In order to interact with space, humans need to integrate spatial information according to new objects encountered and/or new motion information (e.g., object- or self-motion, Wolbers and Hegarty, 2010). Encoding and storing spatial information is supposed to rely on the activity of one of these two types of spatial representation (Burgess, 2006): (a) egocentric, centered on the subject, coding for self-to-object relations; (b) allocentric, centered on the environment, coding for object-to-object relations. Various evidence sources, ranging from single neuron recordings in animals (O'Keefe and Dostrovsky, 1971; Andersen et al., 1985) to fMRI and neuropsychological observations (Bisiach and Luzzatti, 1978; Aguirre and D'Esposito, 1999; Holdstock et al., 2000; Mellet et al., 2000; Spiers et al., 2001; Chokron, 2003; Burgess et al., 2006; Fields and Shelton, 2006; Maguire et al., 2006; Shrager et al., 2008), suggest that the two types of spatial representations rely on the occipito-temporo-parietal cortices. Among these regions, the posterior parietal cortex sustains egocentric processing (Chokron, 2003) while the medio-temporal regions are required for the allocentric processing (O'Keefe and Nadel, 1978; Holdstock et al., 2000; Milner and Goodale, 2008; Finke et al., 2011).

The connection proposed between the allocentric processing and the hippocampus is partly due to the strong relationship between the “Place cells” firing and landmark sensory aspects observed in animal studies (e.g., O'Keefe and Dostrovsky, 1971; O'Keefe and Nadel, 1978; O'Keefe and Burgess, 1996). fMRI studies in humans during route navigation congruently revealed that the activation of hippocampal and parahippocampal regions correlated with relevant landmark information (Janzen and Van Turennout, 2004; Janzen and Weststeijn, 2007; Janzen et al., 2007). However, several data sources suggest that the hippocampus is not always involved or necessary for allocentric processing (Aguirre et al., 1996; Bohbot et al., 1998; Galati et al., 2000; Committeri et al., 2004; Bastin et al., 2013). For instance, Bohbot et al. (1998) reported, that the performance of patients suffering from unilateral hippocampal lesions in an adapted version of the Morris water maze task (assessing landmark-based allocentric representation) was similar to that of control subjects. Moreover, as revealed by fMRI, some variants of allocentric processing seem to be supported by a large network of regions including parietal, retrosplenial, and parahippocampal cortices (e.g., Aguirre et al., 1996; Vallar et al., 1999; Galati et al., 2000; Committeri et al., 2004; Zhang and Ekstrom, 2013; for review Galati et al., 2010).

The observable discrepancy between these results and the exclusive allocentric role proposed for the medio-temporal lobe (O'Keefe and Dostrovsky, 1971; O'Keefe and Nadel, 1978) could be due to the way spatial information is encoded. Specifically, beyond the allocentric vs. egocentric distinction relative to the type of spatial relation involved (object-to-object or self-to-object, respectively), the presence or the absence of self-related changes could be a decisive factor in the hippocampal involvement. Using a short-term spatial task, Gomez et al. (2014, 2012) disentangled different sources of information required during encoding. The authors showed that patients suffering from bilateral hippocampal atrophy were impaired for reproducing a trajectory learned on the basis of self-motion information. However, they performed similarly to controls when the trajectory was learned without any self-related changes (i.e., observing the experimenter producing the trajectory). This result is in-line with the hypothesis that hippocampal activity may be observed when self-related changes are required to build spatial representations.

Functional MRI evidence provides indirect support to this last hypothesis. Studies manipulating the type of visual perspective during spatial encoding itself showed that hippocampal and parahippocampal structures were more activated during route navigation than during survey navigation (Shelton and Gabrieli, 2002; Shelton and Pippitt, 2007). Hence, despite previous behavioral reports of a strong connection between survey perspective and allocentric representation formation (Thorndyke and Hayes-Roth, 1982; Allen, 1999), such processing seems less tightly related to hippocampal activity than during route navigation. The enhanced activation of hippocampal and parahippocampal structures for route navigation may reflect the integration of new information generated by self-related changes during the egocentric spatial processing (Shelton and Pippitt, 2007).

