On the neurocognitive origins of human tool use : A critical review of neuroimaging data

Highlights

• Neuroimaging data from the past decade on tool use brain circuitry is re-analyzed.

• Results stress the central role of the left inferior parietal cortex.

• We propose an innovative theoretical approach to the tool use brain network.

• Left inferior parietal cortex might support reasoning on physical object properties.

Abstract

Since more than a century, neuropsychological models have assumed that the left inferior parietal cortex is central to tool use by storing manipulation knowledge (the manipulation-based approach). Interestingly, recent neuropsychological evidence indicates that the left inferior parietal cortex might rather support the ability to reason about physical object properties (the reasoning-based approach). Historically, these two approaches have been developed from data obtained in left brain-damaged patients. This review is the first one to (1) give an overview of the two aforementioned approaches and (2) reanalyze functional neuroimaging data of the past decade to examine their predictions. Globally, we demonstrate that the left inferior parietal cortex is involved in the understanding of tool-use actions, providing support for the reasoning-based approach. We also discuss the functional involvement of the different regions of the tool-use brain network (left supramarginal gyrus, left intraparietal sulcus, left posterior temporal cortex). Our findings open promising avenues for future research on the neurocognitive basis of human tool use.

Introduction

Since the appearance of symbolic reasoning (e.g., language skills) in human lineage, Man has largely overlooked the physical reasoning skills involved in tool use. A good illustration of this lack of attention is the implicit hierarchy of intellectual work over manual work, as if tool use did not require any kind of specific intelligence or reasoning. This belief is also deeply ingrained in the minds of scholars and scientists alike. For instance, since more than a century, neuropsychological models have assumed that knowledge about manipulation is central to tool use (the manipulation-based approach). This knowledge is supposed to be stored within the left inferior parietal lobe (IPL). By contrast, a more recent perspective, initiated by Goldenberg in the 2000’s, assumes that, to use tools, people reason about the physical object properties in order to generate mechanical actions (the reasoning-based approach). This reasoning would be supported by mechanical knowledge located in the left IPL. Historically, these two approaches have been developed from data obtained in left brain-damaged patients. The goal of the present review is to examine their validity by reanalyzing functional neuroimaging studies of the past decade. To anticipate our conclusions, we demonstrate that the left IPL is clearly involved in the understanding of tool-use actions, providing support for the reasoning-based approach. From our findings, we shall also discuss the functional involvement of the different regions of the tool-use brain network.

Tool use is considered by many to be a hallmark of complex cognitive adaptations (Beck, 1980, Johnson-Frey, 2004). So, understanding the neurocognitive bases of human tool use can provide fundamental insights into the evolution of human lineage. More than a century after the publication of Descent of Man by Darwin in 1871, the issue of the origins of human cognition is still a matter of debate between proponents of continuity versus discontinuity theories of the evolution of human cognition. Both of them are not at odds with regard to the fact that differences exist between human and nonhuman tool use. Nevertheless, they differ greatly as to the interpretations they lend to these differences. Continuity theories consider that these differences are rather quantitative than qualitative, whereas discontinuity theories assume that at least some of them can be viewed as qualitative. We will begin by presenting these theories in order to help the reader to understand how, at an epistemological level, they have framed the manipulation-based and the reasoning-based approach to human tool use. We would also like to add one caveat. Particularly, the interpretations based on neuroimaging studies will be discussed parsimoniously all along the paper, notably to avoid circularity between them and the results obtained from the present meta-analysis. Rather, we decided to present how the manipulation-based and the reasoning-based approach have been developed mainly from the neuropsychological literature. In this context, neuroimaging data are considered as a good means to examine their validity.

Tool use was once considered to be one of the defining characteristics of the genus Homo, dating back 2.5 million years (Oakley, 1949; see Ambrose, 2001). The diversity of tool behavior in several nonhuman species including primates, birds, mammals, and even insects, has forced us to revise this assumption (for reviews, see Baber, 2003, Beck, 1980, Van Lawick-Goodall, 1970). In broad terms, tool use is not unique to humans. This conclusion might appear surprising given the subtle – and sometimes less subtle – differences that exist between human and nonhuman tool use (see below). Nevertheless, these differences can be masked by the way we define what is a tool. Tools are commonly defined as handheld physical objects that are manipulated in order to increase the user’s sensorimotor capabilities (for a critical review about this definition, see Osiurak et al. 2010). This definition puts a heavy emphasis on manipulation and, as a result, can include a wide range of animal behaviors to the extent that there is manipulation. In this frame, it is true that some similarities do exist between nonhuman primates¹ and humans, particularly with regard to prehension skills.

