Introduction
Neuroimaging research holds much promise for elucidating the functioning principles of episodic memory, including neural signatures that might distinguish between true and false memories (Schacter, Reference Schacter1999). Over the past two decades, the DRM paradigm (Roediger & McDermott, Reference Roediger and McDermott1995) has been used extensively in studies with children and adults (see Brainerd, Reyna, & Ceci, Reference Brainerd, Reyna and Ceci2008, and Gallo, Reference Gallo2010, for reviews), as well as with clinical populations (e.g., Paz-Alonso, Ghetti, et al., Reference Paz-Alonso, Ghetti, Ramsay, Solomon, Yoon, Carter and Ragland2013). One advantage of this paradigm is that it reliably induces robust false recognition by probing processes that are central to memory distortion (Zhu, Cheng, Loftus, Lin, & Dong, Reference Zhu, Cheng, Loftus, Lin and Dong2013) without relying on social pressure as other paradigms do (e.g., Loftus & Pickrell, Reference Loftus and Pickrell1995).
In the DRM paradigm, participants study several word lists (e.g., bed, rest, tired, dream, wake,….), converging on a semantic theme captured in a word that is never studied (i.e., sleep; critical lure, CL); participants then perform an old/new recognition test that includes studied words (i.e., bed; targets), CLs (i.e., sleep), and other lures that are non-semantically associated with the studied materials (i.e., flower; unrelated lures, ULs). Adults are frequently as likely to falsely recognize CLs as they are to correctly recognize studied words (McDermott & Roediger, Reference McDermott and Roediger1998). Age-related increases in both true and false recognition have been consistently reported in behavioral studies (e.g., Brainerd et al., Reference Brainerd, Reyna and Ceci2008; Howe, Cicchetti, Toth, & Cerrito, Reference Howe, Cicchetti, Toth and Cerrito2004; but see Carneiro, Albuquerque, Fernandez, & Esteves, Reference Carneiro, Alburquerque, Fernandez and Esteves2007; Ghetti, Qin, & Goodman, Reference Ghetti, Qin and Goodman2002), suggesting that access or activation of semantic associations becomes more automatic during middle childhood (e.g., Wimmer & Howe, Reference Wimmer and Howe2009).
Moreover, the DRM paradigm has been used in several neuroimaging studies to investigate the neural signatures of true and false memory (e.g., Atkins & Reuter-Lorenz, Reference Atkins and Reuter-Lorenz2011; Cabeza, Rao, Wagner, Mayer, & Schacter, Reference Cabeza, Rao, Wagner, Mayer and Schacter2001). The fMRI research using this paradigm with children and adults has typically revealed activations in a left-lateralized set of regions, including lateral PFC, lateral temporal cortex, and parietal cortex (Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Konishi, Wheeler, Donaldson, & Buckner, Reference Konishi, Wheeler, Donaldson and Buckner2000; Paz-Alonso, Ghetti, Donohue, Goodman, & Bunge, Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008; Slotnick & Schacter Reference Slotnick and Schacter2004). Neuroimaging research using this and other paradigms has also demonstrated that the medial temporal lobes (MTL) are involved in the recollection of semantic and sensory properties (e.g., Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Eichenbaum, Yonelinas, & Ranganath, Reference Eichenbaum, Yonelinas and Ranganath2007; Stark & Squire Reference Stark and Squire2000). For example, Cabeza et al. (Reference Cabeza, Rao, Wagner, Mayer and Schacter2001) reported functional differences between anterior and posterior hippocampus suggesting that the former supports the recovery of semantic information and the latter the recovery of specific perceptual features.
The anterior MTL, including hippocampus and perirhinal cortex has been considered a key component of the semantic system (Patterson, Nestor, & Rogers, Reference Patterson, Nestor and Rogers2007), because of its involvement in extracting systematic associations of features that define objects (Henson & Gagnepain, Reference Henson and Gagnepain2010). Unlike the posterior portions of the MTL typically involved in the processing of contextual features (Diana, Yonelinas, & Ranganath, Reference Diana, Yonelinas and Ranganath2007; Ranganath et al., Reference Ranganath, Yonelinas, Cohen, Dy, Tom and D'Esposito2004), the anterior MTL has been associated with familiarity (Henson & Gagnepain, Reference Henson and Gagnepain2010), including illusory familiarity to novel stimuli (Abe et al., Reference Abe, Okuda, Suzuki, Sasaki, Matsuda, Mori and Fujii2008; see also Chadwick, Hassabis, Weiskopf, & Maguire, Reference Chadwick, Hassabis, Weiskopf and Maguire2010).
Recent pediatric functional neuroimaging research has suggested differential involvement of anterior and posterior hippocampus in memory operations (DeMaster & Ghetti, Reference DeMaster and Ghetti2013; Ghetti, DeMaster, Yonelinas, & Bunge, Reference Ghetti, DeMaster, Yonelinas and Bunge2010; Maril et al., Reference Maril, Davis, Koo, Reggev, Zuckerman, Ehrenfeld and Rivkin2010; Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008). The relevance of these findings is underscored by evidence of distinct trajectories of structural development of hippocampal subregions along its anterior–posterior axis (DeMaster, Pathman, Lee, & Ghetti, Reference DeMaster, Pathman, Lee and Ghetti2013; Gogtay et al., Reference Gogtay, Nugent, Herman, Ordonez, Greenstein, Hayashi and Thompson2006).
