Our task, which required encoding of items and associations between items during navigation as well as recognition of those items and their associations, provided an opportunity to contrast the neural basis of item-related and experimental context-related representations. Our paradigm also permitted us to contrast representations formed during navigation with representations accessed during retrieval. It is thus notable that a high percent (29%) of neurons showed changes in firing rate for viewing specific items during encoding which persisted through retrieval, despite differences in experimental context between the different tasks. These responses were typically specific to a few items, suggesting that item-specific representations are formed in the hippocampal area. While selective, item-specific neural responses have been reported previously in the MTL during viewing of famous people and other objects (Heit et al., 1988
; Kreiman et al., 2000
; Quiroga et al., 2005
), we show here that these representations are invoked during both item-learning and during learning of episodes involving formation of item-context associations.
We also found greater numbers of neurons responding during viewing of stores in the ERC compared with the hippocampus and amygdala. We previously described neurons present in the ERC and parahippocampal gyrus (parahippocampal region) that increased firing during viewing of landmarks during navigation (Ekstrom et al., 2003
). Because we did not find significant numbers of store-responsive neurons firing during a control task in which subjects viewed stores and read their names out loud prior to navigation (Ekstrom et al., 2003
), our results suggest that ERC neurons were specifically involved in encoding spatial landmarks. Our current results extend our previous findings to show that neurons in the ERC increase firing rate to specific stores and maintain their firing during both simple recognition of these objects and during retrieval of these objects along with the context in which they were encoded. Thus, our results show that the same ERC neurons that encode spatial landmarks during navigation maintain their altered firing rates when these stores are viewed during retrieval.
In contrast to the significant changes in neural firing rate for items encountered during navigation and retrieval, we did not find significant numbers of item-specific broadband LFPs changing during navigation and retrieval. These data suggest that during navigation and retrieval, the broadband LFP does not convey information about specific stimuli. We did, however, observe item-specific responses in θ- and γ-band. Previous work demonstrated increased θ and γ amplitude during encoding of items that are subsequently recalled (Sederberg et al., 2003
; Mormann et al., 2005
). Because we did not find a significant correlation between neural item responses and θ-band and γ-band LFP responses, our results suggest that θ-band and γ-band LFPs do not directly relate to neural firing rate changes.
We did observe significant LFP power changes to experimental contexts in all frequency bands. A significantly greater number of broadband, θ-band, and γ-band LFPs were active while subjects viewed passengers and stores during retrieval compared with encoding. In contrast, we found significantly greater numbers of neurons active while subjects viewed stores and passengers during encoding compared with retrieval. It is notable that we again dissociated neurons and LFPs based on encoding vs. retrieval. It is possible that the self-timed nature of navigation led to a greater dissociation of LFPs during encoding (while navigating) compared with retrieval. Previous scalp EEG recordings, however, during a timed, continuous recognition task demonstrate greater ERP amplitude during retrieval compared with encoding (Finnigan et al., 2002
). Also, Kahana et al. (1999)
observed greater θ oscillatory power during retrieval of a spatial locations while recalling learned navigational routes than during encoding of these routes (Kahana et al., 1999
). Thus, given that previous studies have also reported greater LFP power during retrieval compared with encoding we believe that our finding of greater broadband, θ, and γ-band LFP power during retrieval compared with encoding is therefore not specific to the self-paced nature of navigation.
While we found greater numbers of electrodes active during retrieval compared with encoding in all frequency bands, we also observed greater numbers of broadband LFPs responding during associative compared with item recognition. This finding further argues that the greater broadband LFP power we observed during retrieval when compared with encoding relates to retrieval-specific processes. Because we found greater numbers of electrodes responding during associative compared with item recognition in the hippocampus and ERC compared with the amygdala, our data further argue that retrieval-specific increases in LFPs during associative retrieval may be unique to memory processing in these two regions. Given that we observed this effect in the LFP, a reflection of synaptic input (Mitzdorf, 1985
; Logothetis, 2003
), but not in neural firing rate, our data further suggest that retrieving associations requiring both item and source information leads to greater input of synaptic activity in the hippocampus and ERC compared with item recognition.