Prehension skills are based on two basic components, namely, reaching (transportation of the hand to the object by the upper limb) and, more relevant to our concerns here, grasping (shaping the hand according to object size and orientation, and applying the correct grip force). Although the nonhuman primate brain is cytoarchitectonically quite different from ours, a significant body of evidence from neurophysiology has suggested considerable functional homologies between the human and the primate brain with regard to grasping (for reviews, see Johnson-Frey and Grafton, 2003, Rizzolatti and Matelli, 2003, Vingerhoets, 2014). In monkeys the anterior intraparietal area (AIP) in combination with area F5 is involved in the transformation of 3D properties of an object into appropriate finger formations and hand orientation for visually guided grasping movements (Jeannerod et al., 1995; see also Nelissen et al., 2011). Interestingly, it has been proposed that the anterior dorsal intraparietal sulcus (DIPSA) and the putative human homologue of AIP (phAIP) together represent the equivalent of monkey AIP. DIPSA corresponds to the more visual, posterior part of AIP and phAIP to its more somatosensory and motor, anterior part (see Orban et al., 2006, Vanduffel et al., 2014). Neuroimaging data also indicate that area phAIP is activated by observing a tool being grasped (Jacobs et al., 2010, Peeters et al., 2013). Different brain areas within the ventral premotor cortex (vPMC) and phAIP are also involved in the representations of different hand movements or handgrips (e.g., precision vs. power grips; Ehrsson et al., 2000, Ehrsson et al., 2001; see also Dinstein et al., 2007; but for discussion see Sawamura et al., 2006).

To sum up, both human and nonhuman primates possess a primate prehension system involved in the elaboration of motor schemas (e.g., preshape, enclose, selection of the number of fingers) that would be potentially elicitable by the extraction of information about the object properties² (e.g., size, shape) (see also Arbib, 1985, Jeannerod et al., 1995). This system is both effector-specific and side-specific and, as a result, is distributed bilaterally (Vingerhoets, 2014). It is supposed to be the basis of non-tool-use actions, such as object transport, or even “simple tool use” as in nonhuman primates³ (Orban and Caruana, 2014).

The story could have ended here: Humans and nonhuman primates share a common neural network for prehension skills – including manipulative, motor schemas –, explaining why both can use tools. Nevertheless, beyond these similarities, it appears that human tool use differs from that known to occur in nonhumans in different ways. For instance, in nonhumans, tool use is incidental and rare in the wild (Beck, 1980, Byrne, 2004, Chappell and Kacelnik, 2002, Van Schaik et al., 1999). By contrast, humans spontaneously engage in object–object manipulations (McGrew, 1992) and employ a wide range of tools everyday and during all the life (Johnson-Frey, 2007), a feature that characterizes humans of all cultures through the ages (Leroi-Gourhan, 1971). In addition, only humans are able to use a tool to create another one (i.e., use of secondary tools; Gibson, 1993, McGrew, 1992, McGrew, 1993, Toth and Schick, 1993). A substantial body of evidence also indicates serious limitations on the ability of nonhuman animals to solve tool-use situations that are relatively simple for humans (e.g., Povinelli, 2000, Visalberghi and Limongelli, 1994) or to transfer the mechanical relationships they learn in one specific situation to other ones (Martin-Ordas et al., 2008, Penn et al., 2008). In short, even if humans are not unique in using tools, there is undoubtedly something unique about the way humans use tools.

Differences also exist at a neuroanatomical level. More specifically, it has been repeatedly assumed that in humans more ventral parts of the parietal cortex may play a central role in the skillful use of tools (Binkofski and Buxbaum, 2013, Buxbaum, 2001, Daprati and Sirigu, 2006, Johnson-Frey and Grafton, 2003, Rizzolatti and Matelli, 2003, Orban and Caruana, 2014, Vingerhoets, 2014). This system is lateralized to the left hemisphere and includes the left IPL, and particularly, the supramarginal gyrus (SMG), a new human brain area that does not exist in monkeys (Orban and Caruana, 2014, Peeters et al., 2009, Peeters et al., 2013, Vingerhoets, 2014). In sum, not only humans possess specific behavioral characteristics, but there is also a uniquely human brain area that might be the basis for this specificity. The important issue is to understand the functional role of this area. This issue has raised considerable interest in the fields of neuropsychology and neuroimaging, leading to the formulation of two different approaches, which mirror the classical distinction between continuity and discontinuity theories of the evolution of human cognition.

