domains are shown in lines that have the same color as the structural element.

N‐tail and C‐tail structures and their interactions can vary greatly in different kinase groups. For example, while AGC group kinases commonly have long C‐tails, the C‐tails of many KDs from the TYR group and the TKL group are quite short. Long C‐tails were also observed in some CAMK group kinases, such as TTN, that form extensive interactions with various KD structural elements (Figure 8b). Compared to the AGC group C‐tail, the C‐tail of TTN is oriented differently. It contains a small α‐helix inserted in the ATP‐binding cleft between PL and CL and forms β‐strand interaction with the ALC element (Figure 8b). Thus, TTN C‐tail can serve as an autoinhibitory region that directly blocks both ATP and substrate binding.

N‐tail to HC interactions are common for the kinase groups CAMK, STE, TKL, and TYR, but are largely missing in the AGC and CMGC groups. Many kinases in the STE group have an N‐tail that interacts with HC and a C‐tail α‐helix that interacts with HD, HE, and CAH (one example in Figure 7c). Most NEK group kinases (except NEK6 and NEK7) have one or more α‐helices in the C‐tail that interact with HE, HI, and CAH (two examples in Figure 6c,e). C‐tails of KDs in the CK1 group mainly form interactions with CAH, HD, and HE.

Examination of N‐tails and C‐tails also revealed distinctive patterns in certain members of a kinase group. For instance, N‐tails of SFKs in the TYR group interact with both the KD and the SH3 domain, providing the docking site for the SH3 domain. Conversely, N‐tails of receptor tyrosine kinases (in the juxtamembrane regions) are oriented differently compared to those in SFKs and mainly interact with HC. These N‐tail to HC interactions may have an autoinhibitory role in some receptor tyrosine kinases, while in other members, they can be essential for the kinase activity (Li & Hristova, 2010). The C‐tails of the MAP kinases of the CMGC group possess an α‐helix that interacts with HC (one example in Figure 9d), which is absent in most of the other members of the CMGC group kinases such as the CDK kinases. C‐tails of the CAMK group kinases with the UBA domains adopt a loop conformation and interact mainly with HE and HI (one example in Figure 7a), while C‐tails of the CAMK group kinases without UBA domains often have an α‐helix that interacts with HD and HF (one example in Figure 8b).

FIGURE 9.

Structures showing intramolecular long‐range interactions with KDs. (a) AGC_PRKG1 (each protein is labeled by the group name [AGC] followed by the protein name [PRKG1]). (b) AGC_PRKCQ. (c) OTHER_HASPIN. (d) CMGC_MAPK7. (e) CMGC_MAPK15. (f) CAMK_SIK1B. Locations of globular domains are labeled. The structural element coloring scheme is the same as in Figure 1a. The N‐ and C‐terminal parts outside the KD are shown in gray and light pink, respectively. The interactions between the KD structural element and residues outside kinase domains are shown in lines that have the same color as the structural element. N‐terminal distal segments with long‐range interactions are shown in black.

2.9. KD interactions from long insertions inside the KDs

We identified about 1800 interactions between long insertion regions inside the KDs (long_insertion, Figure 1c) and KD structural elements. One common structural feature of the CMGC group kinases is a unique insertion region (CMGC insert, an example shown in Figure 8c) that is usually over 20 amino acids and occurs in between HG and HH (Agnew et al., 2019; Kannan & Neuwald, ²⁰⁰⁴). This feature is not found in most members of other kinase groups (CDC7 of the OTHER group and IRAK1/2 of the TKL group are a few exceptions). The CMGC insert provides structural stability for the C‐lobe and is part of the substrate‐binding site (Kannan & Neuwald, 2004).

Other long insertions are specific for certain kinases. For instance, the SRPK1 kinase (CMGC group) has a long insertion in CAH (Figure 8c). Experimental studies showed that the long insertion region of the SRPK1 kinase can affect the dynamics of the KD (Plocinik et al., 2011) and contribute to the binding pocket for selective SRPK1 inhibitors (Batson et al., 2017). This long insertion contains two long α‐helices that interact extensively with the N‐lobe β‐sheet in the AlphaFold model. They may contribute to the stability of the KD or provide allosteric regulation of the kinase activity. The kinase MASTL (AGC group) has the longest insertion region (over 500 residues) among all KDs in human kinases. This long insertion, occurring between ALN and ALC, has been shown to be essential for substrate specificity and allosterically regulates the catalysis reaction (Hermida et al., 2020). One part of the insertion (residues 201–221) interacts with HC and B4 (Figure 8d) in the AlphaFold model and thus could play a direct role in regulating the kinase activity of MASTL.