A final question we wished to address was the simultaneous relation between neurons and LFPs during memory processing. Previous recordings from human auditory cortex suggest a high correlation between neural firing rate and γ-band LFPs during listening to complex sounds (Mukamel et al., 2005
). In the Mukamel et al. (2005)
experiment, however, large numbers of neurons were simultaneously active during the task, suggesting coordinated input and output processes (Logothetis, 2003
). In our experiment, few neurons responded at any given time to items due to their selectivity for certain items and not others. It is thus not surprising that we observed little or no population correlation between neurons and LFPs for items in any frequency band. Although we observed significant θ- and γ-band responses to items, cellular responses to items did not correlate with either θ- or γ-band LFPs. These data suggest that neural and θ/γ band LFP responses were not associated in time, and that θ/γ band LFP responses were to different objects at different times than cellular responses. On the basis of our dissociation for neural responses during encoding and LFP responses during retrieval, we tentatively suggest that our item-related neural responses relate to item-maintenance, while LFP responses may relate to attentional process involved in processing specific objects in memory.
The lack of a correlation between experimental context-related neurons and LFPs is somewhat more surprising because these experimental context-related responses were active over an entire block of the experiment (e.g., during navigation or retrieval). Given, however, that neurons and LFP power (across all bands) increased during complementary phases of the experiment, with neurons increasing firing during encoding, and LFPs increasing in power during retrieval, this lack of a correlation is not surprising based on our behavioral results. In support of our finding of a dissociation between LFP power and neural firing during navigation, a recent spatial learning study in rodents by Robbe et al. (2006)
similarly observed little correlation between LFP power and neural firing rate (e.g., Robbe et al., ), although the authors did not look specifically at encoding or retrieval of landmarks (Robbe et al., 2006
It is important to note here that individual neurons we recorded from may show strong coupling in the γ frequency range even though the behaviorally-responsive population we analyzed, on average, did not. In fact, previous results suggest that subsets of neurons do in some case phase lock with the ongoing γ oscillation in rat and human hippocampus (Bragin et al., 1995
; Jacobs et al., 2007
). Our findings, however, showed that over the population of item and experimental context responses, there was little correlation with γ oscillatory power, a result bolstered by our behavioral findings. Because our primary interest was the relationship between the population of simultaneously recorded neurons and LFPs in the hippocampal area for comparison with previous studies (Buchwald et al., 1965
; Wyler et al., 1982
; Logothetis et al., 2001
; Logothetis, 2003
; Nase et al., 2003
; Mukamel et al., 2005
; Kreiman et al., 2006
), we did not focus on individual neurons that showed strong coupling or anticoupling with γ. What we wish to emphasize is that the correlation over the population of neurons and LFPs in our study was significantly lower than that observed in previous studies in other brain regions, despite the similarity of our methods to previous studies. Correlations between cellular firing rate and LFPs over the population of recording sites in auditory cortex of humans and infero-temporal cortex of monkeys, particularly in the γ-band, are typically significantly higher than what was observed in this study, reported in the range of 0.2–0.8 (Mukamel et al., 2005
; Kreiman et al., 2006
The correlations reported in this study in the hippocampal area and amygdala, in contrast, did not differ significantly from zero. These data therefore suggest that the hippocampus, ERC, and amygdala have a different functional-anatomical layout than sensory regions in the brain. In support of this notion, recordings from rodent hippocampus demonstrate no correlation between neighboring neurons and their behavioral correlates (Redish et al., 2001
), in contrast to the visual cortex, where neurons responding to the orientation of bars are arranged in hypercolumns based on their preferred orientation (Mountcastle, 1997
). The fact that increased selectivity of neurons is often accompanied by decreases in global activity in the hippocampus (Hirase et al., 2001
) further supports our finding of no significant LFP change during selective neural epochs in the hippocampal area.
Because of the behavioral dissociations for neural firing rate and LFPs, and the nonsignificant correlations we found between cellular firing rate and LFPs in the hippocampus compared with the larger correlations reported in previous studies in other brain regions (Mukamel et al., 2005
; Kreiman et al., 2006
), we believe the differences we observed between cellular firing rate and LFPs also in part relate to differences in what these two signals represent. LFPs likely reflect the input to a brain region because they largely represent the summation of excitatory synaptic events (e.g., primarily EPSPs) due to afferent input (Mitzdorf, 1985
). Neural firing rate (e.g., action potentials), however, represents the computations and therefore the output of a region (Koch and Segev, 2000
; Logothetis, 2003
). Our study thus provides preliminary evidence for a distinction between input-related LFP responses and output-related spiking activity during declarative memory processes in the human hippocampus. We suggest therefore that our findings, in conjunction with previous electrophysiological studies showing stimulation-induced differences between LFP-related input and neuron-related output (Mathiesen et al., 1998
; Lauritzen, 2001
; Lauritzen and Gold, 2003
), support distinct functional roles for ensemble activity (LFP and the fMRI BOLD signal) and cellular firing rate in the human hippocampus during declarative memory processes.