Structural and functional differences between anterior and posterior hippocampus suggest a division of labor in memory processes which might depend, at least in part, on their connectivity with different cortical regions (Poppenk & Moscovitch, Reference Poppenk and Moscovitch2011). To further understand how differences in connectivity may be relevant for true and false memory, it is helpful to examine what regions have been previously associated with these phenomena. The only neurodevelopmental imaging study using the DRM paradigm (Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008) found a developmental progression in the functional engagement of left anterior hippocampus, with 8-year-olds recruiting this region for correct identification of new ULs, 12-year-olds for true memories as well as for the correct rejection of ULs, and adults for distinguishing true from false memories. This study also found age-related differences in left-lateralized cortical regions found in prior DRM retrieval studies with adults [i.e., middle temporal gyrus (MTG, BA21), posterior parietal cortex (PPC, BA7), ventrolateral PFC (BA47), and dorsolateral PFC (dlPFC, BA46)], which are thought to support recollection, semantic processing, and memory monitoring during true and false episodic retrieval (e.g., Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Kim & Cabeza, Reference Kim and Cabeza2007; Okado & Stark, Reference Okado and Stark2003; Slotnick & Schacter, Reference Slotnick and Schacter2004). Overall, this previous research provides us with a reasonable basis to guide the identification of relevant connectivity nodes for the present study.
The central goal of the present research is to expand our previous work by examining age-related differences in connectivity between anterior and posterior parts of the hippocampus and cortical regions during true and false recognition. In an analysis of functional connectivity during resting state, Kahn, Andrews-Hanna, Vincent, Snyder, and Buckner (Reference Kahn, Andrews-Hanna, Vincent, Snyder and Buckner2008) found that the anterior hippocampus was strongly correlated with regions in the lateral temporal cortex; in contrast, the more posterior parts of the hippocampus were correlated with lateral parietal cortex, posterior parahippocampal cortex, and regions along the posterior midline and ventral medial PFC. This posterior pathway is involved in recollection (Ranganath & Ritchey, Reference Ranganath and Ritchey2012). These connectivity results concerning the hippocampal sub-regions map well onto the regions engaged in memory retrieval from fMRI studies reviewed earlier.
There is now a small set of studies documenting age-related and task-related differences in functional connectivity between MTL and PFC regions during memory encoding (Menon, Boyett-Anderson, & Reiss, Reference Menon, Boyett-Anderson and Reiss2005) and among MTL, PFC, and PPC during retrieval (Ofen, Chai, Schuil, Whitfield-Gabrieli, S., & Gabrieli, Reference Ofen, Chai, Schuil, Whitfield-Gabrieli and Gabrieli2012; Paz-Alonso, Bunge, Anderson, & Ghetti, Reference Paz-Alonso, Ghetti, Ramsay, Solomon, Yoon, Carter and Ragland2013). However, no study has examined developmental differences in functional connectivity during true and false memory. The main goal of the present research is to examine developmental differences in how hippocampal activity is coupled with activity in other cortical regions involved in true- and false-memory retrieval.
Based on the literature discussed thus far, we predicted that the anterior and posterior hippocampus may be differentially connected with various cortical regions supporting memory, and that the anterior hippocampus may be particularly involved in the DRM task because it might preferentially support processing of semantic relationships (Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001) in connection with other ventral regions, such as MTG (Dennis, Bowman, & Vandekar, Reference Dennis, Bowman and Vandekar2012; Kahn et al., Reference Kahn, Andrews-Hanna, Vincent, Snyder and Buckner2008). In contrast, consistent with recent research (Dennis et al., Reference Dennis, Bowman and Vandekar2012), we expected false recognition to show be associated with a more distributed set of regions including frontal parietal, and posterior cingulate regions.
To help characterize age differences in connectivity in true and false memory retrieval, we manipulated list associative strength. This is relevant because false memories depend on associative activation in the DRM paradigm (e.g., Howe, Reference Howe2005; Howe, Wimmer, & Blease, Reference Howe, Wimmer and Blease2009), and children's domain-specific associative connections among related concepts (as well as the automaticity of the activation of these concepts) develop during middle childhood (Wimmer & Howe, Reference Wimmer and Howe2009). Therefore, we used high- and low-strength DRM lists based on association strength norms (Stadler, Roediger, & McDermott, Reference Stadler, Roediger and McDermott1999). The comparison of age groups in the processing of related concepts, differing in strength of semantic associations, allow for shedding new light onto whether developmental differences in behavioral performance in the DRM paradigm stem from differences in automaticity of access and activation of semantic concepts.
