Introduction
People differ in their ability to memorise information. However, participants of memory championships—memory athletes—exhibit a completely different scale of memory performance. They are able to memorise more information quicker and more reliably than what is within the normal range of memory performance: Remembering 300 random words in only 15 min without a single mistake is not a feat one can just perform. However, the memory athletes tested here are capable of this and similar feats. One central pillar of their success is a mnemonic strategy that is known for its encoding efficacy since ancient Greece: the method of loci (Roediger
1980; Yates
1966). Users of this strategy mentally navigate a familiar route and at separate loci—distinct landmarks along the route—visualise placing the information there. This combination of map-based spatial memory and associative memory has repeatedly been demonstrated to enhance memory for a broad variety of information (Worthen and Hunt
2011).
Successful memory athletes attribute their memory performance mainly to the method of loci (Dresler et al.
2017; Dresler and Konrad
2013; Maguire et al.
2003). Little is known, however, why the method of loci facilitates memory retention so strongly. One explanation might be that the method engages different memory systems synergistically. In the classification of memory subsystems two aspects are often contrasted (Squire
2004): habits or simple stimulus–response association (Jog et al.
1999; Knowlton et al.
1996; Mishkin and Petri
1984; Yin and Knowlton
2006) and more episodic and map-like representations (Eichenbaum
2004; O’Keefe and Nadel
1978). While the former is linked to the caudate nucleus, the latter is linked to the hippocampus. This division is exemplified in the context of navigation: a stimulus response strategy would rely on simple association of landmarks and actions (“turn right at the church”). In contrast, navigation using the map-based system would rely on an internal map of the environment. As efficiency of these two systems depends on the environmental context, they often compete for the task at hand so the ideal system for the task is utilised (Doeller et al.
2008; Poldrack and Packard
2003).
During the method of loci, new information needs to be associated with the loci; and after successful encoding of one piece of information, one needs to navigate to the next locus as quickly as possible (Mallow et al.
2015). For the association (Knowlton et al.
1996; Yin and Knowlton
2006) and the automatic navigation along a well-known, fixed route—characteristically for stimulus–response learning—(Hartley et al.
2003; Packard and Knowlton
2002) the caudate nucleus appears ideally suited. Memory athletes routinely create a vivid visual image for the association of new information on a given locus. For the vividness (Danker et al.
2016), for constructing a visual scene (Hassabis and Maguire
2011; Zeidman and Maguire
2016), and for maintaining a map of the whole set of information along the route (O’Keefe and Nadel
1978) the hippocampus is usually recruited. The routes memory athletes use are ones that they are extremely familiar with containing many different loci. For this kind of representations, the hippocampus with its map-based encoding should be ideal (O’Keefe and Nadel
1978). During training of the method of loci, the memory athletes would train over and over again to use these routes. Memory athletes tend to train multiple routes so that during a competition they do not need to reuse the same route, which potentially might lead to interference. At a competition, the well-rehearsed routes are then used to encode novel information, going along the route. Taken together, on the one hand the rapid navigation from locus to locus would be served well with efficient stimulus–response associations provided by the caudate nucleus. Whereas on the other hand, the vivid scene construction needed for encoding and the maintaining of a global representation of the route could be done by the hippocampus. Integrating these facilities in a frictionless fashion might be what enables memory athletes’ superior memory. Preliminary evidence about the involvement of the caudate nucleus and the hippocampus comes from work on mnemonics: Both the method of loci and the pegword method, a similarly associative but non-spatial mnemonic, show caudate nucleus activity during encoding, however, only the method of loci elicits increased hippocampal activation (Fellner et al.
2016). This supports the specific involvement of the hippocampus in the spatial dimension of the method of loci.
There is substantial evidence for a competitive interaction of the hippocampus and the caudate nucleus: during spatial navigation, lesioning the one system improved performance based on the other system and vice versa (Poldrack and Packard
2003). This double dissociation implies that when both systems are intact, they are competing for the task at hand, which in turn reduces their efficiency (Lee et al.