The theory of a continuity among species was initially formulated more than a century ago by Darwin (1871, 1981 and Romanes (1883), the two founders of comparative psychology. They assumed that there is no qualitative difference or discontinuity between nonhuman and human minds. As Darwin claimed, “the difference, great as it is, certainly is one of degree and not of kind.” (Darwin, 1871, 1981, p. 445). The core assumption is that in cases in which other species exhibit behavior similar to our own, similar cognitive causes are at work. This is the argument by analogy (see also Boesch, 2005, McGrew, 2005, Suddendorf and Whiten, 2001, Van Lawick-Goodall, 1970). To the question as to why human tool use appears to be not completely similar to that of nonhuman animals, the answers generally highlight the importance of environmental resources and pressures (e.g., prey, predators), which could have led humans to use tools more frequently for survival (Boesch and Tomasello, 1998, De Beaune, 2008, Wynn, 1993). It has also been hypothesized that differences between human and primate tool use may reflect hominin enhancements of pre-existing primate prehension systems (Ambrose, 2001, Marzke et al., 1992, Napier, 1956, Napier, 1960, Van Schaik et al., 1999). The manipulation-based approach is highly inspired by these theories, by formulating that in humans the left IPL is central to store sensorimotor knowledge about how to manipulate familiar tools skillfully (e.g., Buxbaum, 2001, Buxbaum and Kalénine, 2010, Thill et al., 2013, Van Elk, 2014). In a way, the left IPL would store the same kind of information as that contained in the primate prehension system (see above). Simply, the difference is that this knowledge is not dedicated to know how to grasp objects to move them, but rather to know how objects have to be grasped in order to use them as tools. In this approach, the main difference between nonhuman primate and human tool use is more one of degree than of kind.

Proponents of the discontinuity theory stress that the argument by analogy has unfortunately led many scientists to document the similarities and downplay the differences between human and primate behavior (Penn et al., 2008, Povinelli et al., 2000). They advocate that it is not because two behaviors are analogous that similar psychological causes are at work, and suggest that those differences are not quantitative but qualitative in that they are the consequences of psychological diversity (Penn et al., 2008, Tomasello et al., 2005). Thus, it has been argued that humans alone are able to understand observable regularities of the environment in terms of unobservable causal forces (gravity, force, shape, mass; e.g., Penn et al., 2008, Povinelli, 2000). The reasoning-based approach is akin to these theories by stressing that in humans the left IPL is critical to store mechanical knowledge necessary to reason about how tools and objects have to be used in a purposeful way (Goldenberg, 2013, Osiurak, 2014c, Osiurak et al., 2010). The corollary is that this kind of reasoning would be unique to humans. In broad terms, the main difference between nonhuman primate and human tool use is rather one of kind than of degree. In the following lines, we shall present in more detail how the neuropsychological literature has documented these two approaches.

The common definition of tool use put a special emphasis on manual actions, as if the main problem the user faces when using a tool is to know how to manipulate it but not to reason about how the tool has to interact with the object. This way of addressing the problem of the neurocognitive bases of human tool use strongly reflects the folk psychology. Since centuries, people needing to use tools to carry out everyday activities have been described as doing manual work, as if there was no need to reason when using tools, contrary to intellectual work. This belief is also profoundly ingrained in the minds of psychologists and neuroscientists alike. For instance, since the first descriptions of brain-damaged patients with tool use disorders in the late 1800’s, the difficulties to use tools have been included in the terminology of apraxia,⁴ thereby suggesting that tool use disorders are first and foremost a matter of gesture. During the last 50 years, several neuropsychological models have been formulated, wherein tool use disorders have been described within a general architecture also useful for explaining difficulties in producing symbolic gestures (e.g., waving goodbye) or imitating meaningless postures (e.g., putting the back of the hand on the front) (Buxbaum, 2001, Cubelli et al., 2000, Geschwind, 1965, Heilman et al., 1982, Heilman and Watson, 2008, Rothi et al., 1991, Roy and Square, 1985, Buxbaum and Kalénine, 2010, Thill et al., 2013, Van Elk, 2014).