Methods
Participants
Our sample consisted of 31 right-handed native English-speaking participants: fourteen 8- to 9-year-olds (M = 9.07 years; range = 8.03–9.82; 6 females) and 17 young adults (M = 21.21 years; range = 19.30–27.62; 9 females). Additionally, data from 10 participants were excluded from analysis due to excessive head motion (6 children and 1 adult whose head motion parameters exceeded 3 mm within at least one functional scan), technical difficulties during fMRI data acquisition (2 adults), or failure to understand the task (1 child who responded with the same button to most of the trials). Participants received either monetary compensation or course credit for their participation. Before participating, informed consent was obtained based on procedures approved by the UC Davis IRB. Children were prescreened with the Child Behavior Checklist (Achenbach, Reference Achenbach1991). We note that none of the participants in the present study participated in our prior developmental DRM study (Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008); however, we included 8- to 9-year-olds and adults here, corresponding to the youngest and oldest group tested in Paz-Alonso et al. (Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008) to better relate the present results to our previous research.
Task and Procedure
Eighteen lists of 12 words each were adapted for use with children and adults from materials used previously in the DRM experimental paradigm (Roediger & McDermott, Reference Roediger and McDermott1995). These lists were selected based on associative strength norms and their effectiveness in producing false recognition to CLs (Stadler et al., Reference Stadler, Roediger and McDermott1999). Half of the DRM lists were classified as having high-associative strength with false-recognition rates ranging from .71 to .84 (M = .80 ± .05). The remaining nine lists were classified as having low-associative strength, with false-recognition rates ranging from .33 to .64 (M = .55 ± .11). These high-associative versus low-associative DRM lists differed statistically in their effectiveness to produce false recognition, t(17) = 11.93; p < .001. Because of this associative strength manipulation, most of the lists used in the present study were different from those used in our previous developmental neuroimaging study (Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008).
During the study phase, participants studied the 18 word lists. Words within each list were presented in order of decreasing associative strength. Participants were instructed to do their best to remember each word. Lists were presented auditorily at a rate of one word every 2500 ms, and the presentation order of the lists was randomized.
The time interval between the completion of the study phase and the beginning of the retrieval phase was 10 min. In preparation for the test phase, participants were instructed to respond “yes” if they remembered the word from the study session, or “no” if they did not. The recognition test included a total of 162 words: 54 studied items (Targets), 54 non-studied semantic associates (CLs), and 54 new unrelated items (ULs). Targets consisted of 3 studied items from each of the 18 lists (from serial position 1, 7 and 9) and CLs were the 1st, 4th, and 9th associate from each original DRM list, which were not presented during the study session. For example, from the original DRM list “sleep, bed, rest, awake, tired, dream, wake, snooze, blanket, doze, slumber, snore….”, the selected CLs to-be-presented only at retrieval were “sleep”, “awake”, and “blanket.” From the studied list of words “bed, rest, tired, dream, wake, snooze, doze, slumber, snore….”, the target items selected to-be-presented at retrieval were “bed”, “doze”, and “snore.” Finally, ULs were selected from non-semantically related words based on the MRC Psycholinguistic Database. These ULs matched the corresponding CLs from studied lists of high and low association strength in frequency, familiarity, concreteness, and age of acquisition norms.
The 162 trials were presented in two separate functional runs, with 16 randomized orders. First, participants viewed a drawing depicting an ear for 1500 ms, which signaled that a word was about to be presented auditorily. Next, the words “Yes” and “No” were projected on the screen for 2000 ms, instructing participants to respond by making left-handed key presses on a two-button fiber-optic box. Finally, a fixation cross-bar was displayed for 500 ms before the start of the next trial. Both encoding and retrieval phases were conducted while functional data were acquired; results from the encoding phase are not included in the present report.
fMRI Data Acquisition
Whole-brain fMRI data were acquired with a 3-Tesla Siemens TRIO whole-body MRI scanner (Siemens Medical Solutions, Erlangen, Germany) at the UC Davis Imaging Research Center using a standard whole-head coil. Functional images during retrieval were acquired in two separate runs using a gradient-echo echo-planar pulse sequence (interleaved acquisition, repetition time [TR] = 2000 ms; echo time [TE] = 25 ms; 35 axial slides; 2.75 × 2.75 × 3.4 mm; no inter-slice gap; flip angle = 90°; field of view = 220 mm; 185 volumes/run). High-resolution T1-weighted anatomical images were also collected. To limit head movement, the area between participants’ heads and the head coil was padded with foam. Snugly fitting headphones dampened background scanner noise and enabled auditory stimulus presentation and communication with experimenters.
Preprocessing of fMRI Data
Standard SPM5 (Wellcome Department of Cognitive Neurology, London) preprocessing routines and analysis methods were used. Images were corrected for differences in timing of slice acquisition, followed by rigid-body motion correction. Structural and functional volumes were spatially normalized to T1 and echo-planar imaging templates, respectively. The normalization algorithm used a 12-parameter affine transformation together with a nonlinear transformation involving cosine basis functions. During normalization, the volumes were resampled to 3 × 3 × 3 mm3 voxels. Templates were based on the MNI305 stereotaxic space (Cocosco, Kollokian, Kwan, & Evans, Reference Cocosco, Kollokian, Kwan and Evans1997). These procedures have been validated with children ages 6 years and above (e.g., Burgund et al., Reference Burgund, Kang, Kelly, Buckner, Snyder, Petersen and Schlaggar2002; Kang, Burgund, Lugar, Petersen, & Schlaggar, Reference Kang, Burgund, Lugar, Petersen and Schlaggar2003). After normalization, functional volumes were spatially smoothed with an 8-mm full width at half maximum isotropic Gaussian kernel.