2008; Packard et al.
1989). However, using early stage Huntington disease as a model for lesions in the caudate nucleus, a compensatory role of the hippocampal system has been observed; while the function of the caudate nucleus decays, the hippocampus can rescue the loss of functionality. Furthermore, in the same study, they observed a cooperative interaction of the memory systems in healthy controls which facilitated route recognition performance (Voermans et al.
2004). We hypothesise a similar cooperative interaction between the hippocampus and the caudate nucleus in memory athletes to facilitate their memory performance as it supports the method of loci optimally.
We investigated 23 athletes out of the Top-50 of the memory sports world ranking and 23 controls matched for age, sex, and IQ. To study whether memory athletes show a stronger synergy between the hippocampus and the caudate nucleus, we combined structural analysis and functional analysis of resting-state brain connectivity. We are not comparing task activation of memory athletes to matched controls as that is confounded by performance differences. Therefore, it is difficult to distinguish whether observed differences in activation are cause or consequence of behavioural differences.
In contrast to matched controls, athletes might exhibit more refined mechanisms for mnemonic processing or utilise a qualitatively different approach in terms of neural processing. To capture both of these differences, our analysis strategy is twofold: comparing our sample to matched controls, we test how they differ structurally and functionally; relating structural and functional variation within the athlete sample to their position in the world ranking, we investigate what predicts their success. Both analyses complement each other. The comparison to the control group can reveal anatomical changes common among the athletes, while the association to the world ranking can identify anatomical patterns that are central to the success of the athletes. As previous work showed a functional gradient along the anterior to posterior axis of the hippocampus (Strange et al.
2014) that is directly implicated in spatial processing (Kjelstrup et al.
2008), we subdivided the hippocampus into anterior and posterior part. The anterior and posterior hippocampus have been dissociated functionally on many aspects of cognition (Poppenk et al.
2013). A secondary reason for this was that an enlarged posterior hippocampus could be accompanied by a shrunken anterior hippocampus—producing no difference on average (Maguire et al.
2006).
We hypothesise that a specific trait or the massive training of the memory athlete is associated to structural differences in volumes of the hippocampus and the caudate nucleus; these should be accompanied by functional interactions that facilitate the synergistic use during the method of loci.
Discussion
Comparing 23 of the world’s leading memory athletes with carefully matched controls, we observed enlarged hippocampal volumes, especially pronounced in the right anterior division. In contrast, volumes of the caudate nucleus volumes did not differ significantly from those of matched controls. The position in the memory sports world ranking was predicted by both the volume of the right caudate nucleus and the right posterior hippocampus. A second feature distinguishing the groups was that for memory athletes, the volumes of the right posterior hippocampus and the right caudate nucleus were more strongly correlated than in matched controls. Using resting state data, we observed an association between the structural group difference in the right anterior hippocampus and correlations with performance. Functional connectivity from the anterior hippocampus to both the right caudate nucleus and the right posterior hippocampus predicted the ranking.
We suggest that these results are best understood in the context of cooperative hippocampal–caudate nucleus interaction that may enable the superior performance seen in memory athletes. We focused on the caudate nucleus and the hippocampus, because both the ability to create simple stimulus response associations—supported by the caudate nucleus—and the utilisation of map-like representations—supported by the hippocampus—are essential aspects of the method of loci. A differential neural architecture regarding these structures that makes memory athletes more apt at utilising the method of loci might manifest itself in two ways. First, athletes might be characterized by enlarged pivotal brain structures. Second, they might utilise neural mechanisms not readily available to normal controls. For this reason, we compared our sample of memory athletes with matched controls, and complementary, we related the structural and functional variation we find in the sample of memory athletes to their position in the world ranking, thus identifying what makes certain memory athletes especially successful.
