Big steps toward understanding dynein

Dynein is a microtubule-based molecular motor that is involved in various biological functions, such as axonal transport, mitosis, and cilia/flagella movement. Although dynein was discovered 50 years ago, the progress of dynein research has been slow due to its large size and flexible structure. Recent progress in understanding the force-generating mechanism of dynein using x-ray crystallography, cryo-electron microscopy, and single molecule studies has provided key insight into the structure and mechanism of action of this complex motor protein.It has been 50 years since dynein was discovered and named by Ian Gibbons as a motor protein that drives cilia/flagella bending (Gibbons, 1963; Gibbons and Rowe, 1965). In the mid-1980s, dynein was also found to power retrograde transport in neurons (Paschal and Vallee, 1987). Subsequently, the primary amino acid sequence of the cytoplasmic dynein heavy chain, which contains the motor domain, was determined from the cDNA sequence (Mikami et al., 1993; Zhang et al., 1993). Like other biological motors, such as kinesins and myosins, the amino acid sequence of the dynein motor domain is well conserved. There are 16 putative genes that encode dynein heavy chains in the human genome (Yagi, 2009). Among these is one gene encoding cytoplasmic dynein heavy chain and one encoding retrograde intraflagellar transport dynein heavy chain, while the rest encode for heavy chains of axonemal dyneins. Most of the genes encoding the human dynein heavy chain have a counterpart in Chlamydomonas reinhardtii, which suggests that their functions are conserved from algae to humans.Dynein is unique compared with kinesin and myosin because dynein molecules form large molecular complexes. For example, one axonemal outer arm dynein molecule of C. reinhardtii is composed of three dynein heavy chains, two intermediate chains, and more than ten light chains (King, 2012). Mammalian cytoplasmic dynein consists of two heavy chains and several smaller subunits (Fig. 1 A; Vallee et al., 1988; Allan, 2011). The cargoes of cytoplasmic dynein are various membranous organelles, including lysosomes, endosomes, phagosomes, and the Golgi complex (Hirokawa, 1998). It is likely that one cytoplasmic dynein heavy chain can adapt to diverse cargos and functions by changing its composition.Open in a separate windowFigure 1.Atomic structures of cytoplasmic dynein. (A) Schematic structure of cytoplasmic dynein complex, adapted from Allan (2011). (B) The primary structure of cytoplasmic dynein. (C and D) The atomic model of D. discoideum cytoplasmic dynein motor domain (PDB accession no. 3VKG) overlaid on a microtubule (EMDB-5193; Sui and Downing, 2010) according to the orientation determined by Mizuno et al. (2007) (C) Side view. (D) View from the plus end of microtubule. (E) Schematic domain structure of dynein.Dynein must have a distinct motor mechanism from kinesin and myosin, because it belongs to the AAA+ family of proteins and does not have the conserved amino acid motifs, called the switch regions, present in kinesins, myosins, and guanine nucleotide-binding proteins (Vale, 1996). Therefore, studying dynein is of great interest because it will reveal new design principles of motor proteins. This review will focus on the mechanism of force generation by cytoplasmic and axonemal dynein heavy chains revealed by recent structural and biophysical studies.

Anatomy of dynein

To understand the chemomechanical cycle of dynein based on its molecular structure, it is important to obtain well-diffracting crystals and build accurate atomic models. Recently, Kon and colleagues determined the crystal structures of Dictyostelium discoideum cytoplasmic dynein motor domain, first at 4.5-Å resolution (Kon et al., 2011), and subsequently at 2.8 Å (without the microtubule binding domain) and 3.8-Å (wild type) resolution (Kon et al., 2012). Carter and colleagues also determined the crystal structures of the Saccharomyces cerevisiae (yeast) cytoplasmic dynein motor domain, first at 6-Å resolution (Carter et al., 2011), and later at 3.3–3.7-Å resolution (Schmidt et al., 2012). According to these crystal structures as well as previous EM studies, the overall structure of the dynein heavy chain is divided into four domains: tail, linker, head, and stalk (Fig. 1, B–E). Simply put, each domain carries out one essential function of a motor protein: the tail is the cargo binding domain, the head is the site of ATP hydrolysis, the linker is the mechanical amplifier, and the stalk is the track-binding domain.The tail, which is not part of the motor domain and is absent from crystal structures, is located at the N-terminal ∼1,400 amino acid residues and involved in cargo binding (gray in Fig. 1, B and E). The next ∼550 residues comprise the “linker” (pink in Fig. 1, B–E), which changes its conformation depending on the nucleotide state (Burgess et al., 2003; Kon et al., 2005). This linker domain was first observed by negative staining EM in combination with single particle analysis of dynein c, an isoform of inner arm dynein from C. reinhardtii flagella (Burgess et al., 2003). According to the crystal structures, the linker is made of bundles of α-helices and lies across the AAA+ head domain, forming a 10-nm-long rod-like structure (Fig. 1, C and D). Recent class averaged images of D. discoideum cytoplasmic dynein show that the linker domain is stiff along its entire length when undocked from the head (Roberts et al., 2012). The head (motor) domain of dynein is composed of six AAA+ (ATPase associated with diverse cellular activities) modules (Neuwald et al., 1999; color-coded in Fig. 1, B–E). Although many AAA+ family proteins are a symmetric homohexamer (Ammelburg et al., 2006), the AAA+ domains of dynein are encoded by a single heavy chain gene and form an asymmetric heterohexamer. Among the six AAA+ domains, hydrolysis at the first AAA domain mainly provides the energy for dynein motility (Imamula et al., 2007; Kon et al., 2012). The hexameric ring has two distinct faces: the linker face and the C-terminal face. The linker face is slightly convex and the linker domain lies across this side (Fig. 1 D, left side). The other side of the ring has the C-terminal domain (Fig. 1 D, right side).The stalk domain of dynein was identified as the microtubule-binding domain (MTBD; Gee et al., 1997). It emanates from the C-terminal face of AAA4 and is composed of antiparallel α-helical coiled-coil domain (yellow in Fig. 1, B–E). The tip of the stalk is the actual MTBD. Interestingly, the crystal structures revealed another antiparallel α-helical coiled coil that emerges from AAA5 (orange in Fig. 1, B–E), and this region is called the buttress (Carter et al., 2011) or strut (Kon et al., 2011), which was also observed as the bifurcation of the stalk by negative-staining EM (Burgess et al., 2003; Roberts et al., 2009). The tip of the buttress/strut is in contact with the middle of the stalk and probably works as a mechanical reinforcement of the stalk.

The chemomechanical cycle of dynein

Based on structural and biochemical data, a putative chemomechanical cycle of dynein is outlined in Fig. 2 (A–E). In the no-nucleotide state, dynein is bound to a microtubule through its stalk domain, and its tail region is bound to cargoes (Fig. 2 A). The crystal structures of yeast dynein are considered to be in this no-nucleotide state. When ATP is bound to the AAA+ head, the MTBD quickly detaches from the microtubule (Fig. 2 B; Porter and Johnson, 1983). The ATP binding also induces “hinging” of the linker from the head (Fig. 2 C). According to the biochemical analysis of recombinant D. discoideum dynein (Imamula et al., 2007), the detachment from the microtubule (Fig. 2, A and B) is faster than the later hinging (Fig. 2, B and C). As a result of these two reactions, the head rotates or shifts toward the minus end of the microtubule (for more discussion about “rotate” versus “shift” see the “Dyneins in the axoneme” section) and the MTBD steps forward. The directionality of stepping seems to be mainly determined by the MTBD, because the direction of dynein movement does not change even if the head domain is rotated relative to the microtubule by insertion or deletion of the stalk (Carter et al., 2008). In the presence of ADP and vanadate, dynein is considered to be in this state (Fig. 2 C).Open in a separate windowFigure 2.Presumed chemomechanical cycle and stepping of dynein. (A–E) Chemomechanical cycle of dynein. The pre- and post-power stroke states are also called the primed and unprimed states, respectively. The registries of the stalk coiled coil are denoted as α and β according to Gibbons et al. (2005). (F and G) Processive movement of kinesin (F) and dynein (G). (F) Hand-over-hand movement of kinesin. A step by one head (red) is always followed by the step of another head (green). The stepping of kinesin is on one protofilament of microtubule. (G) Presumed stepping of dynein. The step size varies and the interhead separation can be large. A step by one head (red) is not always flowed by the step of another head (green). (H) A model of strain-based dynein ATPase activation. (G, top) Without strain, the gap between the AAA1 and AAA2 is open and the motor domain cannot hydrolyze ATP. (G, bottom) Under a strain imposed between MTBD and tail (thin black arrows), the gap becomes smaller (thick black arrows) and turns on ATP hydrolysis by dynein.After the MTBD rebinds to the microtubule at the forward site (Fig. 2 D), release of hydrolysis products from the AAA+ head is activated (Holzbaur and Johnson, 1989) and the hinged linker goes back to the straight conformation (Fig. 2 E; Kon et al., 2005). The crystal structure of D. discoideum dynein is considered to be in the state after phosphate release and before ADP release. This straightening of the linker is considered to be the power-generating step and brings the cargo forward relative to the microtubule.