fMRI Data Analysis
Statistical analyses were performed on individual participants’ data using the general linear model (GLM). The fMRI time series data were modeled by a series of impulses convolved with a canonical hemodynamic response function (HRF). In addition to excluding participants whose head motion exceeded 3 mm in any parameter within at least one functional scan, we also included motion parameters for translation (i.e., x, y, z) and rotation (i.e., yaw, pitch, roll) as covariates of noninterest in the GLM. Although additional approaches to deal with age differences in motion have been proposed (Power, Barnes, Snyder, Schlaggar & Petersen, Reference Power, Barnes, Snyder, Schlaggar and Petersen2012; Van Dijk, Sabuncu, & Buckner, Reference Van Dijk, Sabuncu and Buckner2011), further research is still needed to determine how micro-movement artifacts should be examined (Power et al., Reference Power, Barnes, Snyder, Schlaggar and Petersen2012), thereby justifying our use of the standard approach.
Each trial was modeled as an event, time-locked to the onset of the cue period. The resulting functions were used as covariates in a GLM, along with a basic set of cosine functions that high-pass filtered the data. The least-squares parameter estimates of the height of the best-fitting canonical HRF for each condition were used in pairwise contrasts. Contrast images, computed on a participant-by-participant basis, were submitted to group analyses.
At the group level, whole-brain contrasts between conditions were computed by performing one-sample t tests on these images, treating participants as a random effect. Our standard statistical threshold for All > Null contrast was a False Discovery Rate (FDR) set to q < .01, with of at least 10-contigous voxels extent threshold. All brain coordinates are reported in MNI atlas space (Cocosco et al., Reference Cocosco, Kollokian, Kwan and Evans1997). These contrasts provided the basis for identifying cortical regions to submit to functional connectivity analysis.
Based on evidence from neuroimaging DRM studies with children and adults (e.g., Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Kim & Cabeza, Reference Kim and Cabeza2007; Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008), we conducted functional connectivity analysis on regions-of-interest (ROIs) in left anterior hippocampus (−25, −12, −18; 3728 mm3), left posterior hippocampus (−22, −38, 2; 1376 mm3), left pars triangularis/opercularis (−45, 18, 18; 10152 mm3), left pars orbitalis (−37, 27, −5; 344 mm3), left dlPFC (−43, 13, 39; 1304 mm3), left MTG (−58, −33, 4; 4208 mm3), right MTG (64, −34, 2; 912 mm3), and left PPC (−28, −56, 47; 2628 mm3), with the MARSBAR toolbox for use with SPM (Brett, Anton, Valabretue, & Poline, Reference Brett, Anton, Valabregue and Poline2002). All cortical ROIs consisted of active voxels identified from All > Null across all participants, FDR corrected q < .01 with a 10-voxel threshold extent within the specific MARSBAR anatomical ROI defined above (Supplementary Table 1). In contrast, hippocampal ROIs were not identified functionally from the All > Null contrast, but were anatomically defined. The absence of reliable hippocampal activations in this contrast is not surprising given the absence of an active baseline condition in the present research (Stark & Squire, 2001). Furthermore, our previous studies suggest that anatomically defined hippocampal ROIs may be more sensitive to age differences (e.g., DeMaster & Ghetti, Reference DeMaster and Ghetti2013).
Functional connectivity analysis was conducted via the beta-series correlation method (Rissman, Gazzaley, & D'Esposito, Reference Rissman, Gazzaley and D'Esposito2004), implemented in SPM5 with custom Matlab scripts. The canonical HRF in SPM was fit to each occurrence of each condition, and the resulting parameter estimates (i.e., beta values) were sorted according to the study conditions to produce a condition-specific beta series for each voxel. For each participant, we calculated the beta-series correlation between each ROI for each condition (high and low associative strength) across all types of responses (e.g., hits, false alarms) and for true and false recognition specifically. These indices were used for our analyses.
First, we sought to identify significant coupling strength between ROIs across all the main study conditions. To do so, beta-series correlation values (r values) were averaged across conditions and participants. Given our sample size (N = 31), r-values ≥.36 and <.46 corresponded to a p-value ≤.05, r-values ≥.46 and <.57 corresponded to a p-value ≤.01, and r-values ≥.57 corresponded to a p-value ≤.001; all of them two-tailed. Second, we sought to identify significant coupling strength between ROIs for true and false recognition and high and low association lists in children and adults; to do so, we averaged beta-correlations within each age group separately for each combination of factors (i.e., true-recognition/high-association; true-recognition/low-association, false-recognition/high-association; false-recognition/low-association).
Finally, we examined differential coupling strength between pairs of ROIs (i.e., nodes) during true and false recognition, as a function of Age and Associative Strength. To do so, r values for each pair of ROIs for each participant and condition were converted to Fisher's Z normally distributed values and submitted to t tests. Differential coupling strength was examined only for those pairs of nodes for which coupling strength was significant in at least one of the age groups for a given combination of factors (e.g., false-recognition/high-association). All the reported significant differences in strength of coupling had a Δr value between an of .10 to .25.