2.10. Potential regulatory regions distal to KDs

Long‐range interactions between distal regions and KDs are relatively rare (Figure 1c). Some of these long‐range interactions have been experimentally shown to have regulatory roles. For example, a distal autoinhibitory sequence (residues 51–70) has been experimentally characterized as a pseudosubstrate in the cGMP‐dependent kinase PRKG1 of the AGC group (Hofmann et al., 2006). It lies N‐terminal to the two CNMP‐binding domains (both adopting a jelly roll fold) and is about 300 residues away from the N‐terminus of the PRKG1 KD. The AlphaFold model revealed that this segment interacts with CL, ALN, and ALC (Figure 9a), consistent with its function as a pseudosubstrate (Hofmann et al., 2006). AlphaFold model also helps identify a similar N‐terminal distal segment (residues 140–153) in PRKCQ (protein kinase C, theta type) (Hayashi & Altman, 2007) that interacts with CL, ALN, and ALC and serves as a pseudosubstrate for autoinhibitory regulation (Figure 9b). This segment in PRKCQ lies in between the C2 domain and the C1 domain, consistent with experimental results (Steinberg, 2008).

An N‐terminal segment of HASPIN (residues 376–408) forms long‐range interactions with HASPIN KD at several structural elements. This region adopts an extended helix‐strand‐helix conformation (Figure 9c). The N‐terminal α‐helix of this region interacts with the active site structural elements PL, CL, and ALC, while the β‐strand of this region forms a β‐sheet with ALC. Part of this region has been identified as an autoinhibitory segment named HASPIN basic inhibitory segment (HBIS) (Ghenoiu et al., 2013). While the experimental structure of the HASPIN KD has been determined (Eswaran et al., 2009), the structure of its complex with this autoinhibitory segment has not been solved. AlphaFold model provides structural details of the interaction between HBIS and KD and suggests an orthosteric mechanism of autoinhibition.

Another example of long‐range interactions was observed in the MAP kinase MAPK7 (ERK5) (Nithianandarajah‐Jones et al., 2012). MAPK7 is a CMGC group kinase with the CMGC insert and a long C‐tail. The C‐tail has a helical interaction with HC in the AlphaFold model (Figure 9d). The AlphaFold model also revealed a distal C‐terminal segment (residues 736–747, about 400 residues from the KD C‐terminus) that is inserted in the ATP‐binding cleft between PL and CL (Figure 9d). Thus, this distal segment could serve as an autoinhibitory region for the KD domain by blocking ATP binding. This long‐range intramolecular interaction is consistent with a previous experiment that showed the C‐terminus of MAPK7 regulates its activity through autoinhibition, as deletion of the C‐terminus led to increased kinase activity (Buschbeck & Ullrich, 2005). The structure of this distal segment has not been solved experimentally. This segment exhibits high sequence conservation and offers mechanistic insight into the regulation of MAPK7 by autoinhibition. We also identified long‐range interactions for MAPK15 (Deniz et al., 2023) (residues 453–479, CMGC group, Figure 9e) and SIK1B (Darling & Cohen, 2021) (residues 564–608, CAMK group, Figure 9f). Like the one in MAPK7, the distal segments of MAPK15 and SIK1B interact with PL and CL and thus could provide autoinhibition by blocking ATP binding.

3. DISCUSSION

Understanding the domain architecture of a protein and the intramolecular interplay of its domains and regions is essential for deciphering its biological function and could provide directions for developing strategies to manipulate its function for therapeutic purposes. In our study, we used high‐quality structural models generated by AlphaFold to explore intramolecular interactions between KDs and regions outside of the KD, including globular domains and nonglobular regions. We found interactions from a set of globular domains and prevalent interactions from the flanking regions (N‐tails and C‐tails) of KDs. Many of these regions have been experimentally characterized to have regulatory roles (Abe et al., 1999; Kannan et al., ²⁰⁰⁷; Kovacs et al., ²⁰¹⁵; Taylor et al., ²⁰²²). The relative orientations of the interacting globular domains in a number of AlphaFold models of human kinases were also supported by structural studies (Boggon & Eck, 2004; Lapierre et al., ²⁰¹⁶; Marx et al., ²⁰⁰⁶; Myasnikov et al., ²⁰²¹; Ogawa et al., ²⁰⁰²). AlphaFold models of the interactions between KDs and distal regions offer new insights into the regulation of kinases such as HASPIN, MAPK7, MAPK15, and SIK1B, where experimental structures are not available. However, there are some limitations to our approach. Regulatory regions in disordered parts of the protein may not have strong signals from multiple sequence alignments, from which AlphaFold models explore the coevolution information. The tendency to have low‐quality AlphaFold model scores (pLDDT) in nonglobular and intrinsically disordered regions (David et al., 2022) could hinder the detection of their intramolecular interactions. This could partially explain the low fraction of long‐range intramolecular interactions in our dataset. For example, all nine human protein kinase C proteins (PRKCA/B/D/E/G/H/I/Q/Z) have a distal pseudosubstrate region (Newton, 2018; Steinberg, ²⁰⁰⁸). However, the interactions between the pseudosubstrate region and the KD were only detected in one AlphaFold model (PRKCQ).