Three of our results provide evidence for the model that memory athletes utilise hippocampal–caudate in a cooperative fashion to enhance their ability to memorise information: volumes of the posterior hippocampus and caudate nucleus were associated with the world ranking; these two volumes are more strongly correlated with each other within the athletes compared to the matched controls. Resting state functional connectivity of the anterior hippocampus to both the posterior hippocampus and the caudate nucleus predicted the world ranking. Memory athletes with both a large posterior hippocampus and caudate nucleus were able of more impressive memory feats across different types of material. On top of that, the better athletes showed a stronger functional connectivity between those two regions and the anterior hippocampus, a region that showed the largest volumetric difference relative to matched controls. As memory athletes attribute their exceptional memory abilities to mnemonic strategies, such as the method of loci (Dresler et al.
2017; Dresler and Konrad
2013; Maguire et al.
2003), we propose that our findings reflect the degree to which the athlete’s neural architecture supports the use of mnemonic strategies by an optimised, cooperative utilisation of the caudate nucleus and hippocampus. It is important to note that the memory athletes excel in different memory domains, such as face memory, word list learning, memorising playing cards, not only in competitions but also in laboratory settings (Konrad
2014). However, differing from other forms of superior memory, such as highly superior autobiographic memory (LePort et al.
2016) or superior recognition abilities (Russell et al.
2009), the memory athletes are not intrinsically better at memorising, they need their techniques for their exceptional memory performance (Konrad
2014; Ramon et al.
2016). The method of loci in itself has been applied in diverse sets of context to facilitate memory (Worthen and Hunt
2011). The biggest advantage one would have to apply mnemonics in real life is when there are large amounts of information that have to be learned, especially when the material in itself is not very well structured (as for example a story is). However, to use the methods on the level of the athletes, a large amount of training will be necessary. Some of the memory athletes have told us that they used their mnemonics to learn medical terms or a new language rather quickly. Though even naïve participants can tremendously improve using the method of loci (Dresler et al.
2017) to learn word lists. Thus, if one is willing to practice the mnemonics and has to learn large sets of facts or associations by heart mnemonics seem like a good way of facilitating learning.
In the past, the debate of the interaction between the caudate nucleus and the hippocampal memory systems was focused on a competitive interaction (Poldrack and Packard
2003), with the systems competing for solving the task at hand (Packard and McGaugh
1996). The central evidence for competition that has been replicated multiple times by now is the following: before the rats solve a navigational task in which different task requirements can be fulfilled by either system, one of the relevant structures gets lesioned. Trivially, behaviour depending on this structure drops substantially. But importantly, behaviour that depends on the other structure is improved after the lesion. This increase suggests that the lesioned structure was competing for solving the task (Jacobson et al.
2012; McDonald and White
1994). Compared to the amount of work supporting the competitive notion, there is only preliminary evidence for cooperation of these systems: the hippocampus can compensate for dysfunction of the caudate nucleus during the early stages of Huntington’s disease; but providing even stronger support for cooperation was a functional interaction between the hippocampus and the right caudate nucleus in healthy controls facilitating route recognition (Voermans et al.
2004). However, beyond the competition vs. cooperation dichotomy there is also work that suggests parallel processing that not necessarily implies cooperation or competition (Doeller et al.
2008). Most of the evidence for competition of the memory systems comes from rather simple navigation paradigms in which there are only two choices; one indicating use of the stimulus response system, the other indicating a more spatial hippocampal strategy. However, with the method of loci combining aspects from stimulus-response learning—such as rapid navigation from one locus to the next in a fixed order—and aspects of hippocampal processing—such as scene construction (anterior hippocampus) and maintaining of a spatial representation of the route (posterior hippocampus)—a cooperation of those two systems seems optimal to produce exceptional memory performance. As these aspects utilised in the method of loci are quite complementary, we presume that the different systems can cooperate rather than interfere with each other as was shown in other navigational tasks (Packard and McGaugh
1996).