In line with this approach, it has been posited that central to tool use is the storage of sensorimotor knowledge about tool manipulation (i.e., manipulation knowledge), also called visuo-kinesthetic engrams (Heilman et al., 1982), action lexicons (Rothi et al., 1991), motor engrams (Buxbaum, 2001) or motor programs for tool use skills (Johnson-Frey et al., 2005). This long-term knowledge is supposed to contain the main parameters of the gesture associated with the manipulation of a tool (e.g., the hand posture, the position of the hand in the space, the amplitude of the movement executed by elbow joints) so that it provides a processing advantage by avoiding that each gesture is reconstructed de novo with each use. As discussed, manipulation knowledge might be the basis for an additional prehension system, only present in humans and specifically devoted to tool use. This knowledge has been associated with the left IPL (Binkofski and Buxbaum, 2013, Buxbaum, 2001, Buxbaum and Kalénine, 2010, Daprati and Sirigu, 2006, Gainotti, 2013, Heilman et al., 1982, Johnson-Frey and Grafton, 2003, Rizzolatti and Matelli, 2003, Rothi et al., 1991, Van Elk, 2014). The manipulation knowledge hypothesis has also been repeatedly stressed to account for the activation of the left IPL in neuroimaging studies (e.g., Boronat et al., 2005, Buxbaum et al., 2006, Grèzes and Decety, 2002, Hermsdörfer et al., 2007, Imazu et al., 2007, Johnson-Frey et al., 2005, Kellenbach et al., 2003, Kroliczak and Frey, 2009, Rumiati et al., 2004, Vingerhoets, 2008, Vingerhoets et al., 2009). Importantly, manipulation knowledge is supposed to be associated only with the conventional use of familiar tools, because it is based on the experience we have with those tools. So, this knowledge cannot be useful to use novel tools or familiar tools in a non-conventional way. Moreover, it encodes egocentric relationships, namely, relationships between the user (and particularly his/her hand) and a tool. The manipulation knowledge hypothesis has found resonance in recent years, with the growing interest in the embodied cognition approach, suggesting that knowledge is constituted by information represented within the motor and sensory systems (Barsalou, 2008, Borghi, 2004, Borghi et al., 2013, Pezzulo et al., 2013a, Pezzulo et al., 2013b, Mizelle and Wheaton, 2010, Thill et al., 2013).

As mentioned just above, manipulation knowledge encodes egocentric relationships (i.e., user-tool). The corollary is that it cannot help people to know with which object or in which context a specific tool can be used (i.e., tool-object, allocentric relationships). The storage of long-term, allocentric relationships has been suggested to be specific to the ventral system (e.g., Goodale and Milner, 1992, Milner and Goodale, 2006). In line with this, it has been assumed that the left temporal lobe would be particularly involved in conceptual knowledge about tool function (e.g., Buxbaum, 2001, Thill et al., 2013, Van Elk, 2014; see also Hodges et al., 2000). However, the precise localization of the area concerned is still a matter of debate. Neuropsychological studies in patients with semantic dementia, herpetic encephalitis or left brain damage have indicated the involvement of anterior portions of the left temporal lobe (e.g., Bozeat et al., 2002, Buxbaum et al., 1997, Hodges et al., 1999, Hodges et al., 2000, Goldenberg and Spatt, 2009, Lauro-Grotto et al., 1997, Sirigu et al., 1991). Although some neuroimaging studies have corroborated these findings (e.g., Canessa et al., 2008), others have stressed the potential role of the left posterior temporal lobe (Ebisch et al., 2007; see also Hermsdörfer et al., 2007, Tsuda et al., 2009, Vingerhoets, 2008) or have failed to obtain any significant neural correlates (Boronat et al., 2005, Kellenbach et al., 2003).

Contrary to manipulation knowledge, the study of function knowledge in the field of human tool use has received far less attention in recent years. This might be explained by compelling evidence that function knowledge and real tool use (i.e., the actual use of a tool with its corresponding object) can be impaired independently from each other (Bartolo et al., 2007, Bozeat et al., 2002, Buxbaum et al., 1997, Forde and Humphreys, 2000, Goldenberg and Spatt, 2009, Hodges et al., 2000, Lauro-Grotto et al., 1997, Negri et al., 2007, Osiurak et al., 2008, Osiurak et al., 2009, Osiurak et al., 2011, Silveri and Ciccarelli, 2009). Given that function knowledge is neither necessary, nor sufficient for tool use, the intriguing question is what is the role of this knowledge? It has been posited that even if function knowledge is not central to tool use, it can be useful when manipulation knowledge is impaired, as a means of compensation (the multiple-routes-for-action hypothesis; e.g., Buxbaum, 2001, Buxbaum et al., 1997, Sirigu et al., 1991). The opposite is also true in that impaired function knowledge might be compensated by intact manipulation knowledge.

Finally, the manipulation-based approach assumes the existence of a production system that involves bilateral dorsal structures (particularly both superior parietal lobes [SPL] and IPS) and is specialized for acquiring objects on the basis of visual information about object shape, size and location that is constantly updated in function of the positions of objects with respect to retina, eye, head, torso, limb and hand (Buxbaum and Kalénine, 2010; see also Binkofski and Buxbaum, 2013, Buxbaum, 2001, Buxbaum et al., 2000, Heilman et al., 1986, Thill et al., 2013, Van Elk, 2014). In a way, the production system is very close to the primate prehension system mentioned above. Like manipulation knowledge, the production system encodes egocentric relationships between the tool and the agent. It is also thought to support non-tool-use actions. In addition, even if it does not contain any long-term information about tool-use skills, it can receive input from the left IPL in order to adapt the tool-use representation created from manipulation knowledge to the situational constraints.

