Brain correlates of declarative memory atypicalities in autism: a systematic review of functional neuroimaging findings

Only one study focused on WM processing of visually presented verbal information, using letters. Other reviewed studies concerned mostly visuospatial information (e.g., colored figures, abstract images, or stimulus location), including faces. Most of these studies reported similar memory performance (thereafter, “performance”) between ASD and TD participants, with different patterns of brain activation and connectivity.

Using an fMRI letter N-back task, Koshino et al. [51] identified in the 2-back condition, a pattern of activation and connectivity being lateralized right or bilateral posterior in adults with ASD relative to TD, with reduced activations in left fronto-parietal areas, also in participants with ASD relative to TD. These results suggest a more visual graphical, i.e., less semantic, processing of letters, in spite of similar between-group performance.

Other studies used visuospatial material. Vogan et al. [52] conducted an fMRI 1-back color matching task in preadolescent children and observed that when the number of different colors constituting a figure increased, ASD relative to TD children showed diminished performance and diminished parietal activation while showing similar activations in bilateral occipital areas and fusiform gyri. In a 2-year follow-up study, Vogan et al. [53] observed in the same participants, a greater load-dependent activation in parietal lobes only in TD controls. The minimal changes they observed in the ASD group were limited to occipito-temporal areas, with diminished performance on the higher load conditions. Results from these two studies suggest preserved activation within the visual occipito-temporal areas associated with the “ventral stream,” while an under-recruitment of parietal areas associated with the “dorsal stream,” with a limited integration of these two streams during visual processing over time. Close to these results, Urbain et al. [54] used a visual N-back task with MEG in children and identified a greater load-dependent activation in the left dorsal parietal cortex in the TD children, while in the left ventral parietal cortex in those with ASD. Despite similar performance, this distinction would reflect a more controlled (i.e., top-down) processing of visual stimuli in the TD group, while being more automatic (i.e., bottom-up) in the ASD participants, in line with the distinction between dorsal and ventral attentional systems within the parietal cortex [74]. In summary, these studies suggest that visual processing during WM tasks is associated with increased activations in ventral areas alongside diminished activation in dorsal areas, in individuals with ASD relative to TD.

Three fMRI studies have documented atypical processing of faces in WM in ASD, reporting similar performance between ASD and TD groups. Herrington et al. [55] conducted a 1-back task in children using superimposed images of neutral or emotional faces and houses, with targets containing either same or different faces or houses. They observed a similar activation of the fusiform gyrus in both groups. By contrast, increased activation of the dorsolateral prefrontal cortex was found in the ASD group when the target contained a face, suggesting selective attention in order to further discriminate faces. For the TD group, increased activation was found when the target contained a house, suggesting selective attention in this group was aimed at disregarding the automatic percept of a face. Koshino et al. [56] used an N-back task with mixed neutral and emotional faces, in adolescents and young adults and observed reduced activations in both left inferior prefrontal areas usually associated with verbal processing and right posterior temporal areas usually associated with theory of mind, in participants with ASD compared with TD. Activations within the bilateral fusiform face areas (FFA), a key structure enabling the processing of basic facial features and also involved in object identification [75], were similar but atypically located in ASD relative to TD (i.e., location of peak activation differed from that of the TD group), and also less connected to frontal areas. In adults, Kleinhans et al. [57] used a 1-back task with neutral faces and reported similar activation of bilateral FFA in ASD and TD groups, but was less functionally connected in ASD with both the subcortical face processing structures (amygdala and superior colliculi) associated with fast emotional attention for faces [76] and the limbic structures (precuneus and posterior cingulate gyri) that enable the processing of emotional stimuli [77]. Overall, these results point toward ASD participants having a relatively similar level of FFA activation, but lower activity and connectivity with brain areas related to the attention to faces and their socio-emotional processing.

Included studies reported reduced activation of prefrontal areas (n = 9), reduced deactivation of the DMN (n = 6), and atypical functioning of the WM-related brain network (n = 7), mainly during WM maintenance.