Extending this reasoning to our results suggests that the memory athletes show higher levels of cooperation between the hippocampus and caudate nucleus, thus facilitating the use of the method of loci.
One finding that links especially nicely to our results is that participants who focused stronger on a spatial strategy compared to a response-based strategy in a virtual navigation task showed increased grey matter density in the hippocampus while it was reduced in the caudate nucleus. Additionally, these densities were negatively correlated (Bohbot et al.
2007). This result is in line with the competition account: as the hippocampus and the caudate nucleus compete a high density of the hippocampus entails a relative lower density of the caudate nucleus and in turn there is an associated bias towards the hippocampal spatial strategy. In our memory athletes, we found the opposite pattern: the volumes of the right posterior hippocampus and right caudate nucleus were positively correlated; this correlation was significantly reduced and not apparent in matched controls (Fig.
2). If competition between memory systems leads to an inverse structural relation as described above, cooperation could lead to a positive association between structures. As for an example reported by Voermans et al. (
2004), if the hippocampus is more dominant in a competing scenario it will suppress the caudate nucleus. In a cooperative scenario, it would support it. Whereas the matched controls do not utilise the two systems together frequently, the memory athletes do so, which goes hand in hand with a correlation between the structures involved. The structural consequences of this competition have recently been demonstrated by using video games as a model for spatial navigation (West et al.
2017). Players that relied on stimulus-response strategies showed a reduction in hippocampal grey matter, whereas players with a spatial strategy showed an increase.
The biggest volumetric difference in the memory athletes is the enlarged right anterior hippocampus. Since the work on taxi drivers’ navigational memory (Maguire et al.
2000; Woollett
2011), we know that the hippocampus remains plastic even after maturation. Extensive training in the method of loci could have similar neuroanatomical consequences for memory athletes as the acquisition of navigational memory in taxi drivers. However, since we do not have longitudinal data, we can only speculate whether enlarged hippocampi were a prerequisite or a consequence of the participants becoming world class memory athletes. For the taxi drivers, the hippocampal growth was linked to the acquisition of the complex street layout of London. As we lack a clear intervention in the memory athletes, we can only speculate about the differences in hippocampal volume. One facility that is central to the mnemonics utilised by the memory athletes is the ability to integrate information to enhance remembering it. During the method of loci, athletes have to transform the information they need to remember in a vivid image which is then associated with one of the route points of a very familiar environment. This function of integrating separate elements into a coherent visual scene has been linked to the anterior hippocampus (Zeidman et al.
2015; Zeidman and Maguire
2016).
The structural differences in the hippocampus and the association to the world ranking in the memory athletes was mostly right lateralised. For the caudate nucleus, results seemed stronger for the right hemisphere, however, they did not significantly differ. The right lateralisation for the hippocampus is in line with a substantial body of work showing the right hippocampus to more strongly implicated with spatial processing (Bohbot et al.
1998; Burgess et al.
2002; Kühn and Gallinat
2014; Postma et al.
2008). As the method of loci is a dominantly spatial one, it fits that the right side is more strongly implicated in the exceptional memory exhibited by the memory athletes we studied.
One central limitation of our study is that we did not investigate subdivisions of the caudate nucleus. From animal work, we know that there is spatial differentiation within the caudate nucleus in terms of cooperation and competition (Packard et al.
1989; Sabatino et al.
1992; McDonald and White
1993; Devan et al.
1999). Therefore, for future work it is important to use methods complementary to volumetry as we applied here. For example, voxel-based morphometry or shape analysis could help to dissociate cooperative from competitive sub-regions of the caudate nucleus. Another limitation is that we do not know how specific the cooperation of the caudate nucleus and the hippocampus is for the method of loci. Given how well video games might serve as a model for these spatial learning strategies (West et al.
2017), it might be interesting to have participants play a game that can best be performed if both strategies are integrated, as they are in the method of loci.