The MTBD of dynein

As outlined in Fig. 2, the nucleotide state of the head domain may control the affinity of the MTBD to the microtubule. Conversely, the binding of the MTBD to the microtubule should activate the ATPase activity of the head domain. This two-way communication is transmitted through the simple ∼17-nm-long α-helical coiled-coil stalk and the buttress/strut, and its structural basis has been a puzzling question.Currently there are three independent MTBD atomic structures in the Protein Data Bank (PDB): One of the crystal structures of the D. discoideum dynein motor domain contains the MTBD (Fig. 3 A), and Carter et al. (2008) crystallized the MTBD of mouse cytoplasmic dynein fused with a seryl tRNA-synthetase domain (Fig. 3 C). The MTBD structure of C. reinhardtii axonemal dynein was solved using nuclear magnetic resonance (PDB accession no. 2RR7; Fig. 3 B). The MTBD is mostly composed of α-helices and the three structures are quite similar to each other within the globular MTBD (Fig. 3). Note that dynein c has an additional insert at the MTBD–microtubule interface (Fig. 3 B, inset), whose function is not yet clear. The three structures start to deviate from the junction between the MTBD and the coiled-coil region of the stalk (Fig. 3, A–C, blue arrowheads). Particularly, one of the stalk α-helix (CC2) in D. discoideum dynein motor domain appears to melt at the junction with the MTBD (Fig. 3 A, red arrowhead). This structural deviation suggests that the stalk coiled coil at the junction is flexible, which is consistent with the observation by EM (Roberts et al., 2009).Open in a separate windowFigure 3.Atomic models of the MTBD of dynein. (A) D. discoideum cytoplasmic dynein (PDB accession no. 3VKH). (B) C. reinhardtii dynein c (PDB accession no. 2RR7). The inset shows the side view, highlighting the dynein c–specific insert. (C) Mouse cytoplasmic dynein (PDB accession no. 3ERR). (D) Mouse cytoplasmic dynein fit to the MTBD–microtubule complex derived from cryo-EM (PDB accession no. 3J1T). All the MTBD structures were aligned using least square fits and color-coded with a gradient from the N to C terminus. CC1, coiled coil helix 1; CC2, coiled coil helix 2. The blue arrowheads points to the junction between MTBD and the stalk, where a well-conserved proline residue (colored pink) is located. In C and D, two residues (isoleucine 3269 and leucine 3417) are shown as spheres. The two residues form hydrophobic contacts in the β-registry (C), whereas they are separated in the α-registry (D) because of the sliding between the two α-helices (blue and red arrows). Conformational changes observed in the mouse dynein MTBD in complex with a microtubule by cryo-EM are shown by black arrows. Note that the cryo-EM density map does not have enough resolution to observe sliding between CC1 and CC2. The sliding was modeled based on targeted molecular dynamics (Redwine et al., 2012).Various mechanisms have been proposed to explain how the affinity between the MTBD and a microtubule is controlled. Gibbons et al. (2005) proposed “the helix-sliding hypothesis” (for review see Cho and Vale, 2012). In brief, this hypothesis proposes that the sliding between two α-helices CC1 and CC2 (Fig. 3, C and D; blue and red arrows) may control the affinity of this domain to a microtubule. When Gibbons’s classification (Gibbons et al., 2005) of the sliding state is applied to the three MTBD structures, the stalk in the D. discoideum dynein motor domain is in the “α-registry” state (not visible in Fig. 3 A because of the melting of CC2), which corresponds to the strong binding state. However, the mouse cytoplasmic and C. reinhardtii axonemal MTBDs have the “β-registry” stalk (Fig. 3 C), which corresponds to the weak binding state.To observe conformational changes induced by the α-registry and/or microtubule binding, Redwine et al. (2012) solved the structure of mouse dynein MTBD in complex with a microtubule at 9.7-Å resolution using cryo-EM and single particle analysis. The MTBD was coupled with seryl tRNA-synthetase to fix the stalk helix in the α-registry. At this resolution, α-helices are visible, and they used molecular dynamics to fit the crystal structure of mouse MTBD (β-registry) to the cryo-EM density map. According to this result, the first helix H1 moves ∼10 Å to a position that avoids a clash with the microtubule (Fig. 3 D, black arrows). This also induces opening of the stalk helix (CC1). Together with mutagenesis and single-molecule motility assays, Redwine et al. (2012) proposed that this new structure represents the strong binding state. Currently, it is not clear why the MTBD structure of D. discoideum dynein motor domain (α-registry, Fig. 3 A) is not similar to the new α-registry mouse dynein MTBD, and this problem needs to be addressed by further studies.

Structures around the first ATP binding site

Another central question about motor proteins is how Ångstrom-scale changes around the nucleotide are amplified to generate steps >8 nm. For dynein, the interface between the first nucleotide-binding pocket and the linker seem to be the key force-generating element (Fig. 4). The crystal structures of dynein give us clues about how nucleotide-induced conformational changes may be transmitted to and amplified by the linker domain.Open in a separate windowFigure 4.Structures around the first ATP binding site. (A) Schematic domain structure of the head domain. Regions contacting the linker domain are colored purple. (B) AAA submodules surrounding the first nucleotide-binding pocket (PDB accession no. 3VKG, chain A). The linker is connected to AAA1 domain by the “N-loop.” To highlight that the two finger-like structures are protruding, the shadow of the atomic structure has been cast on the plane parallel to the head domain. (C) Interaction between the linker and the two finger-like structures. The pink arrowhead points to the hinge-like structure of the linker. The pink numbers indicates the subdomain of the linker.The main ATP catalytic site is located between AAA1 and AAA2 (Fig. 4, A and B). There are four ADP molecules in the D. discoideum dynein crystal structures, but the first ATP binding site alone drives the microtubule-activated ATPase activity, based on biochemical experiments on dyneins whose ATP binding sites were mutated (Kon et al., 2012).One AAA+ module is composed of a large submodule and a small α submodule (Fig. 4 B). The large α/β submodule is located inside of the ring and the small α submodule is located outside. The large submodule bulges toward the linker face, and the overall ring forms a dome-like shape (Fig. 1 D).The main ATP catalytic site is surrounded by three submodules: AAA1 large α/β, AAA1 small α, and AAA2 large α/β (Fig. 4, A and B). Based on the structural changes of other AAA+ proteins (Gai et al., 2004; Suno et al., 2006; Wendler et al., 2012), the gap between AAA1 and AAA2 modules is expected to open and close during the ATPase cycle.In fact, the size of the gap varies among the dynein crystal structures. The crystal structures of yeast dyneins show a larger gap between AAA1 and AAA2, which might be the reason why no nucleotide was found in the binding pocket. Although Schmidt et al. (2012) soaked the crystals in a high concentration of various nucleotides (up to 25 mM of ATP), no electron densities corresponding to the nucleotide were observed at the first ATP binding site. Among dynein crystal structures, one of D. discoideum dynein (PDB accession no. 3VKH, chain A) has the smallest gap, but it is still considered to be in an “open state” because the arginine finger in the AAA2 module (Fig. 4 B, red) is far from the phosphates of ADP. Because the arginine finger is essential for ATP hydrolysis in other AAA+ proteins (Ogura et al., 2004), the gap is expected to close and the arginine finger would stabilize the negative charge during the transition state of ATP hydrolysis.The presumed open/close conformational change between AAA1 and AAA2 would result in the movement of two “finger-like” structures protruding from the AAA2 large α/β submodule (Fig. 4 B). The two finger-like structures are composed of the H2 insert β-hairpin and preSensor I (PS-I) insert. In D. discoideum dynein crystal structure, the two finger-like structures are in contact with the “hinge-like cleft” of the linker (Fig. 4 C, pink arrowhead). The hinge-like cleft is one of the thinnest parts of the linker, where only one α-helix is connecting between the linker subdomains 2 and 3.In the yeast crystal structures, which have wider gaps between AAA1 and AAA2, the two finger-like structures are not in direct contact with the linker and separated by 18 Å. Instead, the N-terminal region of the linker is in contact with the AAA5 domain (Fig. 4 A). To test the functional role of the linker–AAA5 interaction, Schmidt et al. (2012) mutated a residue involved in the interaction (Phe3446) and found that the mutation resulted in severe motility defects, showing strong microtubule binding and impaired ATPase activities. In D. discoideum dynein crystals, there is no direct interaction between AAA5 and the linker, which suggests that the gap between AAA1 and AAA2 may influence the interaction between the head and linker domain. The contact between the linker and AAA5 may also influence the gap around AAA5, because the gap between AAA5 and AAA6 is large in yeast dynein crystal, whereas the one between AAA4 and AAA5 is large in D. discoideum dynein.The movement of two finger-like structures would induce remodeling of the linker. According to the recent cryo-EM 3D reconstructions of cytoplasmic dynein and axonemal dynein c (Roberts et al., 2012), the linker is visible across the head and there is a large gap between AAA1 and AAA2 in the no-nucleotide state. This linker structure is considered to be the “straight” state (Fig. 2, A and E). In the presence of ADP vanadate, the gap between AAA1 and AAA2 is closed and the N-terminal region of linker is near AAA3, which corresponds to the pre-power stroke “hinged” state (Fig. 2, C and D). The transition from the hinged state to the straight state of the linker is considered to be the force-generating step of dynein.