Results
Behavioral Results
Recognition memory
A 2 (Age: children vs. adults) × 2 (Associative Strength: low vs. high) × 3 (Response: targets, CLs, ULs) mixed analysis of variance revealed the significant main effect of Associative Strength, F(1,29) = 8.15; p < .01; ηp2 = .22, such that across ages, the proportion of recognized items was higher for high-association lists (M = 47.93 ± 2.09) compared to low-association lists (M = 44.65 ± 2.38). Moreover, a significant main effect of Response was found, F(2,58) = 196.18, p < .001, ηp2 = .87, which was qualified by a significant Age × Response interaction, F(2,58) = 3.25, p < .05, ηp2 = .10 (Figure 1).
Behavioral results. Mean proportion of recognition as a function of age, response type (Hits: CL FAs, or false alarms to critical lures; UL FAs, or false alarms to unrelated lures), and association strength.
Simple-effects analyses revealed that, compared to children, adults produced a higher proportion of hits, F(1,29) = 6.05, p < .05, ηp2 = .17, and false alarms to CLs, F(2,58) = 4.01, p = .05, ηp2 = .12. In contrast, both age groups falsely recognized ULs at comparable rates, F < 1. Thus, adults exhibited higher true recognition and false recognition of CLs than children. For response time results see Supplementary Materials.
fMRI Results
Our analytical approach for the fMRI data included whole-brain contrasts to identify cortical regions generally involved in the task and functional connectivity analyses to establish their functional relation with anatomically defined hippocampal regions. Whole-brain analysis for All > Null across all participants revealed activation in bilateral dlPFC (BA9/46) and pars opercularis, triangularis and orbitalis (BA44/45/47), premotor cortex and frontal eye fields (BA6/8), bilateral anterior cingulate cortex (BA24/32), bilateral middle and superior temporal cortex (BA22/21), bilateral superior temporal pole (BA38), left parietal cortex (BA7/40), and subcortical regions including insula, putamen, and caudate (Supplementary Table 1 reports the results of this whole-brain contrast and other contrasts capturing the discrimination between true and false memories).
Functional Connectivity Supporting True and False Recognition
To better understand the functional dynamics among the regions involved in this task, we first sought to characterize the temporal coupling between brain regions across study conditions. To do so, we extracted pairwise correlations between ROIs for each participant and response condition (e.g., Hits, false alarms to CLs, correct rejections of CLs, correct rejections of ULs), and averaged them across participants, response, and list associative strength (Figure 2).
Functional connectivity analyses for the main study conditions. (A) Axial and sagittal views show statistically significant pairwise correlations between regions of interest, including left anterior hippocampus (L. Ant. Hip.), left posterior hippocampus (L. Post. Hip.), left pars triangularis/opecularis (Tri/Oper), left pars orbitalis (Orb.), left dorsolateral prefrontal cortex (dlPFC), bilateral medial temporal lobe (MTL), and left posterior parietal cortex (PPC). (B) Correlation matrix showing the size of the correlations between pairs of nodes. Stronger pairwise correlations, averaged across participants and conditions, are shown in deeper blue. BrainNet Viewer was the network mapping tool used for this figure (Xia, Wang, & He, Reference Xia, Wang and He2013).
This analysis revealed that activity in the left anterior hippocampus was tightly coupled with activity in the posterior left hippocampus and with the left MTG. Left MTG was tightly coupled with regions within the left ventrolateral PFC (i.e., pars orbitalis, pars triangularis/opercularis) and with the contralateral right MTG (all ps < .05). Right MTG also exhibited statistically significant coupling with the left pars triangularis/opercularis (ps < .05). Finally, we observed significant functional connectivity between left fronto-parietal nodes, including PPC, dlPFC and pars triangularis/opercularis (ps < .01).
True recognition
We then examined the temporal coupling between nodes involved in true recognition and tested for differences in these pairwise functional correlations as a function of age and list associative strength. During true recognition, significant functional connectivity among temporal regions was revealed such that the left anterior hippocampus, but not posterior hippocampus, was functionally coupled with the left MTG, which was connected with its contralateral counterpart (ps < .05; Figure 3A). Fronto-parietal regions were also tightly connected, including dlPFC, pars triangularis/opercularis, pars orbitalis and PPC nodes (ps < .01; Figure 3A). However, none of the nodes within the set of temporal regions and the set of fronto-parietal regions exhibited significant coupling with regions in the other set.
Functional connectivity analyses for true recognition (Hits). (A) Axial and sagittal views show statistically significant pairwise correlations between regions of interest (ROIs), including left anterior hippocampus (L. Ant. Hip.), left posterior hippocampus (L. Post. Hip.), left pars triangularis/opecularis (Tri/Oper), left pars orbitalis (Orb.), left dorsolateral prefrontal cortex (dlPFC), bilateral middle temporal gyrus (MTG), and left posterior parietal cortex (PPC). The correlation matrix shows the size of the correlations between pairs of nodes. Stronger pairwise correlations during hit responses, averaged across participants, are shown in deeper green. (B) Sagittal views show statistically significant pairwise correlations between ROIs in adults and children for hit responses to items from lists of high associative strength. (C) A sagittal view shows increased strength of coupling for adults compared to children for hit responses to items from lists of high associative strength. BrainNet Viewer was the network mapping tool used for this figure (Xia et al., Reference Xia, Wang and He2013).