First, just over half (i.e., 9/16) of fMRI N-back studies found reduced prefrontal activations in ASD relative to TD, associated with diminished or enhanced activations within posterior areas. In studies reporting reduced prefrontal activations in conjunction with enhanced parietal or temporal activation, performance was similar between ASD and TD groups in children and adults [51, 53, 56, 58]. By contrast, other fMRI studies that reported reduced prefrontal activations without enhanced parietal or temporal activations found significantly diminished performance in groups with ASD relative to TD, in children and adults [52, 59‐61]. Close to these findings, in a short report using an fMRI visuospatial mental rotation task requiring manipulation of information, Silk et al. [62] found in adolescents with ASD relative to TD diminished activations within cortical and subcortical (caudate head) frontal regions, with similar parietal activation, while similar between-group performance. Taken together, these results suggest a general, diminished load-dependent recruitment of the prefrontal areas, being partially or fully offset by enhanced activation in posterior sensory and attentional areas.

Second, some studies investigated the DMN, reporting atypical deactivation in the higher load conditions of fMRI N-back tasks in ASD, with similar performance in the ASD and TD groups. In adolescents with ASD relative to TD, Chantiluke et al. [59] found in the verbal 3-back condition, reduced activation of the dorsolateral prefrontal cortex associated with enhanced DMN deactivation, suggesting a compensatory process. In adolescents too, Rahko et al. [63] observed similar activation of WM areas in the visuospatial 2-back condition and similar DMN deactivation in ASD and TD groups. This persisted in the 0-back condition in the ASD group only and was associated with diminished performance. In contrast to these results, Vogan et al. [52, 61] found a similar DMN deactivation in a visual 1-back task with different levels of difficulty in children and adolescents with ASD and TD, suggesting no reduced load-dependent modulation of DMN deactivation. However, in a 2-year follow-up study with the same participants, Vogan et al. [53] reported a greater longitudinal load-dependent DMN deactivation in those with ASD compared with TD. In middle-aged adults with and without ASD, Braden et al. [64] identified with a verbal 0-, 1-, and 2-back task, similar load-dependent DMN deactivation and similar performance between groups. Hence, there are no clear results about DMN deactivation during WM tasks in ASD.

Third, more recent studies have explored memory networks in ASD focusing on slow frequency bands in EEG/MEG. Urbain et al. [58] identified reduced fronto-temporal alpha connectivity associated with diminished performance during a visual 2-back condition in children with ASD relative to TD. This points to a less efficient WM network during maintenance. Using the same paradigm in young and middle-aged adults, Yuk et al. [65] found similar performance across ASD and groups TD, but connectivity results differed for each phase. Maintenance was associated with an alpha connectivity network common to both groups, with an additional recruitment of fronto-parietal areas and increased coherence, and recognition was associated with diminished theta coherence between right frontal and left parietal areas, when comparing groups with ASD and TD. In adults too, Larrain-Valenzuela et al. [66] manipulated memory load during a modified Sternberg task and found that for TD participants, there was enhanced alpha power in occipital areas and enhanced theta power in bilateral frontal areas when load increased. There was, however, no load-dependent modulation of alpha and theta power in the ASD group, indicating that atypical oscillatory activity in ASD may contribute to diminished WM performance. More recently, Audrain et al. [67] identified in a visual 1-back task in adults with ASD and TD with similar performance, that connectivity in the theta band during maintenance phase consisted mainly in interhemispheric posterior connections in ASD, while being mainly antero-posterior in TD. In addition, connectivity in the alpha band differed in ASD and TD in all WM phases. These results point toward a more atypically functioning WM maintenance relative to encoding and retrieval stages in ASD (theta band), but a global lack of inhibition of task-irrelevant processes (alpha band). Overall, in spite of contrasting results, a decrease in antero-posterior theta connectivity seems to characterize the WM maintenance phase in autistic individuals. Enhanced recruitment of the alpha band may correspond to a compensatory process, enabling greater inhibition to avoid interference.

Other studies on WM networks used fMRI during N-back tasks. In middle-aged adults, Braden et al. [64] identified a similar cortico-striatal–thalamic–cortical network of activation in ASD and TD groups but a reduced functional connectivity within this network in ASD alongside similar performance, when switching from 0- to 2-back condition. Using the method of connectivity matrices, Barendse et al. [68] identified a lower global efficiency (i.e., diminished transmission of neural information within the network in adolescents with ASD) along with performance in normal range. More recently, Hawco et al. [69] reported within-group differences in the location of prefrontal activations in young adults with ASD, associated with a trend for reduced performance when compared with TD participants, suggesting a link between idiosyncratic prefrontal activation during WM and diminished performance.

Only one study focused on visually presented verbal information. The studies reviewed used mainly visuospatial material including faces. We first describe the electrophysiological and then the fMRI results.