The critical role given to manipulation in tool use by the manipulation-based approach may appear quite surprising for developmental psychologists, for whom tool use is viewed as an instance of problem-solving situation supporting by mechanical reasoning skills (e.g., Beck et al., 2011, Mounoud, 1996). The same is true in the field of animal cognition, wherein animal tool users have been shown to fail to solve tool-use situations that are relatively simple for young children (see above; e.g., Povinelli, 2000, Visalberghi and Limongelli, 1994). In this field, too, reasoning is viewed as central to tool behavior (Penn et al., 2008, Penn and Povinelli, 2007; but see also Wolpert, 2003). The reasoning-based approach has been elaborated in line with this perspective, mainly in the light of studies in left brain-damaged patients with tool use disorders (Goldenberg, 2009, Goldenberg, 2013, Goldenberg and Hagmann, 1998, Goldenberg and Spatt, 2009, Osiurak, 2013, Osiurak, 2014a, Osiurak, 2014c, Osiurak et al., 2009, Osiurak et al., 2010, Osiurak et al., 2011, Osiurak et al., 2013). The Four Constraints theory (Osiurak, 2014c) and the dialectical theory of human tool use (Osiurak et al., 2010; see also Osiurak and Badets, in press) correspond to the most recent versions of this approach. Orban and Caruana (2014) recently proposed a more precise neural model of this approach. Globally, this approach is based on the core assumption that, in everyday life, people use tools to solve problems (e.g., preparing a meal). To do so, they have to use mechanical knowledge to reason about how to solve them.

The reasoning-based approach has been developed because of theoretical and empirical limitations inherent to the manipulation-based approach. First, the multiple-routes-for-action hypothesis associated with the manipulation-based approach is questionable (see Osiurak, 2014c, Osiurak et al., 2010, Osiurak et al., 2011). According to this hypothesis, a patient with impaired manipulation knowledge is still able to select and use tools appropriately because function knowledge can compensate. However, function knowledge is supposed to store information about allocentric relationships, but not about how to manipulate a tool skillfully (i.e., egocentric relationship). To solve this theoretical problem, the manipulation-based approach assumes that the production system can be in charge of adapting the movements from input from the left temporal cortex (i.e., function knowledge). If so, the issue is why the human brain possesses manipulation knowledge that is not necessary for tool use? Likewise, impaired function knowledge can be compensated by intact manipulation knowledge. But, manipulation knowledge only encodes user-tool, egocentric relationships and is not thought to contain information about how to select and use the appropriate tools and objects (i.e., allocentric relationships). In other words, even if the multiple-routes-for-action hypothesis may appear very attractive at first glance, it suffers from theoretical problems that question the functional roles associated with function knowledge and manipulation knowledge.

Second, a significant body of evidence has indicated a strong link in left brain-damaged patients between the ability to actually use tools and to solve mechanical problems⁵ (Goldenberg and Hagmann, 1998, Goldenberg and Spatt, 2009, Hartmann et al., 2005, Jarry et al., 2013, Jarry et al., 2015, Osiurak et al., 2009, Osiurak et al., 2013; for reviews see Goldenberg, 2013, Osiurak, 2014c). Given that mechanical problem solving tasks involve the use of novel tools, it can be hypothesized that they put a heavy demand on executive functions. However, mechanical problem solving tasks are not impaired after frontal lobe lesions (Goldenberg and Hagmann, 1998, Goldenberg and Spatt, 2009) and are not correlated with performance on “executive” tasks (e.g., Tower of London; Hartmann et al., 2005, Jarry et al., 2013). In addition, patients with dysexecutive syndrome perform relatively well on these tasks as compared with healthy subjects (Goldenberg et al., 2007). Rather, neuropsychological evidence suggests that a common cognitive process is involved in any use situation, whatever tools are familiar or novel. This process could be supported by the left IPL (Goldenberg and Hagmann, 1998, Goldenberg and Spatt, 2009). Neuroimaging studies have corroborated this finding (Peeters et al., 2009, Peeters et al., 2013, Vingerhoets et al., 2011; see also Fridman et al., 2006, Orban and Caruana, 2014).

The strong link between real tool use and mechanical problem solving is difficult to explain within the manipulation-based approach. Given that the tools used in mechanical problem solving tasks are novel, manipulation knowledge cannot be activated to determine how to manipulate them skillfully. Moreover, to solve mechanical problems, people have to form an allocentric representation of the tool solution (e.g., a hooking action involves the relationship between a hook and something that can be hooked; see Osiurak, 2013). Manipulation knowledge encodes egocentric relationships, so it cannot be employed to form this allocentric representation. Another possibility is that function knowledge is the common process underlying both real tool use and mechanical problem solving, notably because this kind of knowledge contains information about allocentric relationships. However, neuropsychological evidence demonstrates that function knowledge and mechanical problem solving skills can be disrupted independently (Goldenberg and Spatt, 2009, Hodges et al., 1999, Hodges et al., 2000, Jarry et al., 2013; Lesourd et al., in press; Spatt et al., 2002), ruling out this possibility.