To investigate age differences in connectivity for high and low associative strength, average beta-series correlations for each age group at each level of associative strength were computed. Age differences in connectivity were apparent for high associative strength (Figure 3B), which was confirmed when we submitted Fisher's Z values to t tests.
Compared to children, adults exhibited stronger positive anterior hippocampus-dlPFC and anterior hippocampus-PPC coupling for true recognition of studied items from high-association lists (p < .05; Figure 3C). No age differences in coupling strength were observed for low-association lists. Similarly, no differences in coupling strength were observed between high- and low-association lists.
False recognition
Analysis of the functional connectivity during false recognition was conducted with the same methods as used for true recognition. The results, however, showed a more complex pattern with significant coupling strength observed across fronto-temporal-parietal regions (Figure 4A). More specifically, the left anterior hippocampus showed statistically significant temporal coupling with posterior hippocampus, left MTG, and left pars orbitalis (ps < .05). The left MTG was then also tightly coupled with all the PFC nodes (ps < .05). Finally, the left pars triangularis/opercularis exhibited significant functional connections with the remaining PFC nodes (ps < .05), and left PPC (p < .01).
Functional connectivity analyses for false recognition (critical lure false alarms, CL FAs). (A) Axial and sagittal views show statistically significant pairwise correlations between regions of interest (ROIs), including left anterior hippocampus (L. Ant. Hip.), left posterior hippocampus (L. Post. Hip.), left pars triangularis/opecularis (Tri/Oper), left pars orbitalis (Orb.), left dorsolateral prefrontal cortex (dlPFC), bilateral middle temporal gyrus (MTG), and left posterior parietal cortex (PPC). The correlation matrix shows the size of the correlations between pairs of nodes. Stronger pairwise correlations for false alarms to CLs, averaged across participants, are shown in deeper red. (B) Sagittal views show statistically significant correlations between pairs of ROIs in adults and children for false alarms to items from list of high-associative strength. (C) A sagittal view shows increased strength of coupling for children compared to adults for false recognition of CLs from lists of high associative strength. BrainNet Viewer was the network mapping tool used for this figure (Xia et al., Reference Xia, Wang and He2013).
To investigate age differences in connectivity for high and low associative strength, we again used the same methods used for the examination of true recognition. Results again showed that age differences were apparent for high-associative strength lists (Figure 4B), which was confirmed when we submitted Fisher's Z values to t tests.
Children exhibited significantly stronger positive hippocampal-pars orbitalis, and bilateral MTG-pars triangularis/opercularis coupling relative to adults for CLs from high association lists (ps < .05; Figure 4C). In contrast, no age differences were observed for false recognition of CLs in low-association lists. Finally, no differences were observed between high- and low-association lists.
Although the present study was not aimed at examining whether individual differences in coupling strength predicted performance, an exploratory analysis revealed that overall false recognition of CLs was negatively associated with anterior hippocampus-pars orbitalis connectivity across participants, r(29) = −.43, p < .05; this correlation persisted even when participants’ age was partialled out, r(28) = −.41, p = .05. No other significant correlations between behavioral performance and connectivity were found.
Discussion
The goal of the present study was to investigate age-related and semantic strength-related differences in functional connectivity between the hippocampus and cortical regions involved in true and false recognition. Informed by our previous study (Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008) and other neuroimaging studies using the DRM paradigm with adults (e.g., Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Kim & Cabeza, Reference Kim and Cabeza2007), we focused on a left-lateralized set of regions including anterior and posterior hippocampus, lateral temporal cortex, parietal cortex, and lateral PFC. Furthermore, we included right MTG because prior neurodevelopmental evidence suggests that children recruit this region to access semantic representations to support their not yet fully developed semantic system (e.g., Blumenfeld, Booth, & Burman, Reference Blumenfeld, Booth and Burman2006; Chou et al., Reference Chou, Booth, Burman, Bitan, Bigio, Lu and Cone2006). These cortical regions emerged from our initial unbiased whole-brain contrast analysis fully consistent with previous evidence concerning episodic retrieval (e.g., Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Dennis, Kim, & Cabeza, Reference Dennis, Kim and Cabeza2008; Okado & Stark, Reference Okado and Stark2003; Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008).
Previous neurodevelopmental studies have reported functional connectivity analyses between hippocampal regions and frontal, parietal and lateral temporal cortices during memory retrieval in general (e.g., Ofen et al., Reference Ofen, Chai, Schuil, Whitfield-Gabrieli and Gabrieli2012; Paz-Alonso, Bunge, et al., Reference Zhu, Cheng, Loftus, Lin and Dong2013), and during true and false memories in particular (Dennis et al., Reference Dennis, Bowman and Vandekar2012). However, no prior study has examined differences in functional connectivity as a function of age and semantic association strength.