Beyond the type of visual perspective during spatial encoding, the participants' spatial processing during each encoding condition should be further controlled to insure that the differential hippocampal activity is related to new self-related changes. Actually, the foreseen retrievals tasks can modulate the strategies (allocentric vs. egocentric) used by participants during encoding itself (Shelton and Gabrieli, 2004). From a methodological point of view, this control can be achieved by warning participants that they will have to perform an object-to-object relation retrieval of the spatial information before they enter the survey perspective encoding. On the contrary, participants can be instructed to encode spatial information to subsequently perform a self-to-object relation retrieval task which will insure that the two tasks will in fact differ on the integration (or not) of self-related changes.

If the presence/absence of self-related change is indeed crucial, the information about how the self-related changes are generated could be relevant in providing a thorough understanding of the mechanisms responsible for the different involvements in the spatial memory network. Specifically, given areas could be involved when a change in head direction modifies orientation cues without updating the observer's location. On the reverse, different ones could be involved when both a change in orientation cues and an update in the observer's position occur.

Regarding the head orientation, studies on single cell recordings in animals revealed that several cells fired selectively when the animal oriented its head in a given direction. Cells' localization was first described in post-subiculum (Ranck, 1984), limbic system (Taube, 1998; Robertson et al., 1999) including the retrosplenial cortex (Cho and Sharp, 2001), and hippocampus (Leutgeb et al., 2000) of animals. In humans, heading disorientation was observed after lesion of the retrosplenial cortex (Aguirre and D'Esposito, 1999). More recently, using single-neuron recordings, Jacobs et al. (2010) showed that human entorhinal cortex could contain path cells encoding in a clockwise or counterclockwise route direction during navigation in circular environments. Functional MRI studies have revealed that retrosplenial activity was modulated by previously learnt variations of the head direction (Baumann and Mattingley, 2010).

Farrell and Robertson (1998) defined as the egocentric-updating processing, changes related to both head orientation (i.e., head direction changes) and observer's spatial position. Several animal studies and models of spatial neuronal networks have involved hippocampal “Place cells” during such spatial changes. In fact, based on neuronal recordings in animals, these models indicate that place cell firing is also related to path integration (Mittelstaedt and Mittelstaedt, 1982), which refers to the ability to return directly to a starting point from any location relying mainly on idiothetic cues (i.e., self-motion cues). Path integration can be considered as a subtype of the general process of egocentric-updating where objects are updated relative to the individual (Burgess, 2008). These models emphasized the need to combine landmark sensory aspects and idiothetic cues to build a place code (e.g., Redish and Touretzky, 1997; McNaughton et al., 2006).

These various approaches suggest that the retrosplenial cortex may be crucial in processing heading changes within egocentric referencing, while the hippocampus may be related to egocentric-updating. No fMRI evidence in humans has clearly disentangled the underlying network subtending these two types of processing.

In the present fMRI study, we aim to characterize the neural substrate of spatial memory encoding. Beyond an allocentric (object-to-object) vs. egocentric (self-to-object) distinction, we further disentangled the egocentric information by varying the types of self-related changes during encoding (egocentric-updating vs. rotation only). As such, we defined three spatial encoding conditions: Allocentric, Egocentric-updating, and Egocentric with rotation only.

One hypothesis is that the cerebral organization evolution of cognitive functions from animals to humans could lead to a different neural specialization of spatial memory encoding in specific areas such as the hippocampal and retrosplenial areas. Alternatively, given the neuropsychological evidence, we expect that spatial cognitive functions share similar structures in animals and humans. Therefore, we can expect that the egocentric-updating condition reveals additional activity in the hippocampal structure compared to the allocentric condition. This supports the idea that the human hippocampal area could also code for self-related changes. Additionally, compared to egocentric-updating, the egocentric with rotation only condition should involve retrosplenial activation, supporting the idea that the human retrosplenial cortex codes specifically for orientation changes.

The experiment consisted of two phases: a training phase performed outside the MR magnet and a testing phase performed during the MR examination. The training phase had three objectives: (1) familiarization with the virtual presentation via free navigation in the environment, (2) instructions on how to perform each task, (3) task execution with performance feedback. During the testing phase, three spatial encoding conditions and a control condition were performed during a block-design paradigm.

Our results confirmed the activation of a large occipito-parieto-temporal network common for spatial encoding conditions (e.g., Ghaem et al., 1997). Beyond the classical allocentric-egocentric distinction, the type of referencing induced by the visual perspective during encoding and the task-demand, allowed the identification of essential regions within the spatial memory network. The major result was the specific involvement of two areas during the self-related conditions, which were the retrosplenial cortex for orientation changes, and the right hippocampus for self-location changes.