We identified numerous interactions with functional elements (HC, PL, CL, ALN, and ALC) in our analysis. Interactions at these orthosteric sites can have a direct impact on kinase activity, for example, autoinhibition by blocking the ATP‐binding site between PL and CL, or the use of pseudosubstrates to obstruct substrate access. These inhibitory effects can be directly inferred from the structural models. However, the effects of the interactions outside functional elements could be more challenging to interpret solely based on AlphaFold models (Nussinov et al., 2022, 2023). Experimental studies have shown that many of these interactions allow for allosteric regulations of kinase activity. Interactions at allosteric sites can stabilize the kinase structure in either the active or inactive conformation, resulting in kinase activation or inhibition. The intramolecular interactions detected in our study may also have roles in other aspects of the kinase function, such as providing sites for protein–protein interactions and facilitating subcellular localization.

A small fraction of KD‐interacting residues likely falls into the category of disordered residues, as predicted by the IUPred2A program (Figure 2c). Interdomain contacts formed between these predicted disordered residues and residues from KDs display relatively higher PAE values compared to predicted ordered residues (Figure S2). For example, about 68% of KD‐interacting residues predicted to be disordered have a PAE value less than or equal to 3, as compared to about 80% for KD‐interacting residues predicted to be ordered. As residues in disordered regions are flexible and could adopt a variety of conformations, the conformation of these residues in the AlphaFold model and their interactions with the KDs should be interpreted with caution (Ruff & Pappu, 2021).

To fully understand the regulatory mechanisms, including autoinhibition, it is important to consider conformational changes and the dynamic properties of the proteins (Grant et al., 2010; Trudeau et al., ²⁰¹³; Wright & Dyson, ²⁰¹⁵; Xie et al., ²⁰²⁰). However, in our study, we used a single AlphaFold model for each kinase, which is limited in its ability to capture all possible conformational changes and competing interactions between multiple regulatory regions. In addition, posttranslational modifications, such as phosphorylation, play critical roles in the regulation of kinase activity, yet these modifications are not included in current AlphaFold models. As a result, our approach may have missed some regulatory regions that rely on posttranslational modifications. Future improvements to AlphaFold models, such as the consideration of protein dynamics (Amatya et al., 2019; Ozkirimli et al., ²⁰⁰⁸) and protein modifications (Bagdonas et al., 2021), could help overcome these limitations and enhance our understanding of kinase regulation.

4. MATERIALS AND METHODS

4.1. Defining human KDs and their structural elements

The structural alignment of 497 human KDs was used to define the structural cores of the KDs and their structural elements (http://dunbrack.fccc.edu/kincore/alignment) (Modi & Dunbrack, 2019). We removed one KD (PLK5) that is a fragment. Compared to the previously defined structural elements from this alignment (Modi & Dunbrack, 2019), we separated out the ATP‐binding P‐loop (motif: GXGXFG) as an additional structural element. It consists of the C‐terminus of B1 (β‐strand 1) and N‐terminus of B2 and is named PL (P‐Loop). In addition, we defined the loop region (usually a β‐hairpin) connecting CL (catalytic loop) and ALN (activation loop, N‐terminal part) as a new structural element as it lies on the opposite side of the active site and usually makes different intramolecular contacts compared to other residues in CL and ALN. This new structural element is termed CAH (Catalytic loop‐Activation loop Hairpin). The 18 structural elements of KDs include five β‐strands (B1–B5), seven α‐helices (HC, HD, HE, HF, HG, HH, and HI), PL, CL, CAH, ALN, ALC (activation loop, C‐terminal part), and FL (F‐loop, which occurs in between HF and HG). Residue ranges of structural elements of human KDs are available in Table S1.