Neumann et al. [78] were interested in the old/new effects with MEG elicited by the recognition of pseudowords or abstract shapes. The memory performance of the participants with ASD was lower than TD for the former type of stimuli and similar for the latter. The authors identified an early right occipital activation (100–200 ms) in ASD with both kinds of stimuli that did not occur in controls, followed by a bilateral old/new effect (200–500 ms) with shapes in ASD participants reflecting familiarity-based recognition that was more left hemispheric in controls. These results suggest that pseudowords were processed and recognized similarly to visual material in ASD relative to TD participants.

Desaunay et al. [79] used a visual associative recognition task and identified a similar ERP time-course on the occipital P2 (220–270 ms), mid-central FN400 (350–470 ms), and parietal LPC (600–700 ms) potentials in adolescents and young ASD and TD adults, whereas performance for ASD participants was lower. Amplitudes were reduced in the group with ASD relative to TD on both the P2 potential, thought to index an intermediate processing stage linking elementary perceptual processes with higher-level semantic processes [93], and the FN400 potential, that reflects both conceptual priming and familiarity process [94]. Two ERP studies on face recognition reported similar performance between ASD and TD, but differences on both the temporal N170 potential (peaking 170 ms, and indexing attention and structural perception of faces), and the temporal N250 potential (also labeled as EPN, peaking 250 ms, and indexing familiarity-based recognition for faces [95]). In children, Gunji et al. [80] found a shorter latency on the N170 potential for self, familiar and unfamiliar faces in ASD relative to TD. Churches et al. [81] investigated the recognition of unfamiliar faces after a study phase and found reduced amplitudes of the N170 and N250 potentials in participants with ASD relative to TD. In summary, these amplitude decrements suggest reduced attention to visual stimuli and reduced integration of low-level perceptual into high-level conceptual information and the semantic memory system, with a decrease in familiarity strength during visual recognition in ASD.

Other studies used fMRI. Greimel et al. [82] studied in children and adolescents, the encoding of visually presented objects superimposed with either faces or houses. Despite similar performance at test, encoding of subsequently recognized objects presented with a face was associated, in ASD relative to TD, with reduced activations within bilateral inferior and medial frontal gyri and the right intraparietal lobule, suggesting that encoding of faces in the ASD group was less automatic and less associated with social information than in TD. Recently, Lynn et al. [83] evaluated the functional connectivity of the FFA for faces versus non-faces (cars) stimuli, at encoding and recognition stages in children, adolescents, and adults. TD participants tended to improve performance from adolescence to adulthood, while those with ASD did not. With faces, the authors found reduced connectivity between bilateral FFA with both prefrontal and primary visual cortices in ASD relative to TD. This was independent of age and occurred at both stages, suggesting that reduced integration of information between distributed brain areas may lead to weaker representation of faces in memory. These authors also identified in ASD, age-related reduced connectivity between the right FFA and the visual cortex only during the recognition phase, suggesting that memory difficulties with faces may arise more during recognition than encoding. Extending these results in an updated study, the same team investigated the age-related changes in the similarity of FFA activation [84]. For TD participants, they found an increasing overlap of activations within a category (faces, cars) from adolescence to adulthood in the right FFA—that processes faces holistically while the left FFA processes them more featurally [96]—reflecting brain maturation for holistic processing. In ASD. Similarity of FFA activation increased from childhood to adolescence, presumably reflecting maturation, but then decreased from adolescence to adulthood. Recognition performance was more related to the similarity score in the left FFA—while in the right FFA in TD—which together suggests reduced FFA specialization for faces and object identification in ASD, and a more featural rather than configural processing of these stimuli. Thus, fMRI studies complement those in EEG, by showing diminished FFA specialization for faces and objects and reduced integration of information between distributed brain areas including bilateral FFA and the visual cortex.

Studies focusing on encoding and retrieval processes in LTM explored either the functional connectivity during single item recognition (n = 2), the electrophysiological recollective old/new effect (n = 3), or associative processes with fMRI (n = 4).

First, Noonan et al. [85] conducted an fMRI source recognition paradigm in adolescents and adults using visually presented target words previously presented in a visual or auditory modality. Functional connectivity was analyzed between three seed regions of interest (left middle frontal, left superior parietal, and left middle occipital cortex) and the whole brain. The authors found, in spite of diminished performance in ASD, a large pattern of similar connectivity in ASD and TD groups and unexpectedly enhanced left fronto-parietal functional connectivity in participants with ASD relative to TD. Close to these results, Chan et al. [86] investigated in children and adults, the coherence in the theta band—associated with WM and LTM memory processes—during the recognition phase of previously seen pictures. Antero-posterior theta connectivity was greater in the left hemisphere in ASD participants and in the right hemisphere for TD ones, with a negative correlation between coherence and performance in the ASD group only, suggesting a threshold of connectivity beyond which performance decreased. Taken together, these two studies identified enhanced hemispheric left connectivity during episodic recognition in the ASD compared with the TD participants.