The first crucial result of the study reveals an enhanced retrosplenial activity when contrasting ERO to EU. Encoding new spatial information during the ERO condition only relied on rotation changes, while, during the EU condition, the weight of rotation information during the integration of new spatial information was weakened by the concomitant change of location. This result therefore provides evidence that the retrosplenial cortex in humans specifically processes changes in head direction. Moreover, although the performances for A and EU conditions were correlated, the ERO performance was not correlated with the allocentric or the egocentric-updated performance. Therefore, participants that were good at processing head direction changes were not necessarily good at processing location changes or allocentric landmark-based information. This supports the idea that relatively independent cognitive mechanisms are at play in those tasks.

Although the retrosplenial cortex (BA 30) is known to contain head-direction cells in rodents (Cho and Sharp, 2001; Sharp et al., 2001; Wiener and Taube, 2005), only limited experimental evidence was reported in humans. Concerning clinical evidence, it has been suggested that right retrosplenial lesions lead to pure topographical disorientations (Valenstein et al., 1987; Yasuda et al., 1997; Aguirre and D'Esposito, 1999; Maguire, 2001; Vann and Aggleton, 2004; Vann et al., 2009), reflecting a type of heading disorientation (i.e., an inability to represent the orientation direction with respect to the external environment).

Functional MRI studies have provided evidence suggesting that the retrosplenial cortex was strongly activated during scene viewing, scene imagery, and scene memory (Epstein et al., 2007). Moreover, a meta-analysis of navigation fMRI studies reported bilateral involvement of BA 30 in humans (Maguire, 2001; for an extensive review see Vann et al., 2009). Congruently, we report a common spatial encoding network encompassing the bilateral retrosplenial cortex (when compared to control condition). Overall, these results make it possible to pinpoint the implication of the retrosplenial region in spatial processing.

However, with regards to head-direction, only one fMRI study aimed at identifying neural correlates of orientation This study reported a region situated close to the anatomically defined retrosplenial cortex (BA 30, see Maguire, 2001, for a review), in BA 31 (Baumann and Mattingley, 2010). In our current study we report for the first time the activation of the right retrosplenial cortex (BA 30) for the rotation. This result is in agreement with animal studies of spatial encoding. Additionally, we observed the BA 30 activation during ERO vs. A contrast, next to the bilateral posterior cingulate (BA 31), in line with Baumann and Mattingley (2010)'s findings.

An influential model of spatial memory proposes that the retrosplenial cortex supports, in humans, stimulus conversion from an egocentric reference frame (in the parietal cortex) to an allocentric reference frame (in the medial temporal regions) and vice versa (Burgess et al., 2001a,b; Burgess, 2008; Vann et al., 2009). However, with regards to retrosplenial function, this model was mainly based on animal studies evidence. Few fMRI studies have reported retrosplenial activity congruent with this model (Lambrey et al., 2012; Zhang et al., 2012). Our result supports the hypothesis of retrosplenial cortex encompassing cells activated for head-direction in humans. Neuronal electrical recordings in humans are necessary to provide further direct support of the retrosplenial involvement.

The second crucial activated region reported in this study is the right hippocampal area. It was preferentially involved in spatial encoding, which required an online integration of self-location transformation in the spatial representation, rather than encoding the schematic spatial relation amongst objects (O'Keefe and Nadel, 1978). In fact, the contrast of the EU condition to the A condition significantly enhanced the right hippocampal activity. This result is in line with previous observations of right hippocampal activation during path integration encoding in humans (Wolbers et al., 2007). Nonetheless, during retrieval, the left hippocampal structure might be further involved, as suggested by Iglói et al. (2010). Moreover, this hippocampal activation is congruent with Shelton and Gabrieli (2002)'s results which compared cerebral activities during encoding from a route perspective and a survey perspective. They observed bilateral hippocampal activation specific to the route condition. Differences in the type of subsequent task-demand (recognition task vs. spatial self-to-object pointing in this study) and the environment complexity (large with multiple boundaries and 17 landmarks vs. medium with an average of 5 landmarks in this study) can account for the right lateralization of the hippocampal activation in the present study.