4.2. Identification of globular domains and nonglobular regions interacting with KDs

We downloaded AlphaFold models of human protein kinases from the online AlphaFold database (https://alphafold.ebi.ac.uk). PAE values of these proteins in the format of matrices were also downloaded. AlphaFold models of long proteins with more than 2700 residues were available as segments of 1400 residues, while PAE files for them were not available from the online AlphaFold database. For these long kinases (KALRN, OBSCN, SPEG, TRIO, and TTN), all from the CAMK group, we obtained the PAE values by modeling a manually selected segment containing the KD using locally installed AlphaFold (downloaded from https://github.com/deepmind/alphafold). The multiple sequence alignments used for modeling these proteins have been previously described (Pei et al., 2022).

We assigned the globular domains in AlphaFold models of human kinases by the DPAM program (Schaeffer et al., 2023; Zhang et al., ²⁰²³). Positions outside globular domains were defined as nonglobular regions. In nonglobular regions, we defined residues outside KDs and within 80 residues from the N‐terminus as the N‐terminal tail (N‐tail) and residues outside KDs and within 80 residues from the C‐terminus as the C‐terminal tail (C‐tail). Residue contacts involving residues beyond the 80 residues from the N‐ and C‐termini of the KD are defined as N‐terminal long‐range contacts (N‐long_range) and C‐terminal long‐range contacts (C‐long_range), respectively. This manually selected cutoff (80 residues) was based on manual examination of N‐ and C‐tails of human kinases, especially the AGC group kinases that are known to have long C‐tails. In addition, residues in long insertions (more than five residues from either end of the two structural elements flanking the insertion) between KD structural elements are defined as long_insertion contacts if they make contact with residues in KD structural elements.

We employed two criteria to define intramolecular residue contacts between residues in the KD elements and residues outside the KD in AlphaFold models: distance between residue atoms and PAE values reported by AlphaFold (Jumper et al., 2021). PAE estimates a distance error for each residue pair in the AlphaFold model. Since the PAE matrix is not symmetric, for any residue pair with indices i and j, we chose the lowest value of pae_matrix[i][j] and pae_matrix[j][i] as its PAE value. Off‐diagonal PAE values have a range from 1 to 31 Å (Figure S1). The lower PAE value suggests better relative positioning of two residues. We used a cutoff value of 12 Å as a requirement for two residues to be in contact. This cutoff is based on the distribution of PAE values in all studied AlphaFold models which showed a low‐density and flattened dip from 12 to 17 Å (Figure S1). In addition to the PAE criterion, the distance between two residues (defined as the shortest distance among atom pairs) cannot exceed 5Å in order for them to be considered in contact.

4.3. Obtaining sequence and structure properties of residues that interact with KDs

We obtained pLDDT values (in the B‐factor field) for each residue in the AlphaFold models of human kinases (Tunyasuvunakool et al., 2021). The pLDDT score on a scale from 0 to 100 provides an estimation of the accuracy of the model at any residue position. These scores were binned into four categories as suggested in the AlphaFold online database: very high: pLDDT ≥ 90, confident: 90 > pLDDT ≥ 70, low: 70 > pLDDT ≥ 50, very low: pLDDT < 50. Secondary structures of AlphaFold models were calculated by STRIDE (Frishman & Argos, 1995). The per‐residue disorder predictions were obtained by the program IUPred2A (Meszaros et al., 2018). Sequence conservation values were calculated by AL2CO (Pei & Grishin, 2001). The conservation values were scaled to an integer from 0 to 9 (with the t option of AL2CO). Categories of high, medium, and low conservation have scores of 8–9, 6–7, and 0–5, respectively. We calculated background fractions of these properties by using all positions in the kinase dataset that show at least one PAE score less than 12 Å. The PAE scores of position pairs within 10 residues of each other were ignored in this position selection process to help remove those positions that only show low PAE scores with neighboring positions.

AUTHOR CONTRIBUTIONS

Jimin Pei: Conceptualization; Methodology; Software; Investigation; Writing ‐ original draft; Formal analysis. Qian Cong: Conceptualization; Methodology; Writing ‐ review & editing; Supervision.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Supporting information

FIGURE S1. The distribution of PAE values of AlphaFold models of human protein kinases.

FIGURE S2. The distribution of PAE values for KD‐interacting residues predicted to be ordered and those predicted to be disordered.

AlphaFold structural models and PAE matrices of human kinases, intramolecular interactions of human protein KDs, and the PyMOL sessions showing these interactions are available for download at: http://conglab.swmed.edu/kinreg.

TABLE S1. The list of human protein kinase domains and residue ranges of their structural elements.

Pei J, Cong Q. Computational analysis of regulatory regions in human protein kinases. Protein Science. 2023;32(10):e4764. 10.1002/pro.4764