Using ERPs, Massand et al. [87] and Massand and Bowler [88] conducted two successive studies focusing on the familiarity-related FN400 and recollection-related LPC potentials; overall memory performance was similar in ASD and TD groups in both studies. Massand et al. [87] employed a single-word recognition paradigm and found a parietal rather than anterior early FN400 old/new effect (300–500 ms) in adults with ASD relative to TD, followed by a similar parietal LPC recollective process (500–800 ms) in both groups. Massand and Bowler [88] used a single-picture recognition test followed by a phase where the color of the studied items had to be recalled. The authors found a posterior and attenuated FN400 old/new effect in the ASD group, occurring in a large time-window (300–650 ms), followed by a similar LPC old/new effect. More recently, Desaunay et al. [79] used a visual associative recognition task in a study of adolescents with ASD. They found in conjunction with diminished performance in the ASD relative to the TD participants, a reduced FN400 amplitude followed by a LPC old/new effect presenting a parietal extension. These findings suggest that effortful retrieval of associative information compensates for the lower familiarity strength in the ASD participants. Overall, results from these three ERP studies indicate that recognition in ASD is qualitatively similar to that seen in TD and relies on the same dual-process, but may, however, differ quantitatively, with diminished familiarity-related potentials and preserved or even enhanced recollection-related potential in ASD.

Other fMRI studies explored associative memory in ASD, with a focus on hippocampus as a key structure underlying inter-item and item-context associations [97]. Gaigg et al. [89] evidenced a strong overlap of brain activations in adults with ASD and TD—including the left inferior frontal gyrus and left hippocampus—with similar signal change with the degree of relational encoding of word triplets accompanied by diminished subsequent recognition. Between-group difference was limited to absent or reduced signal change in prefrontal and posterior hippocampal regions in the ASD group relative to TD, when contrasting the degree of awareness (familiarity/recollection) for subsequent retrieval. Three recent studies have investigated associative memory with visual material. First, using a paradigm designed to assess memory precision, Cooper et al. [90] identified similar levels of activity and functional connectivity among prefrontal, hippocampal and parietal areas during encoding, in adults with ASD and TD. By contrast, the recognition phase included several differences albeit close performance, including reduced left prefrontal activity, suggesting limited pre-retrieval search or post-retrieval monitoring, and reduced hippocampal connectivity with the fronto-parietal network. Second, Hogeveen et al. [91] investigated the associative encoding of paired pictures in adolescents and young adults and identified increased hippocampal activity along with diminished connectivity between the medial temporal lobe and postero-medial brain regions in the ASD relative to the TD group, alongside similar between-group performance. This enhanced activity and reduced connectivity were significantly inversely correlated, indicating that the decreased inter-regional connectivity in ASD would be efficiently compensated for by hippocampal hyper-recruitment to support preserved associative memory. Fronto-hippocampal connectivity was also reduced in the ASD group compared with TD group. Third, Cook et al. [92] performed an associative encoding task, asking participants from late childhood to adolescence to rate if object–scene pairs were congruent, incongruent, or intermediate. Participants with ASD and TD showed both a similar congruency rating during encoding, and the same facilitation for better subsequently recognizing congruent than incongruent pairs. In the TD group, encoding of subsequently recognized pairs was associated with greater activation of the bilateral medial prefrontal cortex for intermediate relative to congruent pairs, and with greater activation of the left medial temporal lobe (including the hippocampus) for congruent relative to intermediate pairs. By contrast, activations of these two regions did not vary according to the level of congruency in the ASD group. As congruent information encoding is progressively supported by the medial prefrontal cortex throughout development, the lack of modulation observed in the ASD group may reflect developmental immaturity; participant with ASD with greater behavioral flexibility had higher activations in bilateral, prevailing left medial prefrontal areas. Taken together, these fMRI results extend those in EEG, showing that processing of associative information is qualitatively similar in ASD as in TD. Notably, hippocampal activation is similar or enhanced in ASD compared with TD; between-group differences mainly consist in reduced prefrontal activation and hippocampal connectivity.