Moreover, the modulation by between-subject retrieval performance of a hippocampal-caudate cluster observed in the correlation analysis indicates a critical role for these structures in the subsequent task-resolution. During encoding, good egocentric-updating pointers showed stronger hippocampal activation (extending to caudate). The hippocampal activity has previously been reported to correlate to accuracy of self-location or self-orientation, in a path integration task (Wolbers et al., 2007, using an egocentric pointing task) or during active navigation, (Maguire et al., 1998; Hartley et al., 2003). We therefore interpret this hippocampal activity in good memory performers as highlighting the importance of the right hippocampus for processing self-location changes.

Because the EU vs. A contrast provides the neural substrate engaged by the egocentric-updating and due to self-motion information (i.e., idiothetic) rather than landmark-based information, we suggest that the right hippocampal region may be more specifically devoted to location related self-related changes. Additionally, because allocentric processing such as landmarks and room geometry processing were likely to occur in the EU condition as well, it is possible that the hippocampal involvement could also arise from the relative flexibility of the representation constituted during the EU encoding compared to the more schematic representation during the A encoding. In fact, Zhang and Ekstrom (2013) have shown that such differences in the use of an allocentric representation could be responsible for a differential hippocampal involvement in favor of a flexible representation. As this right hippocampal activity is also observed when directly contrasting EU vs. ERO within egocentric referencing, it is likely that self-motion location change in the EU condition plays a crucial role in this hippocampal activation. This additional evidence supports the hypothesis that, in humans, location self-related changes involve the right hippocampal region.

This result is new and central, as a wide body of evidence from animal literature, and in particular the discovery of place cells, have led to the proposal that the hippocampal structure underlies cognitive mapping in an allocentric reference frame (O'Keefe and Dostrovsky, 1971; O'Keefe and Nadel, 1978; O'Keefe et al., 1998). In humans, single cells neuron recordings have shown that place cells exist in the human hippocampus (Ekstrom et al., 2003).

In the present experiment, we report substantial hippocampal activation when participants must encode allocentric landmark-based spatial information (there is a significant activation of the hippocampus in the A encoding condition compared to the C encoding condition). However, given previous neuropsychological results, we expected that the hippocampal structure could be enhanced when idiothetic self-location changes are involved (Worsley et al., 2001; Philbeck et al., 2004; Gomez et al., 2012). From a behavioral perspective, participants who performed well in the EU condition also performed well in the A condition. Together with the conjunction results, this observation supports the hypothesis that a unique structure, the hippocampus, may underlie different spatial processing, in the present case, egocentric-updating and allocentric processing, which could explain this correlational effect.

Overall, these findings support models emphasizing the need to combine allocentric landmark sensory aspects to egocentric-updating to form the place code of Place cells (e.g., Redish and Touretzky, 1997). Several proposals have suggested that the hippocampus might be involved in complementary types of spatial processing such as path integration processing (Redish and Touretzky, 1997; Whishaw et al., 1997, 2001; Worsley et al., 2001; Wolbers et al., 2007). These models have underlined that the hippocampal structure is well-suited to underlie such spatial processing as it is interconnected to various structures making it possible to represent, together, the body's position: it receives input from systems assumed to represent head-direction (such as the retrosplenial cortex, and thalamus), and from systems processing self-motion, such as the human motion complex (Redish and Touretzky, 1997; Wiener and Taube, 2005).

In closing, our results refine previous findings on navigation in large-scale environments (Maguire et al., 1998; Ekstrom et al., 2003; Hartley et al., 2003) by suggesting that hippocampal activity should be extended to represent self-related location transformations. In keeping with existing models (Burgess et al., 2001a; Buzsaki, 2005), we suggest that, in humans, during spatial encoding: (1) the retrosplenial cortex processes heading-direction; (2) the hippocampus processes self-related location transformations and combines it with landmark information to allow place computations. Such a combination would allow for correcting accumulated errors during egocentric-updating.

To conclude, this experiment highlighted that, in some circumstances determined by the experimental conditions, hippocampal and retrosplenial structures known to be involved in allocentric environmental coding (Galati et al., 2010) could demonstrate a preferential involvement in an egocentric coding of space. Consequently the differentiation between allocentric vs. egocentric representation no longer seems to be sufficient in understanding the complexity of the mechanisms at play during spatial encoding.