On the basis of these findings, it has been assumed that human tool use might be supported by the ability to reason about the physical object properties (Osiurak, 2014c; see also Goldenberg, 2013, Goldenberg and Hagmann, 1998, Goldenberg and Spatt, 2009, Osiurak et al., 2010, Osiurak et al., 2011, Osiurak et al., 2013). This reasoning is based on mechanical knowledge (e.g., cutting, lever, percussion), which corresponds to abstract knowledge about physical principles and, as a result, encodes allocentric relationships. This knowledge is viewed as abstract for two reasons. First, there is no overlapping between the physical reality and the technical reality. The same physical matter (e.g., glass) can possess distinct properties (resistant, sharp, transparent, etc.). Conversely, distinct physical matters (plastic, wood, metal, etc.) can have the same property (e.g., resistant). Moreover, the same physical matter does not always offer the property appropriate for a given action. For example, the lead of a pencil is friable when applied to paper but not to leather. Second, if people stored the properties of a given tool or object (e.g., friability of the lead of a pencil) in an absolute way, then they would not be able to transfer the mechanical principles they learned in a given situation to another one. Yet, as mentioned above, one of the specificities of human tool use lies in the transfer ability (Leroi-Gourhan, 1971, Martin-Ordas et al., 2008, Penn et al., 2008). In addition, this knowledge is supposed to be contained within the left IPL (e.g., Goldenberg, 2009, Goldenberg, 2013, Goldenberg and Spatt, 2009, Osiurak, 2014c, Osiurak and Lesourd, 2014) and particularly within the area PF of the SMG (Orban and Caruana, 2014; see also Caspers et al., 2006, Caspers et al., 2008; see also below). This contrasts with the manipulation-based approach according to which the left IPL is associated with manipulation knowledge.

The reasoning-based approach also offers another interpretation of the role of function knowledge located within the left temporal cortex. As discussed, the manipulation-based approach suggests that function knowledge is useful only when manipulation knowledge is impaired, as a means of compensation. In this view, no prediction is emitted about the idea that function knowledge may be differentially involved in real (i.e., the actual use of a tool with its corresponding object) versus single tool use (i.e., the use of a tool presented in isolation). By contrast, according to the reasoning-based approach, function knowledge might help people to organize the search in memory in order to get tools and objects that are not here now. When people engage in everyday activities, all the needed tools and objects are not at hand in the workspace, forcing them to get them either before or during the activity. In this view, Osiurak (2014c) (see also Osiurak et al., 2008, Osiurak et al., 2010, Osiurak et al., 2011) proposed that real tool use is mainly supported by mechanical knowledge, as demonstrated by the strong link between real tool use and mechanical problem solving. Nevertheless, function knowledge might be particularly involved in single tool use. When people are presented with a tool in isolation, they have to form a representation of the examiner’s expectations. Even though function knowledge is based on personal experience, it is also a vehicle for social knowledge given that daily life activities are culturally shared. Therefore, people can access information from function knowledge to represent the examiner’s expectations and, as a consequence, identify the category to which the tool belongs as well as a potential usage and an object with which it can be used. In line with this, patients with a selective impairment of function knowledge should encounter difficulties to show the conventional use of familiar tools presented in isolation, leading them to use mechanical knowledge to infer potential uses from the surrounding environment. However, performance should be improved by the addition of the corresponding object because mechanical knowledge enables to infer a potential use. Neuropsychological evidence supports this view.

Osiurak et al. (2008) (see also Sirigu et al., 1991) described a patient (MJC) with left temporal lobe lesions and bilateral frontal lobe lesions following a closed-head injury. She had a severe semantic impairment and met difficulties in single tool use tasks. Interestingly, MJC used the desk to demonstrate how to use the tools. For example, she used a screwdriver as a kind of gimlet, saying: “One can make a hole with it”. The performance was normal when both the tool and its corresponding object were present (real tool use) as well as when she had to show how to use familiar tools in a non-conventional way. Other studies in patients with a selective impairment of function knowledge have corroborated this observation by documenting a strong relationship between function knowledge and single tool use as well as intact ability to use novel tools (e.g., Hodges et al., 2000, Silveri and Ciccarelli, 2009, Sirigu et al., 1991, Spatt et al., 2002; for a review, see Osiurak et al., 2011; see also Lesourd et al., in press).