Processivity of dynein

As the structure of dynein is different from other motor proteins, dynein’s stepping mechanism is also distinct. Both dynein and kinesin are microtubule-based motors and move processively. Based on the single molecule tracking experiment with nanometer accuracy (Yildiz et al., 2004), it is widely accepted that kinesin moves processively by using its two motor domain alternately, called the “hand-over-hand” mechanism. To test whether dynein uses a similar mechanism to kinesin or not, recently Qiu et al. (2012) and DeWitt et al. (2012) applied similar single-molecule approaches to dynein.To observe the stepping, the two head domains of yeast recombinant cytoplasmic dynein were labeled with different colors and the movement of two head domains was tracked simultaneously. If dynein walks by the hand-over-hand mechanism, the step size would be 16 nm and the stepping of one head domain would always be followed by the stepping of another head domain (alternating pattern), and the trailing head would always take a step (Fig. 2 F). Contrary to this prediction, both groups found that the stepping of the head domains is not coordinated when the two head domains are close together. These observations indicated that the chances of a leading or trailing head domain stepping are not significantly different (Fig. 2 G; DeWitt et al., 2012; Qiu et al., 2012).This stepping pattern predicts that the distance between the head domains can be long. In fact, the distance between the two head domains is on average ∼18 ± 11 nm (Qiu et al., 2012) or 28.4 ± 10.7 nm (DeWitt et al., 2012), and as large as ∼50 nm (DeWitt et al., 2012). When the two head domains are separated, there is a tendency where stepping of the trailing head is preferred over that of the forward head.In addition, even though the recombinant cytoplasmic dynein is a homodimer, the two heavy chains do not function equally. While walking along the microtubule, the leading head tends to walk on the right side, whereas the trailing head walks on the left side (DeWitt et al., 2012; Qiu et al., 2012). This arrangement suggests that the stepping mechanism is different between the two heads. In fact, when one of the two dynein heavy chains is mutated to abolish the ATPase activity at AAA1, the heterodimeric dynein still moves processively (DeWitt et al., 2012), with the AAA1-mutated dynein heavy chain remaining mostly in the trailing position. This result clearly demonstrates that allosteric communication between the two AAA1 domains is not required for processivity of dynein. It is likely that the mutated head acts as a tether to the microtubule, as it is known that wild-type dynein can step processively along microtubules under external load even in the absence of ATP (Gennerich et al., 2007).These results collectively show that dynein moves by a different mechanism from kinesin. It is likely that the long stalk and tail allow dynein to move in a more flexible manner.

Dyneins in the axoneme

As mentioned in the introduction, >10 dyneins work in motile flagella and cilia. The core of flagella and cilia is the axoneme, which is typically made of nine outer doublet microtubules and two central pair microtubules (“9 + 2,” Fig. 5 A). The axonemes are found in various eukaryotic cells ranging from the single-cell algae C. reinhardtii to human. Recent extensive cryo-electron tomography (cryo-ET) in combination with genetics revealed the highly organized and complex structures of axonemes that are potentially important for regulating dynein activities (Fig. 5, C and D; Nicastro et al., 2006; Bui et al., 2008, 2009, 2012; Heuser et al., 2009, 2012; Movassagh et al., 2010; Lin et al., 2012; Carbajal-González et al., 2013; Yamamoto et al., 2013).Open in a separate windowFigure 5.Arrangement of axonemal dyneins. (A) The schematic structure of the motile 9 + 2 axoneme, viewed from the base of flagella. (B) Quasi-planar asymmetric movement of the 9 + 2 axoneme typically observed in trachea cilia or in C. reinhardtii flagella. (C and D) 3D structure of a 96-nm repeat of doublet microtubules in the distal/central region of C. reinhardtii flagella (EMDB-2132; Bui et al., 2012). N-DRC, the nexin-dynein regulatory complex; ICLC, intermediate chain/light chain complex. Inner arm dynein subspecies are labeled according to Bui et al. (2012) and Lin et al. (2012). To avoid the confusion with the linker domain of dynein, the structures connecting between outer and inner arm dyneins are labeled as “connecters,” which are normally called “linkers.” Putative ATP binding sites of outer arm dynein determined by biotin-ADP (Oda et al., 2013) are indicated by orange circles. The atomic structure of cytoplasmic dynein is placed into the β-heavy chain of outer arm dynein and its enlarged view is shown in the inset. (D) Two doublet microtubules, viewed from the base of flagella.The basic mechanochemical cycles of axonemal dyneins are believed to be shared with cytoplasmic dynein. Dynein c is an inner arm dynein of C. reinhardtii and used extensively to investigate the conformational changes of dynein, as shown in Fig. 2 (A–E), by combining EM and single-particle analysis (Burgess et al., 2003; Roberts et al., 2012). Structural changes of axonemal dyneins complexed with microtubules are also observed by quick-freeze and deep-etch EM (Goodenough and Heuser, 1982; Burgess, 1995), cryo-EM (Oda et al., 2007), negative-staining EM (Ueno et al., 2008), and cryo-ET (Movassagh et al., 2010). According to these studies, the AAA+ head domains are constrained near the A-tubule in the no-nucleotide state. In the presence of nucleotide, the head domains move closer to the B-tubule and/or the minus end of microtubule, and their appearance becomes heterogeneous, which is consistent with the observation of isolated dynein c that shows greater flexibility between tail and stalk in the ADP/vanadate state (Burgess et al., 2003).One of the controversies about the structural changes of axonemal dyneins is whether their stepping involves “rotation” or “shift” of the head (Fig. 2, B to D). The stalk angle relative to the microtubule seems to be a constant ∼60° irrespective of the nucleotide state (Ueno et al., 2008; Movassagh et al., 2010). This angle is similar to the angle obtained from cryo-EM study of the MTBD–microtubule complex (Redwine et al., 2012). Based on these observations, Ueno et al. (2008) and Movassagh et al. (2010) hypothesize that the “shift” of the head pulls the B-microtubule toward the distal end. However, Roberts et al. (2012) propose that the “rotation” of head and stalk is involved in the stepping based on the docking of dynein c head into an averaged flagella tomogram obtained by Movassagh et al. (2010). This issue needs to be resolved by more reliable and high-resolution data, but these two models may not be mutually exclusive. For example, averaged tomograms may be biased toward the microtubule-bound stalk because tomograms are aligned using microtubules.To interpret these structural changes of axonemal dyneins, docking atomic models of dynein is necessary. According to Roberts et al. (2012), the linker face of inner arm dynein c is oriented outside of axoneme (Fig. 5 D). For outer arm dyneins, we used cryo-EM in combination with biotin-ADP-streptavidin labeling and showed that the ATP binding site, most likely AAA1, is on the left side of the AAA+ head (Fig. 5 C; Oda et al. (2013)). Assuming that the stalks extend out of the plane toward the viewer, the linker face of outer arm dynein is oriented outside of axoneme (Fig. 5 C, inset; and Fig. 5 D). If it were the opposite, the AAA1 would be located on the right side of the AAA+ head. In summary, both inner and outer arm dynein seem to have the same arrangement, with their linker face oriented outside of the axoneme (Fig. 5 D).A unique characteristic of axonemal dyneins is that these dyneins are under precise temporal and spatial control. To generate a planer beating motion (Fig. 5 B), dyneins should be asymmetrically controlled, because the dyneins located on doublets 2–4 drive the effective stroke, whereas the ones on doublets 6–8 drive the recovery stroke (Fig. 5 A). Based on the cryo-ET observation of axonemes, Nicastro et al. (2006) proposed that “linkers” between dyneins provide hard-wiring to coordinate motor activities. Because the linkers in axonemes are distinct structures from the linker domain of dynein, for clarity, here we call them “connecters.” According to the recent cryo-ET of proximal region of C. reinhardtii flagella (Bui et al., 2012), there are in fact asymmetries among nine doublets that are localized to the connecters between outer and inner arm dynein, called the outer-inner dynein (OID) connecters (Fig. 5, A and C). Recently we identified that the intermediate chain 2 (IC2) of outer arm dynein is a part of the OID connecters, and a mutation of the N-terminal region of IC2 affects functions of both outer and inner arm dyneins (Oda et al., 2013), which supports the idea that the connecters between dyneins are involved in axonemal dynein regulation.

Closing remarks

Thanks to the crystal structures, we can now design and interpret experiments such as single molecule assays and EM based on the atomic models of dynein. Our understanding of the molecular mechanism and cellular functions of dyneins will be significantly advanced by these experiments in the near future.One important direction of dynein research is to understand the motor mechanisms closer to the in vivo state. For example, the step sizes of cytoplasmic dynein purified from porcine brain is ∼8 nm independent of load (Toba et al., 2006). This result suggests that intermediate and light chain bound to the dynein heavy chain may modulate the motor activity of dynein. To address such questions, Trokter et al. (2012) reconstituted human cytoplasmic dynein complex from recombinant proteins, although the reconstituted dynein did not show robust processive movement. Further studies are required to understand the movement of cytoplasmic dynein. Similarly, axonemal dyneins should also be studied using mutations in a specific gene that does not affect the overall flagella structure, rather than depending on null mutants that cause the loss of large protein complexes.Detailed full chemomechanical cycle of dynein and its regulation are of great importance. Currently, open/closed states of the gap between AAA1 and AAA2 are not clearly correlated with the chemomechanical cycle of dynein. Soaking dynein crystal with nucleotides showed that the presence of ATP alone is not sufficient to close the gap, at least in the crystal (Schmidt et al., 2012). This result suggests that other factors such as a conformational change of the linker are required. For other motors, ATP hydrolysis is an irreversible chemical step, which is often “gated” by strain. In the case of kinesin, ATP is hydrolyzed by a motor domain only when a forward strain is applied by the other motor domain through the neck linker (Cross, 2004; Kikkawa, 2008). A similar strain-based gating mechanism may play important roles in controlling the dynein ATPase. Upon MTBD binding to the forward binding site, a strain between MTBD and tail would be applied to the dynein molecule. The Y-shaped stalk and strut/buttress under the strain would force the head domain to close the gap between AAA4 and AAA5 (Fig. 2 H). Similarly, the linker under the strain would be hooked onto the two finger-like structures and close the gap between AAA1 and AAA2 (Fig. 2 H). The gap closure then triggers ATP hydrolysis by dynein. This strain-based gating of dynein is consistent with the observation that the rate of nonadvancing backward steps, which would depend on ATP hydrolysis, is increased by load applied to dynein (Gennerich et al., 2007). To explain cilia and flagella movement, the geometric clutch hypothesis has been proposed (Lindemann, 2007), which contends that the forces transverse (t-force) to the axonemal axis act on the dynein to regulate dynein activities. In the axoneme, dynein itself can be the sensor of the t-force by the strain-based gating mechanism. Further experiments are required to test this idea, but the strain-based gating could be a shared property of biological motors.     