True recognition was associated with coupling among ventral regions including hippocampus and temporal cortices consistent with previous research (Dennis et al., Reference Dennis, Bowman and Vandekar2012), as well as coupling among a more dorsal fronto-parietal set of regions. These sets of regions appeared to be segregated: nodes within hippocampal-temporal and fronto-parietal sets were tightly coupled, but none of the nodes within each of these two sets exhibited significant coupling with nodes in the other set. In contrast, false recognition was associated with a more distributed hippocampal-temporal-parietal-frontal set of regions.
The apparent segregation evident in true recognition might suggest that accurate item retrieval emerges from processes that might be more readily dissociable than those involved in false recognition (e.g., tighter coupling of hippocampus and regions in the MTG might support retrieval of item-specific semantic features; tighter fronto-parietal coupling might support retrieval monitoring or post-retrieval decision processes; e.g., Dobbins, Simons, & Schacter, Reference Dobbins, Simons and Schacter2004; Shannon & Buckner, Reference Shannon and Buckner2004). However, we note that this apparent segregation is no longer evident when the analysis was restricted to high-association lists and when age differences were examined. These findings are discussed next.
Differential Coupling Strength during True and False Recognition as a Function of Age and Association Strength
A direct examination of age differences associated with true recognition showed that adults, compared to children, exhibited stronger hippocampus-PPC and hippocampus-dlPFC connectivity in high association lists. Left PPC activation has been associated with successful episodic retrieval (Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Shannon & Buckner, Reference Shannon and Buckner2004; Slotnick & Schacter, Reference Slotnick and Schacter2004; Yonelinas, Otten, Shaw, & Rugg, Reference Yonelinas, Otten, Shaw and Rugg2005), and amount of retrieved information (e.g., Cabeza, Reference Cabeza2008; Donaldson, Wheeler, & Petersen, Reference Donaldson, Wheeler and Petersen2010; Konishi et al., Reference Konishi, Wheeler, Donaldson and Buckner2000; McDermott, Jones, Petersen, Lageman, & Roediger, Reference McDermott, Jones, Petersen, Lageman and Roediger2000). Furthermore, dlPFC might support subsequent post-retrieval decision processes associated with further attempts to recollect studied items (e.g., Dobbins et al., Reference Dobbins, Simons and Schacter2004; Nolde, Johnson, & D'Esposito, Reference Nolde, Johnson and D'Esposito1998). Finally, there is evidence for age-related differences in activation in left PPC and dlPFC from middle childhood to adulthood during memory retrieval of DRM items (Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008).
Thus, age differences in connectivity might reflect stronger engagement in retrieval searches in adults to support accurate post-retrieval decision processes. Given that adults are more likely to fall prey to false memories with the DRM paradigm, it is conceivable that greater retrieval resources may be necessary to identify true memories. However, these differences might also reflect more frequent recollective experiences in adults compared to children (Ghetti & Angelini, Reference Ghetti and Angelini2008); for example, stronger connectivity between hippocampus and PPC in adults, compared to children, may be suggestive of the stronger capture of attention for vivid memories characteristic of recollection (e.g., Cabeza, Ciaramelli, Olson, & Moscovith, Reference Cabeza, Ciaramelli, Olson and Moscovith2008). Additionally, retrieval of qualitative details has been found to increase with age during childhood for true recognition in the DRM paradigm (Reference Del Prete, Mirandola, Cornoldi and GhettiDel Prete, Mirandola, Cornoldi, & Ghetti, in press) and in general (Ghetti & Angelini, Reference Ghetti and Angelini2008). Our functional connectivity results are fully consistent with these possibilities, and with prior evidence showing age-related increases in hippocampal-PPC and hippocampal-dlPFC functional connectivity during memory retrieval (e.g., Paz-Alonso, Bunge et al., Reference Paz-Alonso, Bunge, Anderson and Ghetti2013). These different, but not mutually exclusive interpretations of age differences in connectivity cannot be disambiguated in the present research.
Age-related differences in functional connectivity also emerged for false recognition of high-associative strength items. In children, compared to adults, activity of the anterior hippocampus was more tightly coupled with activity in pars orbitalis, which is involved in the controlled retrieval of stored conceptual representations (e.g., Badre & Wagner, Reference Badre and Wagner2007); moreover, activity in MTG was more strongly connected with activity in pars triangularis/ operacularis, which are engaged in control processes surrounding the selection of activated representations. Therefore, false recognition of CLs in children seemed to emerge from tighter coupling among regions supporting semantic representation, access, and elaboration of items relative to adults (e.g., Chou et al., Reference Chou, Booth, Burman, Bitan, Bigio, Lu and Cone2006; Kim & Cabeza, Reference Kim and Cabeza2007; Prince, Daselaar, & Cabeza, Reference Prince, Daselaar and Cabeza2005; Wagner, Desmond, Demb, Glover, & Gabrieli, Reference Wagner, Desmond, Demb, Glover and Gabrieli1997). Consistent with this interpretation is the negative correlation across participants between false recognition and hippocampal-pars orbitalis connectivity: just as children, as a group, generally succumbed less to the DRM memory illusion and showed tighter hippocampus-pars orbitalis coupling, individuals who are generally less prone to the illusion exhibit tighter coupling among these nodes. Furthermore, pars orbitalis and MTG exhibited age-related differences in activation profiles (Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008), likely due to increasing automatic access to semantic representations with age (Blumenfeld et al., Reference Blumenfeld, Booth and Burman2006; Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008).