The reasoning-based approach posits that the movements associated with manipulation are reconstructed de novo on the basis of (1) the mental simulation of the tool-use action generated by mechanical knowledge⁶ and (2) on-line information about the physical environment and the position of the body in the space. This reconstruction would take place within the production system, a system very close to the aforementioned primate prehension system. So, the production system might be essentially supported by the IPS (phAIP and DIPSA; see Orban and Caruana, 2014). This perspective is very close to the one proposed by the manipulation-based approach. However, contrary to the manipulation-based approach, the production system is thought here as the only system that encodes egocentric relationships between the user and the tools and objects. Besides, Orban and Caruana (2014) (see also Peeters et al., 2009, Peeters et al., 2013) proposed that an anterior portion of the left SMG (aSMG) might play a critical role by integrating the information coming from phAIP (i.e., production system) and from the area PF of the left SMG (i.e., mechanical knowledge). In this view, the left aSMG might send some biasing signals to phAIP to favor the selection of the handgrip (i.e., egocentric relationship) that best suits the correct use of the tool generated by mechanical knowledge (i.e., allocentric relationship).

A significant body of literature has demonstrated considerable functional homologies between the human and the primate brain with regard to prehension skills. These homologies can explain why both humans and nonhuman primates are able to manipulate objects and, as a result, to use tools. However, despite this similarity, human tool use appears to differ from nonhuman primate tool use in several respects (i.e., frequent use, wide repertoire of tools, use of secondary tools, and transfer abilities). These differences go along with the existence of evolutionary new neural substrates peculiar to humans within the left IPL/SMG. The manipulation-based approach claims that this area might be responsible for the storage of sensorimotor knowledge about tool manipulation, a perspective close to the continuity theory by considering that the main difference between human and nonhuman tool use is more one of degree than of kind. By contrast, for the reasoning-based approach, this area might be involved in the ability to reason about physical object properties on the basis of mechanical knowledge. This approach is more akin to the discontinuity theory by suggesting that the main difference is of kind rather than of degree.

As discussed so far, most of our understanding of the neurocognitive bases of human tool use has come from the studies of brain-damaged patients with tool use disorders. Although these studies have been fruitful to generate a certain number of theoretical proposals, they are generally insufficient to characterize with a great deal of precision the neural substrate associated with tool use skills. Particularly, the role of the left IPL remains a matter of debate, notably because no less than seven cytoarchitectonic areas have been identified within this area. Five of them (Pfop, PFt, PF, PFm, and PFcm) are located approximately at the position of BA 40 on the SMG, the remaining two areas (PGa and PGp) approximately cover the region of BA 39 on the angular gyrus (Caspers et al., 2006, Caspers et al., 2008, Zilles et al., 2002). As discussed by Orban and Caruana (2014), special attention has to be paid to PF (mechanical knowledge) as well as aSMG (integration area), which largely overlaps with cytoarchitectonic area PFt (Caspers et al., 2008, Peeters et al., 2013). Interestingly, many neuroimaging studies have been carried out on the topic in the last decade. So, even if no theoretical model of human tool use has been formulated on the sole basis of these studies, they offer a good opportunity to reach this level of precision.

To sum up, the goal of the present review is to examine the predictions derived from the manipulation-based and the reasoning-based approach by conducting a comprehensive meta-analysis on functional neuroimaging data, based on activation likelihood estimation (Eickhoff et al., 2012). Since the development of neuroimaging techniques and the subsequent increase in the number of functional imaging studies, the scientific community has been faced with the need to synthesize results from the literature. Therefore we aim to provide here an overview of previous experiments on the functional brain activity related to tool use. We wished to integrate functional neuroimaging results across a large number of selected studies through a quantitative meta-analytical approach. To do so, we used a coordinate-based meta-analysis (CBMA, Chein et al., 2002, Turkeltaub et al., 2002) with the aim of identifying anatomical locations where an effect can be observed consistently across experiments.

The numerous published neuroimaging studies concerning human tool use have employed a wide range of tasks and task comparisons, also called contrasts. These contrasts can be qualified as general or specific (see Binder et al., 2009). A general contrast is one between a condition that elicits cognitive processes involved in tool use and a “baseline” condition (e.g., performing a meaningless gesture). A specific contrast entails a comparison between two conditions involving tool-use cognitive processes (e.g., conditions with correct manipulation versus conditions with incorrect manipulation). In this review, we only included general contrasts, because the number of specific contrasts of the same kind was too low to run a meta-analysis. A critical criterion concerned the nature of the task. In this review, we included both use and non-use tasks. Use tasks correspond with tasks wherein participants have to imagine, plan or execute tool-use actions either with the object in hand (real use) or not (pantomime). In non-use tasks, participants are not asked to perform or even imagine tool-use actions. Nevertheless, they are confronted with tool stimuli and have to decide, for instance, whether the tool-use action shown is correct or not (complex observation) or simply observe the tool-use action while performing an n-back memory task (simple observation). We focused on data obtained with visual objects to avoid the involvement of language-related neural circuit. Thus, all the studies using words and silent reading were excluded from the analysis. Only two studies using words but not visual objects were included because the task was to determine whether the tools shared the same action, gesture or context. On the basis of these criteria, the general contrasts (hereafter referred to as conditions) included in the present meta-analysis were as follows.