The 2015 Bioinformatics Open Source Conference (BOSC 2015)

The Bioinformatics Open Source Conference (BOSC) is organized by the Open Bioinformatics Foundation (OBF), a nonprofit group dedicated to promoting the practice and philosophy of open source software development and open science within the biological research community. Since its inception in 2000, BOSC has provided bioinformatics developers with a forum for communicating the results of their latest efforts to the wider research community. BOSC offers a focused environment for developers and users to interact and share ideas about standards; software development practices; practical techniques for solving bioinformatics problems; and approaches that promote open science and sharing of data, results, and software. BOSC is run as a two-day special interest group (SIG) before the annual Intelligent Systems in Molecular Biology (ISMB) conference. BOSC 2015 took place in Dublin, Ireland, and was attended by over 125 people, about half of whom were first-time attendees. Session topics included “Data Science;” “Standards and Interoperability;” “Open Science and Reproducibility;” “Translational Bioinformatics;” “Visualization;” and “Bioinformatics Open Source Project Updates”. In addition to two keynote talks and dozens of shorter talks chosen from submitted abstracts, BOSC 2015 included a panel, titled “Open Source, Open Door: Increasing Diversity in the Bioinformatics Open Source Community,” that provided an opportunity for open discussion about ways to increase the diversity of participants in BOSC in particular, and in open source bioinformatics in general. The complete program of BOSC 2015 is available online at http://www.open-bio.org/wiki/BOSC_2015_Schedule.Open in a separate window     

YCF1: A Green TIC: Response to the de Vries et al. Commentary

This response to a recent Commentary article by de Vries et al. highlights critical errors in the annotation and identification of Ycf1 homologs in the sequenced chloroplast genomes. Contrary to what is reported by de Vries et al., the majority of chloroplast genomes sequenced to date appear to have retained a typical Ycf1 sequence (i.e., including the N-terminal 6TM domain and a variable hydrophilic C-terminal domain) as my group previously reported. Our evidence continues to support the model that Ycf1 forms an essential component of a “green TIC” that is largely conserved among the Chlorophyta and land plants. Since the establishment of this green TIC with Tic20 as the core component, some cases of loss of Ycf1 during the evolution of the green lineages might be regarded as modifications or alterations of the complex. Here, I discuss our working model that the presence of an alternative “nonphotosynthetic-type” or “ancestral-type” TIC might explain other (or specific) cases of the lack of Ycf1, not only in early lineages, including Glaucophyta and Rhodophyta, but also in the grasses.Virtually all chloroplasts/plastids in today’s photosynthetic, plastid-containing eukaryotes derive from one successful primary endosymbiotic event with a cyanobacterium-like ancestor. During evolution, massive transfer of genes from the endosymbiont to the host’s nuclear genome occurred concomitant with the establishment of a protein transport system that allows these nucleus-encoded proteins back into the endosymbiotic organelle. Two successive protein translocons at the outer and inner envelope membranes of chloroplast, termed TOC and TIC, respectively, are responsible for this protein transport. My group recently identified a novel TIC complex consisting of Tic20, Tic56, Tic100, and Tic214 in Arabidopsis thaliana (Kikuchi et al., 2013). We found that Tic214 is encoded by the previously enigmatic essential chloroplast gene ycf1 (Boudreau et al., 1997; Drescher et al., 2000). This discovery led us to an extensively revised view of the molecular mechanisms of chloroplast protein import (Nakai, 2015). We found this TIC complex to be absolutely required for photosynthetic protein import in green tissues and, thus, essential for plant viability in Arabidopsis (Hirabayashi et al., 2011; Kikuchi et al., 2013). Experimental evidence for the presence of a similar TIC complex in pea (Pisum sativum) (Kikuchi et al., 2009) supports the idea that this complex is a general feature of the TIC apparatus. The TIC complex and Ycf1 appear to have evolved dramatically along with the evolution of the green lineage, including green algae and land plants; thus, they can be regarded as a “general green TIC.”In their commentary, de Vries et al. (2015) repeated a phylogenetic analysis of Ycf1 and asked why 30 years of TOC/TIC research has missed this green TIC complex. Our response is that many things have been missed over the last 30 years. Several “old” Tic proteins, such as Tic40 and Tic110, were mistakenly identified very early (probably because of their high abundance in the inner envelope membrane and/or of the use of inadequate methodology) and have long been believed to be true translocon components without further concrete evidence for their direct involvement in protein transport (Bölter et al., 2015; Paila et al., 2015). It has become clear that it is time to reevaluate the functions of these historic Tic proteins as well as of the so-called redox regulators, namely, Tic32, Tic55, and Tic62 (Nakai, 2015). To address this, we developed a highly specific method to purify and identify functional TIC complex components (Supplemental Figure 1; Kikuchi et al., 2013; Nakai, 2015). For now, until the historic TIC components are demonstrated to function directly in protein transport, we have removed them from our working model (Figure 1).Open in a separate windowFigure 1.Revised Working Model for the Coordinated Function of the Photosynthetic-Type Major TIC Complex and the Alternative Nonphotosynthetic-Type TIC Complex in Substrate-Specific or Tissue-Specific Protein Import in Concert with Different Types of TOC Complexes in Most Land Plants.The alternative TIC complex is predicted to have direct evolutionary relationships with a distinct Tic20-containing TIC complex functioning in grasses and also with simpler ancestral-type TIC complexes that probably have been retained in all plastid-containing lineages, including Glaucophyta and Rhodophyta. Substrate specificities and redundancies of distinct TOC and TIC complexes still remain to be elucidated. See text for details. (Adapted from Nakai [2015], Figure 3, with permission from Elsevier.)While we need to recognize the limitations of predicting the essentiality of a gene/protein based solely on its presence or absence in available genome sequence and annotation databases, de Vries et al. (2015) raise several important questions about Ycf1 function and the evolution of the TIC system. Principal among these are: (1) What is the true evolutionary history of Ycf1, and (2) how can the absences of this “essential” function be explained in the grasses and certain other lineages? Here, we highlight what we believe to be critical errors presented by de Vries et al. (2015) in the annotation and identification of Ycf1 homologs in the sequenced chloroplast genomes and summarize hypotheses that offer reasonable answers to both of these questions.     

Science Translator: An Interview with Louisa Stark

The Genetics Society of America’s Elizabeth W. Jones Award for Excellence in Education recognizes significant and sustained impact on genetics education. The 2015 awardee, Louisa Stark, has made a major impact on global access to genetics education through her work as director of the University of Utah Genetic Science Learning Center. The Center’s Learn.Genetics and Teach.Genetics websites are the most widely used online genetic education resources in the world. In 2014, they were visited by 18 million students, educators, scientists, and members of the public. With over 60 million page views annually, Learn.Genetics is among the most used sites on the Web.Open in a separate window     