Overall, developmental differences in hippocampal-cortical connectivity were evident for high-associative strength items. For veridical memory, adults exhibited stronger connectivity than children, suggesting stronger involvement of controlled processes and/or stronger recollective experiences in adults. For false memories, children exhibited stronger connectivity than adults, suggesting that access to and elaboration of semantic representations may be less automatic in children.
Role of Anterior Hippocampus in True and False Recognition
Connectivity between hippocampus and cortical regions was evident in both true and false recognition. However, whereas anterior hippocampus was functionally connected with other regions of the temporal, parietal, and prefrontal cortex, the posterior hippocampus was not. Its sole statistically significant coupling was observed with anterior hippocampus. The role of anterior hippocampus in veridical and illusory retrieval of semantically related information is consistent with our hypothesis. Our hypothesis is based on evidence that the anterior hippocampus is involved in episodic retrieval of semantic information (e.g., Abe et al., Reference Abe, Okuda, Suzuki, Sasaki, Matsuda, Mori and Fujii2008; Cabeza et al., Reference Cabeza, Rao, Wagner, Mayer and Schacter2001; Chadwick et al., Reference Chadwick, Hassabis, Weiskopf and Maguire2010; Paz-Alonso et al., Reference Paz-Alonso, Ghetti, Donohue, Goodman and Bunge2008; Prince et al., Reference Prince, Daselaar and Cabeza2005) and is functionally connected to temporal regions storing semantic representations (e.g., Kahn et al., Reference Kahn, Andrews-Hanna, Vincent, Snyder and Buckner2008). These findings are in turn consistent with models proposing that the anterior hippocampus is a key component of an anterior network of regions that support semantic processing, because it can extract and form representations of systematic combinations of item features that co-occur repeatedly and is involved in various cognitive operations including episodic memory (e.g., Ranganath & Ritchey, Reference Ranganath and Ritchey2012).
Future studies should relate the differences in functional connectivity reported here to underlying structural development. The anterior hippocampus has been shown to undergo volumetric changes during childhood (DeMaster et al., Reference DeMaster, Pathman, Lee and Ghetti2013; Gogtay et al., Reference Gogtay, Nugent, Herman, Ordonez, Greenstein, Hayashi and Thompson2006), which might influence network connectivity. Furthermore, white matter tracts connecting the anterior hippocampus with PFC achieve stronger integrity during childhood (Lebel & Beaulieu, Reference Lebel and Beaulieu2011). All of these changes may influence functional connectivity.
Future studies should also characterize individual differences in addition to developmental differences. The preliminary evidence of associations between individual differences in connectivity and individual differences in false recognition, which was the only moderate-size association we found between connectivity and performance, suggests that future studies should be powered to examine individual variation more thoroughly.
Caveats and Conclusions
Before we conclude, we note several caveats in the present study. First, age differences in connectivity were observed with high-association lists. It is not clear why low-association lists did not yield consistent age differences in connectivity. Second, we elected to focus on functional connectivity among left-lateralized regions (except for right MTG). Although this approach is justifiable as discussed earlier, a more holistic approach that includes regions across the entire brain might yield additional or relatively different results. Third, we identified cortical regions to conduct connectivity analyses from all participants and across all conditions; although this approach is unbiased, it selected, by definition, regions that are involved in both true and false recognition in both children and adults, thereby potentially overlooking the contribution of regions uniquely involved in true or false recognition, or regions uniquely recruited by either children or adults. These problems do not apply to hippocampal regions, as they were identified anatomically, but the use of the hippocampal anatomical templates provided by the Automated Anatomic Labeling (AAL) atlas may reduce precision. Fourth, age differences in motion may have affected the current results despite the rigorous measures we used to limit this problem. The fact that adults exhibited stronger hippocampal–cortical connectivity for true recognition, but weaker for false recognition, makes it less likely that movement is a central source of contamination in the present study; if that were the case, adults should have exhibited overall stronger coupling than children between the nodes examined here (Power et al., Reference Power, Barnes, Snyder, Schlaggar and Petersen2012).
Despite these limitations, age-related differences in connectivity in the present study suggest that hippocampal activity might be more or less temporally coupled with cortical regions depending on memory veridicality and strength of semantic associations. These variables may have implications for understanding not only how children discriminate true from false memories, but also for how they extract and retain meaning from such common sources of information such as texts and stories.
Acknowledgments
This research was supported by a grant from the National Institute of Child and Human Development (HD054636-01 to S.G.), Juan de la Cierva and PSI2012-32093 grants from the Spanish Ministry of Economy and Competitiveness, and a grant PI2012-15 from the Department of Education, University and Research from the Basque Government (to P.P.). We thank Silvia A. Bunge for initial discussions on the study design. We also thank the editor of this Symposium, Mary Pat McAndrews, and the editor of JINS, Kathleen Y. Haaland, for their guidance through the preparation of this manuscript. Conflicts of Interest: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