The first category of conditions only concerned “non-use tasks”. We identified three sub-types for this first category.

• Action. The task focuses on the understanding of the action made by the tool with the object (allocentric relationship; e.g., is it correct to use this pair of scissors to cut this sheet of paper? Bach et al., 2010). Here, no judgment has to be made on the appropriateness of the manipulation.

• Gesture. The task is to determine whether the manipulation is correct or not (egocentric relationship), without taking into account the action made by the tool with the object (e.g., Does this hand posture – for instance a pinch posture – matches the action goal – for instance, throwing a dart? Vingerhoets et al., 2013a, Vingerhoets et al., 2013b).

• Context. The task is to determine with which object or in which context a given tool can be used (allocentric relationship; e.g., Can these two tools – poultry shears and hand spiral beater – be used in the same context? Canessa et al., 2008).

The second category of conditions was based on the nature of the tools presented, whatever the task required (both use and non-use tasks). Two sub-types of conditions were concerned.

• Familiar. The task focuses on the conventional use of familiar tools (e.g., pantomiming the use of a pair of scissors; Vingerhoets et al., 2011).

• Unfamiliar. The task includes novel tools or familiar tools used in a non-conventional way (e.g., pantomiming the use of screwdriver-like tool; Vingerhoets et al., 2011).

Finally, the third category only included use tasks. Three sub-types could be identified.

• Planning. This corresponds to tasks wherein participants have to imagine using a tool or to a specific condition of the tasks where they have to plan a subsequent tool-use action (e.g., imagining both grasping and using a toothbrush; Vingerhoets et al., 2009).

• Execution. This corresponds to a specific condition of a task where participants execute the tool-use action (e.g., performing the real tool-use actions, such as using a hammer to pound a nail; Hermsdörfer et al., 2007).

• Planning/Execution. Here, the distinction between planning and execution is not made (e.g., transporting pins with chopsticks; Tsuda et al., 2009).

On the basis of these different categories of conditions, several predictions can be emitted from each approach that will be detailed in the following lines (Table 1).

• The role of the left IPL diverges according to the two approaches. To explore this, we compared data from studies exploring how people judge the appropriateness of use gestures (GESTURE) versus how they understand the actions performed by tools (ACTION). The manipulation-based approach predicts that GESTURE conditions should show an activation of left IPL/SMG (e.g., Buxbaum, 2001, Buxbaum and Kalénine, 2010, Thill et al., 2013, Van Elk, 2014). By contrast, the reasoning-based approach predicts that the left IPL (particularly the area PF of SMG) should be only involved in ACTION conditions (e.g., Goldenberg, 2013, Orban and Caruana, 2014, Osiurak, 2014c). Moreover, the reasoning-based approach assumes that only the production system encodes tool-hand interaction (i.e., egocentric relationship). So, in line with this, this approach predicts that the IPS (Orban and Caruana, 2014) should be more involved in GESTURE conditions.

• The manipulation-based approach remains silent about the strong link between real tool use and mechanical problem solving. Consequently, it does not predict that FAMILIAR and UNFAMILIAR use of tools should involve different cerebral regions (e.g., Buxbaum, 2001, Van Elk, 2014). At best, it can be expected that both the left IPL (i.e., manipulation knowledge) and the left temporal cortex (i.e., function knowledge) are preferentially activated by FAMILIAR use as compared to UNFAMILIAR use. By contrast, the reasoning-based approach suggests that only left temporal lobe regions should be more involved in FAMILIAR use than in UNFAMILIAR use (e.g., Goldenberg, 2013, Osiurak, 2014c). This rationale is based on the idea that the left temporal lobe contains function knowledge that is of particular interest for familiar tools (see above). Moreover, the left IPL (particularly the area PF of SMG; see Orban and Caruana, 2014) should be more activated in UNFAMILIAR use than in FAMILIAR use, because only mechanical knowledge is involved when people are confronted with unfamiliar use.

• Both the manipulation-based and the reasoning-based approach assume that, besides the areas supporting the production system (mainly the left IPS because the participants included were right-handed), the left IPL should be particularly involved in PLANNING tool-use actions, because manipulation knowledge or mechanical knowledge is thought to be the basis for the conception of intended actions. Interestingly, according to the reasoning-based approach, the left aSMG should be specifically activated in PLANNING because of its integrative role (Orban and Caruana, 2014). In addition, both approaches agree on that the production system should be engaged in EXECUTION.