Mutation in Bovine β-Carotene Oxygenase 2 Affects Milk Color

β-Carotene biochemistry is a fundamental process in mammalian biology. Aberrations either through malnutrition or potentially through genetic variation may lead to vitamin A deficiency, which is a substantial public health burden. In addition, understanding the genetic regulation of this process may enable bovine improvement. While many bovine QTL have been reported, few of the causative genes and mutations have been identified. We discovered a QTL for milk β-carotene and subsequently identified a premature stop codon in bovine β-carotene oxygenase 2 (BCO2), which also affects serum β-carotene content. The BCO2 enzyme is thereby identified as a key regulator of β-carotene metabolism.THE metabolism of β-carotene to form vitamin A is nutritionally important, and vitamin A deficiency remains a significant public health burden. Genetic variation may underlie individual differences in β-carotene metabolism and contribute to the etiology of vitamin A deficiency. Within an agricultural species, genetic variation provides opportunity for production improvements, disease resistance, and product specialization options. We have previously shown that natural genetic variation can be successfully used to inform bovine breeding decisions (Grisart et al. 2002; Blott et al. 2003). Despite numerous reports of quantitative trait loci (QTL), few causative mutations have been identified. We discovered a QTL for milk β-carotene content and report here the identification of a mutation in the bovine β-carotene oxygenase 2 (BCO2) gene responsible for this QTL. The mutation, which results in a premature stop codon, supports a key role for BCO2 in β-carotene metabolism.The QTL trial consisted of a Holstein-Friesian × Jersey cross in an F₂ design and a half-sibling family structure (Spelman et al. 2001). Six F₁ sires and 850 F₂ female progeny formed the trial herd. To construct the genetic map, the pedigree (including the F₁ sires, F₁ dams, F₂ daughters, and selected F₀ grandsires: n = 1679) was genotyped, initially with 237 microsatellite markers, and subsequently, with 6634 SNP markers (Affymetrix Bovine 10K SNP GeneChip). A wide range of phenotypic measures relating to growth and development, health and disease, milk composition, fertility, and metabolism were scored on the F₂ animals from birth to 6 years of age.To facilitate the discovery of QTL and genes regulating β-carotene metabolism, milk concentration of β-carotene was measured during week 6 of the animals'' second lactation (n = 651). Using regression methodology in a half-sib model (Haley et al. 1994; Baret et al. 1998), a QTL on bovine chromosome 15 (P < 0.0001; Figure 1A) was discovered. The β-carotene QTL effect on chromosome 15 was also significant (P < 0.0001) at two additional time points, in months 4 and 7 of lactation. Three of the six F₁ sire families segregated for the QTL, suggesting that these three F₁ sires would be heterozygous for the QTL allele (“Q”). To further define the most likely region within the QTL that would harbor the causative mutation, we undertook association mapping, using the 225 SNP markers that formed the chromosome 15 genetic map (Figure 1A). One SNP (“PAR351319”) was more closely associated with the β-carotene phenotype than any other marker (P = 2.522^E−18). This SNP was located beneath the QTL peak. Further, the SNP was heterozygous in the three F₁ sires that segregated for the QTL, and homozygous in the remaining three sires. On this basis, we hypothesized that the milk β-carotene phenotype would differ between animals on the basis of the genotype of SNP PAR351319.Open in a separate windowFigure 1.—Discovery of BCO2 mutation affecting milk β-carotene concentration. (A) The β-carotene QTL on bovine chromosome 15 (P < 0.0001) is shown by the red line. The maximum F-value at 21 cM was 7.15. The 95% confidence interval is shown by the shaded box. The association of each marker with milk β-carotene is shown by the blue dots, and the association of the BCO2 genotype is shown by the green diamond. A total of 233 informative markers (8 microsatellite markers and 225 single nucleotide polymorphisms) were included on the genetic map for BTA15. QTL detection was conducted using regression methodology in a line of descent model (Haley et al. 1994) and a half-sib model (Baret et al. 1998). Threshold levels were determined at the chromosomewide level using permutation testing (Churchill and Doerge 1998) and confidence intervals estimated using bootstrapping (Visscher et al. 1996). (B) The haplotypes of 10 representative animals for “QQ” and “qq” are shown for the SNP markers encompassing the SNP (“PAR351319”) most closely associated with the milk β-carotene phenotype. Light and dark gray boxes represent homozygous SNPs, while white boxes represent heterozygous SNPs. The genes present within the defined region are also shown. (C) The mutation in the bovine BCO2 gene is shown. The structure of the BCO2 gene is indicated by the horizontal bar, with vertical bars representing exons 1–12. The A > G mutation in exon 3 (red) causes a premature termination codon at amino acid position 80. (D) The mean concentration of β-carotene in the milk fat of “QQ,” “Qq,” and “qq” cows is shown. β-Carotene was measured by absorbance at 450 nm as previously described (Winkelman et al. 1999). Data are means ± SEM. The statistical significance was determined using ANOVA (***P < 0.0001; n = 651).We then made the following assumptions: that the effect of the QTL was additive, that the Q allele was present in the dam population, allowing the occurrence of homozygous (“QQ”) offspring, and that the QTL was caused by a single mutation, acting with a dominant effect on the milk β-carotene phenotype. Haplotypes encompassing the PAR351319 SNP were determined in the F₂ offspring. A comparison of the phenotypic effect of homozygous Q, heterozygous and homozygous q individuals revealed that indeed, animals with the “QQ” genotype had a higher concentration of milk β-carotene than animals with the “qq” genotype (Figure 1D). We predicted that the region of homozygosity was likely to contain the causative gene and mutation. The extent of this region and the candidate genes contained within it are shown in Figure 1B. A total of 10 genes with known function, including BCO2, were located within the region. This information, combined with knowledge of the role BCO2 plays in β-carotene metabolism in other species (Kiefer et al. 2001), made BCO2 a good positional candidate for the QTL. We therefore sequenced the entire coding region (12 exons, {"type":"entrez-nucleotide","attrs":{"text":"NC_007313.3","term_id":"194719391","term_text":"NC_007313.3"}}NC_007313.3) of the BCO2 gene in each of the six F₁ sires. An A > G mutation, which was heterozygous in the three F₁ sires that segregated for the QTL, was discovered in exon three, 240 bp from the translation initiation site (Figure 1C). The three remaining sires were homozygous for the G allele, which encodes the 530-amino-acid BCO2 protein ({"type":"entrez-protein","attrs":{"text":"NP_001101987","term_id":"241666456","term_text":"NP_001101987"}}NP_001101987). The A allele creates a premature stop codon resulting in a truncated protein of 79 amino acids. To determine whether this mutation was associated with the QTL, the remainder of the pedigree was genotyped. The BCO2 genotype was significantly associated with the milk β-carotene phenotype (P = 8.195^E−29) The AA genotype (referred to as BCO2^−/−) was present in 3.4% (n = 28) of the F₂ population. The AG and GG genotypes (subsequently referred to as BCO2^−/+ and BCO2^+/+, respectively) were present in 32.8% (n = 269) and 63.8% (n = 523), respectively, of the F₂ population.The effect of the premature stop codon on milk β-carotene content was striking. BCO2^−/− cows produced milk with 78 and 55% more β-carotene than homozygous (GG) and heterozygous (AG) wild-type animals, respectively (P < 0.0001; Figure 2A). Consequently, the yellow color of the milk fat varied greatly (Figure 2B). The genotype effect on milk β-carotene content was similar at the other two time points measured during lactation (78 and 68% more β-carotene in milk from BCO2^−/− cows compared to BCO2^+/+ cows; data not shown).Open in a separate windowFigure 2.—Effect of BCO2 genotype on milk β-carotene content. (A) The mean concentration of β-carotene in the milk fat of BCO2^−/−, BCO2^−/+, and BCO2^+/+ cows is shown. β-Carotene was measured by absorbance at 450 nm as previously described (Winkelman et al. 1999). Data are means ± SEM. The statistical significance was determined using ANOVA (***P < 0.0001; n = 651). (B) The effect of the BCO2 genotype on milk fat color is illustrated.No adverse developmental or health affects as a result of the A allele were observed at any stage throughout the lifespan of the animals. The BCO2^−/− cows were fertile and milk yield was normal throughout lactation. Interestingly, quantitative real-time PCR showed fourfold lower levels of the BCO2 mRNA in liver tissue from BCO2^−/− cows (data not shown).β-Carotene and vitamin A (retinol) concentrations were also measured in serum, liver, and adipose tissue samples, and vitamin A concentration was measured in milk samples from 14 F₂ cows of each genotype. Serum β-carotene concentration was higher in BCO2^−/− cows compared to the heterozygous and homozygous wild-type cows (P = 0.003; Figure 3A). Thus, the effect of the mutation on β-carotene concentration was similar for both milk and serum, showing that this effect was not confined to the mammary gland. Vitamin A concentration was higher in serum from BCO2^−/− cows (P = 0.001; Figure 3B); however, the concentration did not differ in milk (13.1 μg/g fat vs. 14.1 μg/g fat for BCO2^−/− and BCO2^+/+ cows, respectively; P > 0.1). Liver β-carotene concentration did not differ between genotype groups (Figure 3C), but liver vitamin A was lower in BCO2^−/− cows compared to BCO2^+/+ cows (P < 0.03; Figure 3D). β-Carotene and vitamin A concentration did not differ between the genotype groups in adipose tissue (data not shown), suggesting tissue-specific effects of the BCO2 enzyme.Open in a separate windowFigure 3.—Effect of the BCO2 genotypes on concentration of β-carotene (A and C), and retinol (B and D), in serum (A and B), and liver (C and D). Subcutaneous adipose tissue biopsies (∼500 mg tissue), liver biopsies (∼100 mg tissue), and serum samples (10 ml) were taken from a subset of 42 cows (14 animals each BCO2^−/−, BCO2^−/+, and BCO2^+/+ genotypes). β-Carotene and retinol measurements were determined using HPLC with commercial standards, on the basis of a published method (Hulshof et al. 2006). Data shown are means ± SEM. Significant differences are indicated by asterisks (*P < 0.05; **P < 0.01; ANOVA, n = 14 per genotype).While previous studies have shown a key role for β-carotene 15, 15′ monooxygenase (BCMO1) in catalyzing the symmetrical cleavage of β-carotene to vitamin A (von Lintig and Vogt 2000; von Lintig et al. 2001; Hessel et al. 2007) similar evidence for the role of the BCO2 enzyme in β-carotene metabolism is lacking. The physiological relevance of BCO2 has therefore been a topic of debate (Wolf 1995; Lakshman 2004; Wyss 2004). BCO2 mRNA and protein have been detected in several human tissues (Lindqvist et al. 2005), and the in vitro cleavage of β-carotene to vitamin A has been demonstrated (Kiefer et al. 2001; Hu et al. 2006). Our results provide in vivo evidence for BCO2-mediated conversion of β-carotene to vitamin A. BCO2^−/− cows had more β-carotene in serum and milk and less vitamin A in liver, the main storage site for this vitamin.Our results show that a simple genetic test will allow the selection of cows for milk β-carotene content. Thus, milk fat color may be increased or decreased for specific industrial applications. Market preference for milk fat color varies across the world. Further, β-carotene enriched dairy foods may assuage vitamin A deficiency. Milk may be an ideal food for delivery of β-carotene, which is fat soluble and most efficiently absorbed in the presence of a fat component (Ribaya-Mercado 2002).In conclusion, we have discovered a naturally occurring premature stop codon in the bovine BCO2 gene strongly suggesting a key role of BCO2 in β-carotene metabolism. This discovery has industrial applications in the selection of cows producing milks with β-carotene content optimized for specific dairy products or to address a widespread dietary deficiency. More speculatively, it would be interesting to investigate possible effects of BCO2 variation in humans on the etiology of vitamin A deficiency.     

The yoga of Rag GTPases: Dynamic structural poses confer amino acid sensing by mTORC1

Heterodimeric Rag GTPases play a critical role in relaying fluctuating levels of cellular amino acids to the sensor mechanistic target of rapamycin complex 1. Important mechanistic questions remain unresolved, however, regarding how guanine nucleotide binding enables Rag GTPases to transition dynamically between distinct yoga-like structural poses that control activation state. Egri and Shen identified a critical interdomain hydrogen bond within RagA and RagC that stabilizes their GDP-bound states. They demonstrate that this long-distance interaction controls Rag structure and function to confer appropriate amino acid sensing by mechanistic target of rapamycin complex 1.

Mechanistic target of rapamycin complex 1 (mTORC1) integrates diverse cellular cues to promote cell growth and proliferation (1, 2). Sufficient levels of nutrients such as amino acids are required for growth factors and hormones (e.g., IGF-1 and insulin) to activate mTORC1 via PI3K, Akt, Ras homolog enriched in the brain (Rheb) (a small GTPase), and tuberous sclerosis complex (a GTPase-activating protein for Rheb) (Fig. 1A). mTORC1 signaling in turn drives anabolic (e.g., protein synthesis) and suppresses catabolic (e.g., autophagy) cellular processes. Evolutionarily conserved Rag GTPases play a critical role in amino acid sensing by mTORC1 (3, 4). Despite advances in understanding Rag structure and function, important mechanistic questions remain regarding how dynamic structural states of Rag proteins controlled by guanine nucleotide binding confer amino acid sensing by mTORC1. Egri and Shen used elegant kinetic and cell-based methods to quantitatively dissect dynamic structural elements within Rag subunits that enable mTORC1 to respond to fluctuating levels of amino acids appropriately and rapidly (5).Open in a separate windowFigure 1mTORC1 activation by growth factors (GFs) requires sufficient levels of amino acids (AAs). GFs and hormones (e.g., IGF-1; insulin) signal through PI3K, Akt, and TSC and activate Rheb through increased GTP loading (A). AAs drive Rag heterodimers toward a RagA/B^GTP–RagC/D^GDP “on” state; conversely, AA deprivation induces a switch toward a RagA/B^GDP–RagC/D^GTP “off” state. In the “on” state, Rag heterodimers bind to and recruit mTORC1 to the surface of lysosomes, where Rheb resides. Therefore, AAs and GFs activate mTORC1 cooperatively because of an induced proximity mechanism mediated by Rags and Rheb. A critical hydrogen bond (blue bar) between the NBD and CRD of RagA or RagC plays a critical role in maintaining the two stable “on” and “off” states (B). CRD, C-terminal roadblock domain; mTORC1, mechanistic target of rapamycin complex 1; NBD, nucleotide-binding domain; Rheb, Ras homolog enriched in the brain; TSC, tuberous sclerosis complex.Rag proteins function as obligate heterodimers, whereby mammalian RagA or RagB dimerizes with RagC or RagD. Rag proteins localize to lysosomal membranes by tethering to the LAMTOR/Ragulator complex (Fig. 1A) (6). In the active RagA/B^GTP–RagC/D^GDP state formed in amino acid–replete conditions, the Rag heterodimer recruits mTORC1 to the lysosomal surface through direct binding (6). Such recruitment enables Rheb to associate with and activate mTORC1 by an induced proximity mechanism (7). Upon amino acid withdrawal, GTP on RagA/B hydrolyzes to GDP, and GTP exchanges for GDP on RagC/D. This inactive RagA/B^GDP–RagC/D^GTP heterodimer releases mTORC1 into the cytosol. Thus, Rags function as dynamic molecular switches that control mTORC1 signaling in accordance with amino acid levels.Prior work (8) demonstrated that the two GTPase subunits of the Rag heterodimer (RagA/B and RagC/D) communicate with each other. GTP binding to one subunit limits binding of GTP to the other subunit and increases GTP hydrolysis if binding were to occur, and vice versa. Such intersubunit crosstalk prevents dual GTP loading, thus maintaining an opposite guanine nucleotide–loaded state and driving Rag heterodimers into two stable “on” or “off” states. The crystal structure of Rag heterodimers from budding yeast bound to GDP or GTP provided important structural information regarding how guanine nucleotide binding controls Rag architecture (9, 10). An individual Rag subunit consists of a nucleotide-binding domain (NBD) and a C-terminal roadblock domain (CRD) that mediates heterodimerization. In the GDP-bound state, the switch I domain within the NBD forms an alpha helix that orients toward the CRD; in the GTP-bound state, the switch I domain swings upward to the top of the nucleotide-binding pocket, away from the CRD. From the yeast Rag crystal structures (9, 10), Egri and Shen predicted that in the GDP- but not GTP-bound state, the hydroxyl group of Ser266 in the RagC CRB forms hydrogen bonds with Lys84 in the switch I alpha helix of the RagC NBD. As RagA Thr210 is analogous to RagC Ser266, they also predicted that Thr210 in the RagA CRB forms hydrogen bonds with Asn30 in the NBD. In the GTP-bound state, the switch I domain swings up and away from the CRD, preventing formation of these hydrogen bonds (Fig. 1B).Egri and Shen coupled these predictions with elegant quantitative kinetic in vitro assays of guanine nucleotide loading and GTP hydrolysis to demonstrate that a critical interdomain interaction in RagA and RagC maintains an opposite nucleotide-loading state in heterodimers and regulates mTORC1 activity (5). They first mutated RagA Thr210 and RagC Ser266 to Ala to abrogate the hydrogen bond and then biochemically purified WT and mutant Rag heterodimers. Ablation of the hydrogen bond had no effect on guanine nucleotide binding. When only one GTP was bound to the heterodimer, rates of GTP hydrolysis were similar on WT and mutant Rag heterodimers. When both Rag subunits of the heterodimer were forced to bind GTP, WT heterodimers displayed an increased rate of GTP hydrolysis compared with those loaded with a single GTP, indicating that the heterodimer actively resolves the dual GTP problem by hydrolyzing GTP on one subunit, consistent with prior work (8). GTP hydrolysis was increased even more for the RagA(T210A)–RagC and RagA–RagC(S266A) mutant heterodimers, suggesting that the mutations mimic a constitutive GTP-loaded conformation, driving faster GTP hydrolysis on the other subunit. In WT heterodimers, preloading the first subunit with GTP increased GTP hydrolysis on the other subunit relative to preloading with GDP. Interestingly, radioactive GTP hydrolysis in mutant heterodimers was strikingly faster than that of the WT when preloaded with either GTP or GDP, indicating that the RagA(T210) and RagC(S266A) mutations shift the heterodimer toward the GTP-loaded conformation. These results suggest that the hydrogen bond stabilizes the GDP-loaded state, and in its absence, Rag proteins tend to adopt a GTP-bound conformation even when bound to GDP, which accelerates GTP hydrolysis on the other subunit.Egri and Shen also investigated the functional significance of the RagA and RagC hydrogen bond in the control of mTORC1 signaling. Coimmunoprecipitation experiments and analysis of mTORC1 signaling to its well-established substrate S6K1 in intact cells demonstrated that the RagA(T210A)–RagC mutant associated with and activated mTORC1 inappropriately in the absence of amino acids. Upon amino acid stimulation, the RagA–RagC(S266A) mutant displayed reduced mTORC1 binding and failed to activate mTORC1 signaling. These results are consistent with RagA(T210A) mimicking a RagA^GTP “on” state and RagC(S266A) mimicking a RagC^GTP “off” state. Taken together, these results reveal the functional significance of the RagA and RagC interdomain hydrogen bond, demonstrating that it plays a critical role in regulation of mTORC1 signaling in accordance with amino acid levels.Mechanistic understanding of Rag heterodimer asanas (i.e., postures and poses) will improve our understanding of the role of mTORC1 in tumorigenesis and metabolism. For example, cancer-associated mutations have been identified in RagC, which increase mTORC1 binding (2). In addition, the physiologic importance of Rag proteins in metabolic control was demonstrated in mice engineered with an active RagA knock-in allele conferring constitutive GTP loading. These mice die perinatally, as they are unable to suppress mTORC1 signaling appropriately upon severance of the placental nutrient supply at birth. These mice fail to suppress energy expenditure, fail to induce autophagy and liberate amino acids as substrates for gluconeogenesis, and consequently fail to upregulate hepatic glucose production, responses essential for survival during fasting, unlike WT neonates (2). Thus, Rag GTPases play critical roles in cell and organismal physiology. Moving forward, deeper mechanistic insight into the yoga of Rag GTPases will improve our understanding of nutrient sensing, how its aberrant regulation contributes to a host of diseases such as cancer, obesity, and type II diabetes, and how its therapeutic targeting could treat these disorders. Namaste.     

Multi-state Modeling of Biomolecules

Multi-state modeling of biomolecules refers to a series of techniques used to represent and compute the behavior of biological molecules or complexes that can adopt a large number of possible functional states. Biological signaling systems often rely on complexes of biological macromolecules that can undergo several functionally significant modifications that are mutually compatible. Thus, they can exist in a very large number of functionally different states. Modeling such multi-state systems poses two problems: the problem of how to describe and specify a multi-state system (the “specification problem”) and the problem of how to use a computer to simulate the progress of the system over time (the “computation problem”). To address the specification problem, modelers have in recent years moved away from explicit specification of all possible states and towards rule-based formalisms that allow for implicit model specification, including the κ-calculus [1], BioNetGen [2]–[5], the Allosteric Network Compiler [6], and others [7], [8]. To tackle the computation problem, they have turned to particle-based methods that have in many cases proved more computationally efficient than population-based methods based on ordinary differential equations, partial differential equations, or the Gillespie stochastic simulation algorithm [9], [10]. Given current computing technology, particle-based methods are sometimes the only possible option. Particle-based simulators fall into two further categories: nonspatial simulators, such as StochSim [11], DYNSTOC [12], RuleMonkey [9], [13], and the Network-Free Stochastic Simulator (NFSim) [14], and spatial simulators, including Meredys [15], SRSim [16], [17], and MCell [18]–[20]. Modelers can thus choose from a variety of tools, the best choice depending on the particular problem. Development of faster and more powerful methods is ongoing, promising the ability to simulate ever more complex signaling processes in the future.

This is a “Topic Page” article for PLOS Computational Biology.

     

Coaligned Dual-Channel STED Nanoscopy and Molecular Diffusion Analysis at 20 nm Resolution

We report on a fiber laser-based stimulated emission-depletion microscope providing down to ∼20 nm resolution in raw data images as well as 15–19 nm diameter probing areas in fluorescence correlation spectroscopy. Stimulated emission depletion pulses of nanosecond duration and 775 nm wavelength are used to silence two fluorophores simultaneously, ensuring offset-free colocalization analysis. The versatility of this superresolution method is exemplified by revealing the octameric arrangement of Xenopus nuclear pore complexes and by quantifying the diffusion of labeled lipid molecules in artificial and living cell membranes.Since its first demonstration in (live) cell imaging (1), stimulated emission depletion (STED) fluorescence microscopy has been realized in many variants. Particularly, the key phenomenon employed in this method, namely switching fluorophores transiently off by stimulated emission, has been accomplished with laser pulses varying from picoseconds to nanoseconds in duration, and from kHz to MHz in repetition rate. Because continuous-wave beams are suitable as well (2), STED microscopy has been implemented with rather different laser systems, ranging from model-locked femtosecond to continuous-wave laser diodes (3,4). Although it underscores the versatility of STED to modulate the fluorescence capability of a fluorophore, this wide range of options may confuse adopters when balancing simplicity, applicability, and resolution gain. The situation is exacerbated when implementing pairs of excitation and STED beams for dual-color colocalization studies (5,6).Here we report on a simple arrangement providing dual-color STED nanoscopy (Fig. 1) and molecular diffusion quantification down to ∼20 nm in (living) cells. The presented dual-channel STED microscope utilizes a single fiber laser providing a 20-MHz train of 775 nm wavelength pulses of 1.2-ns duration. This compact laser source enables STED on fluorophores emitting in the orange to red range. Specifically, we applied this laser on the orange dyes Atto590 and Atto594 (excitation: 595 nm; detection: 620 ± 20 nm), and the red dyes KK114 and Abberior Star635P (excitation: 640 nm; detection: 670 ± 20 nm). Although the spectra of the dyes are partially overlapping, the individual color channels can be separated without data processing (see Fig. S1 and Fig. S2 in the Supporting Material). Both channels are recorded simultaneously within 50 ns, using temporally interleaved pulsed excitation in combination with time-gated detection (5,7,8).Open in a separate windowFigure 1Fluorescence nanoscopy of protein complexes with a compact near-infrared nanosecond-pulsed STED microscope. (A) STED reveals immunolabeled subunits in amphibian NPC; raw data smoothed with a Gaussian filter extending over 14 nm in FWHM. The diameter of the octameric gp210 ring is established as ∼160 nm. Scale bar, 500 nm. (B) Individual NPC image showing eight antibody-labeled gp210 homodimers as 20–40 nm sized units and a 80 nm-sized localization of the subunits in the central channel.Because in STED microscopy, the STED doughnuts firmly determine the position of the fluorescently active molecules, the use of a single doughnut for both fluorophores guarantees that the two color channels are almost perfectly coaligned. The use of the doughnut even counteracts misalignments of the confocal excitation and detection channels (Fig. 2, and see Fig. S3), making STED microscopy particularly powerful for colocalization measurements.Open in a separate windowFigure 2Determination of the colocalization accuracy. Xenopus A6 cells, labeled with an antiserum against multiple NUP subunits in the central NPC channel and two secondary antibodies decorated with the fluorophores Abberior STAR635P and Atto594 were imaged by STED microscopy. (A) Upon overlaying both channels, a high degree of colocalization is directly visible. Scale bar, 200 nm. (B) Quantification of the colocalization by cross correlation of much larger images (see Fig. S3). The correlation is maximal for zero displacement of the images, proving colocalization. (C) Confocal image of monocolored fluorescent beads taken with improperly coaligned excitation beams (left). Improper coalignment spoils the colocalization accuracy in confocal imaging; the two channels should be perfectly coaligned, but they show a false offset as indicated by the color difference. The offset is quantified by the cross correlation of the two channels (right). (D) The STED image of the same beads (left) not only shows 10-fold improved resolution over the confocal image in panel C, but also improved colocalization, again quantified by cross correlation (right). Thus, by predetermining the position of emission, the STED doughnut counteracts errors induced by imperfect coalignment of the two confocal color channels (for details, see Fig. S3). Scale bars = 100 nm.The cross section for stimulated emission is lower at 775 nm as compared to that found at somewhat shorter wavelengths (5), yet STED pulse energies of ∼7 nJ in the focus are sufficient to yield a resolution of ∼30 nm and ∼20 nm in the orange and red channels, respectively (see Fig. S4). In addition, due to the lower peak intensity, the 1.2 ns pulses are likely to induce less nonlinear absorption and hence less photostress as compared to their more commonly used <0.2 ns counterparts (8,9). On the other hand, the pulses are only 2–4 times shorter than the typical lifetime of the excited state, which lessens their STED efficiency. This slight reduction is neutralized here by detecting photons emitted ∼1 ns after excitation (5,7,8).The potential of this straightforward implementation of STED microscopy is evident when imaging immunolabeled nuclear pore complexes (NPCs) of cultured Xenopus cells. Contrary to the confocal recording, STED microscopy reveals subunits of this protein complex, specifically the typical eightfold symmetry of its peripheral transmembrane protein gp210, along with a set of proteins in the central pore channel (Fig. 1, and see Fig. S5 and Fig. S6). Unlike in stochastic superresolution imaging of gp210 (10), the color channels are inherently coaligned and simultaneously recorded simply by executing a single scan. Apart from a weak smoothing and background subtraction applied to enhance image contrast, the images are raw.Because fluorescence off-switching by STED is an instant process, STED microscopy can be employed to study fast spatial translocations, such as the diffusion of molecules on the nanoscale (3). To benchmark the performance of our setup, we analyzed the diffusion of a fluorescent glycerophospholipid analog (11) by fluorescence correlation spectroscopy (FCS) in membranes of living mammalian PtK2-cells (Fig. 3). STED allowed us to reduce the diameter of the probed area from the 250 nm-sized diffraction limit down to 19 nm (FWHM), representing σ = 8 nm in standard deviation of a Gaussian fit. The attained subdiffraction area is 2.5 times smaller as compared to what has been reported in living cells to date (4). In model membranes, the smallest diameter was 15 nm (σ = 6.4 nm).Open in a separate windowFigure 3Nanoscale molecular diffusion analyzed by STED FCS. (A) For moderate and larger STED beam power P_STED, the resolution scales inversely with its square-root, attaining 15 nm in FWHM of the distribution of fluorescence emission in space, describing the measurement area. Note the relatively small threshold power P_S = 1.4 mW, which implies that a large resolution gain is already attained for P_STED < 100 mW. (Inset) The resolution was determined by measuring the transit time of a fluorescent phospholipid-analog (DSPE-PEG-KK114) in a lipid model membrane through the detection area by FCS. (B) In living mammalian Ptk2-cells, the transit time of the lipid analog scales linearly with the detection area, revealing a diffusion constant D_lat = 0.33 μm²/s, and showing that this lipid analog diffuses largely freely in the plasma membrane down to <20 nm scales.In both measurements, the molecular transit time depends linearly on the probed area, indicating that the labeled lipid molecules diffuse essentially freely down to spatial scales of 20 nm. Accordingly, the anomaly exponent α was close to 1 with values of α > 0.85, showing only minor deviations from free diffusion (see Fig. S7). Because the diameter is inversely proportional to the square-root of the STED beam power, the resolution can be adapted to a particular application need (Fig. 3, A and B).In summary, our arrangement provides up-to-date STED microscopy resolution in offset-free colocalization recordings. The ready-to-use near-infrared laser pulses keep undesired single and multiphoton absorption low and leave the visible spectrum amenable for further studies.     

Rainy weather and medical school admission interviews

Mood can influence behaviour and consumer choice in diverse settings. We found that such cognitive influences extend to candidate admission interviews at a Canadian medical school. We suggest that an awareness of this fallibility might lead to more reasonable medical school admission practices.Admission offers to medical school are competitive and sometimes based on an interview. Psychology research suggests, however, that interviews are prone to subconscious biases from extraneous factors unrelated to the candidate.¹ One of the most fundamental observations is that people interviewed on rainy days tend to receive lower ratings than people interviewed on sunny days.² We studied whether this bias also extends to admission interviews at a large Canadian medical school.We analyzed the results of consecutive medical school interviews at the University of Toronto between 2004 and 2009. We included all data available with no exclusions. Almost all interviews occurred in the early spring. Scores for each interview were obtained from the admissions office as recorded from 0 to 20.³ This Likert scale was anchored with integer values where 10 denoted “unsuitable,” 12 denoted “marginal,” 14 denoted “fair,” 16 denoted “good,” 18 denoted “excellent” and 20 denoted “outstanding.”We obtained weather data from the official government archive and defined a priori the day as “rainy” if precipitation (including freezing rain, snow and hail) occurred in the morning or afternoon.⁴ Otherwise, we defined the day as “sunny.” We did not examine more complex combinations with time lags, such as when a sunny day followed multiple rainy days.A total of 2926 candidates were interviewed over the 6-year period. As expected, their demographic characteristics were unrelated to the weather (Appendix 1, available online at www.cmaj.ca/cgi/content/full/cmaj.091546/DC1). Overall, those interviewed on rainy days received about a 1% lower score than those interviewed on sunny days (average score 16.31 v. 16.49, p = 0.042). This pattern was consistent for both senior interviewers (16.39 v. 16.55, p = 0.08) and junior interviewers (16.23 v. 16.42, p = 0.041). We next used logistic regression to analyze subsequent admission decisions. The difference in scores was equivalent to about a 10% lower total mark on the Medical College Admission Test.Open in a separate windowWe suggest that cognitive patterns evident in controlled psychology laboratories can also occur in regular medical settings. The magnitude of the specific influence may be modest, but such small differences can be important in some cases because each year there are about 100 candidates who receive a score within 1% of the admission threshold.⁵ In this study, we examined only one extraneous influence on mood. Many additional factors may also affect mood (e.g., ambiance, deportment, humour and scent).² Calling attention to these issues may diminish their impact on judgment.¹     

CircHIPK3 sponges miR‐558 to suppress heparanase expression in bladder cancer cells

Correction to: EMBO Reports (2017) 18(9): 1646–1659. DOI: 10.15252/embr.201643581 ¦ Published online 9 August 2017The authors contacted the journal after being alerted to issues in the figures. The authors state that while preparing the figures, images were mislabelled leading to partial duplications in two figure panels. The authors requested to withdraw the affected panels and to replace them with correct representative images that had been generated at the time of the original experimentation. The panels listed below are therefore withdrawn and replaced. The related source data are published with this note.Figure 4DThe transwell assay image for UMUC3 cells showing invasion behaviour upon miR‐558 mimic treatment (“miR‐558”) had been incorrect. An image showing the invasion behaviour of UMUC3 cells upon depletion of circHIPK3 (“si‐circHIPK3#1”) showing the same cells as depicted in Fig 2H was erroneously used. A representative image of the correct data is now displayed in the paper.Figure 4EThe transwell assay image for UMUC3 cells showing migration behaviour upon treatment with an miR‐588 anti‐miR (“anti‐miR‐558”) had been incorrect. An image showing the migration behaviour of UMUC3 cells upon circHIPK3 overexpression (“circHIPK3”) showing the same cells as those depicted in Fig 2D was erroneously used. A representative image of the correct data is now displayed in the paper.Figure 5CThe Western blot image showing the β‐actin loading control for T24T cells had been incorrect. A representative image of the correct data is now displayed in the paper.Figure 5FThe image for UMUC3 cells showing tube formation upon treatment with a control mimic and overexpression of circHIPK3 “mimicNC+circHIPK3” had been incorrect. A representative image of the correct data is now displayed in the paper.The figure issues described above are herewith corrected. The authors state that the errors do not affect the results or conclusions of the study and apologize for any confusion these errors may have caused. Figure 4D. Original. Figure 4D. Corrected. Figure 4E. Original. Figure 4E. Corrected. Figure 5C. Original. Figure 5C. Corrected. Figure 5F. Original. Figure 5F. Corrected.      

Structural diversity of ABC transporters

ATP-binding cassette (ABC) transporters form a large superfamily of ATP-dependent protein complexes that mediate transport of a vast array of substrates across membranes. The 14 currently available structures of ABC transporters have greatly advanced insight into the transport mechanism and revealed a tremendous structural diversity. Whereas the domains that hydrolyze ATP are structurally related in all ABC transporters, the membrane-embedded domains, where the substrates are translocated, adopt four different unrelated folds. Here, we review the structural characteristics of ABC transporters and discuss the implications of this structural diversity for mechanistic diversity.ATP-binding cassette (ABC) transporters are a large group of membrane protein complexes that couple transport of a substrate across the membrane to the hydrolysis of the phosphate bond between the γ- and the β-phosphate of ATP (Ames et al., 1990; Higgins, 1992; Davidson et al., 2008; Rees et al., 2009). The free energy released when ATP is converted into ADP and orthophosphate (P_i; approximately −50 kJ mol⁻¹ in many cells) can be used to accumulate the transported substrates in, or to remove them from, cellular compartments.In prokaryotes, ABC transporters are localized to the plasma membrane, and ATP is hydrolyzed on the cytoplasmic side. In eukaryotes, ABC transporters are also found in organellar membranes. ATP hydrolysis by organellar ABC transporters takes place on the cytosolic side of the membrane, except for transporters from mitochondria and chloroplasts where the ATP-binding domains of the transporters are located on the matrix or stroma side. The side of the membrane where ATP is bound and hydrolyzed is termed the cis-side, and the opposite side is called the trans-side.ABC transporters can be classified as exporters or importers. Exporters move substrates from the cis-side to the trans-side of the membrane, from the hydrophobic core of the lipid bilayer to the trans-side, or transfer substrates between the inner and outer leaflets of the bilayer. In contrast, importers move substrates from the trans-side to the cis-side. There are a few ABC transporters that do not have a bona fide transport function. Notable examples include the CFTR, which is a gated chloride channel (Aleksandrov et al., 2007), and the sulfonylurea receptor SUR1, which is a regulatory complex associated with a potassium channel (Bryan et al., 2007). There has been tremendous interest in ABC transporters not only from a mechanistic point of view but also because malfunctioning of human ABC transporters leads to a plethora of diseases (see for instance, Silverton et al., 2011), and some ABC exporters are involved in the drug resistance of bacteria and cancer cells (Ambudkar et al., 2003; Davidson et al., 2008).Many excellent reviews on ABC transporters have been published over the past few years and cover the history, structure, mechanism, physiology, and pharmacology of these proteins (Davidson et al., 2008; Rees et al., 2009; Parcej and Tampé, 2010; Eitinger et al., 2011; George and Jones, 2012; Lewis et al., 2012). Because a wealth of new crystal structures has been determined lately, here we provide an update on the structural diversity of ABC transporters.

Overview

Similar to other membrane transport proteins, ABC transporters adopt at least two conformations in which the substrate-binding site is accessible from either the cis-side or the trans-side. Alternation between the two conformations allows substrate translocation across the membrane (“alternating access” model; Jardetzky, 1966; Tanford, 1982). The binding of substrate on one side of the membrane and release on the opposite side are coordinated by ATP binding and hydrolysis, and ADP and P_i release. Several ABC transporters have been crystallized in different conformations (see

Functional Genetics in the Post-Genomics Era: Building a Better Roadmap in Drosophila

In this commentary, Rob Kulathinal describes two articles from the Perrimon lab, each describing a new online resource that can assist geneticists with the design of their RNA interference (RNAi) experiments. Hu et al.’s “UP-TORR: online tool for accurate and up-to-date annotation of RNAi reagents” and “FlyPrimerBank: An online database for Drosophila melanogaster gene expression analysis and knockdown evaluation of RNAi reagents” are published, respectively, in this month’s issues of GENETICS and G3.     

FORMIDABEL: The Belgian Ants Database

FORMIDABEL is a database of Belgian Ants containing more than 27.000 occurrence records. These records originate from collections, field sampling and literature. The database gives information on 76 native and 9 introduced ant species found in Belgium. The collection records originated mainly from the ants collection in Royal Belgian Institute of Natural Sciences (RBINS), the ‘Gaspar’ Ants collection in Gembloux and the zoological collection of the University of Liège (ULG). The oldest occurrences date back from May 1866, the most recent refer to August 2012. FORMIDABEL is a work in progress and the database is updated twice a year.The latest version of the dataset is publicly and freely accessible through this url: http://ipt.biodiversity.be/resource.do?r=formidabel. The dataset is also retrievable via the GBIF data portal through this link: http://data.gbif.org/datasets/resource/14697A dedicated geo-portal, developed by the Belgian Biodiversity Platform is accessible at: http://www.formicidae-atlas.bePurpose: FORMIDABEL is a joint cooperation of the Flemish ants working group “Polyergus” (http://formicidae.be) and the Wallonian ants working group “FourmisWalBru” (http://fourmiswalbru.be). The original database was created in 2002 in the context of the preliminary red data book of Flemish Ants (Dekoninck et al. 2003). Later, in 2005, data from the Southern part of Belgium; Wallonia and Brussels were added. In 2012 this dataset was again updated for the creation of the first Belgian Ants Atlas (Figure 1) (Dekoninck et al. 2012). The main purpose of this atlas was to generate maps for all outdoor-living ant species in Belgium using an overlay of the standard Belgian ecoregions. By using this overlay for most species, we can discern a clear and often restricted distribution pattern in Belgium, mainly based on vegetation and soil types.Open in a separate windowFigure 1.www.formicidae-atlas.be