2.1. Working Memory Deficits
The prefrontal cortex is one of the most widely explored regions in functional imaging studies of working memory. Reduced activation patterns in the DLPFC in schizophrenia in working memory tasks appear to be most widely replicated; with the extent of reduction being correlated with poorer performance [
67,
68]. For example, Callicott
et al. report a failure of activation of the DLPFC in five of six patients with schizophrenia using a working memory task [
69]. These abnormalities were later confirmed in groups of chronically ill and first episode patients, in medicated and medication naive subjects [
70,
71,
72,
73]. The failure of activity increases during task in medicated patients subsequently was reported only in high, but not in low, levels of working memory load. Not surprisingly, an association between decreased task performance and failure of activation was observed [
74]. However, several studies are inconsistent with the finding of decreased prefrontal cortex activation during working memory tasks. No abnormalities in DLPFC activation patterns were found in acutely ill, medicated patients and hospitalized subjects in partial remission [
75,
76], and even hyperactivation was reported in mildly to moderately ill chronic patients with schizophrenia [
77,
78,
79].
It has been suggested that the inconsistencies in the literature could be related to task requirements or may be explained by poor task performance in schizophrenia [
28,
80]. Many argue for the conceptualization of task related activation patterns as a non-linear, inverted U-shaped function that relates the fMRI signal to working memory load and is shifted to the left in patients with schizophrenia. Assuming an initial relative overactivation of this region with relatively low memory demands, a decline in processing capacity with increasing demand would then be accompanied by a relative underactivation [
81,
82,
83]. A disruption in this activation-performance relationship in schizophrenia has indeed been reported [
28,
84]. Furthermore, a left shift of the curve was confirmed in a study of ten chronically medicated patients reaching peak activation on the working memory system at lower processing loads than healthy controls and a subsequent decline of DLPFC activity at high processing loads [
85].
Honey
et al. attempted to control for possible task performance confounds by recruiting 20 male patients with intact performance on a low-load verbal memory task and matched controls. In controls, functional response and behavioral performance correlated, but a de-coupling between these parameters was present in schizophrenia [
86]. Accounting for performance differences on measures of cognition between patients with schizophrenia and healthy controls is a challenging problem. Controlling for these differences has been handled in different ways, including limiting analyses to correct trials or blocks of trials with acceptable accuracy, using task performance as a covariate, matching groups based on performance or adjusting disparity between task presentations to individual ability levels [
75,
83,
87,
88,
89,
90,
91].
Interestingly, it appears that even patients who are able to keep up with processing demands tend to engage greater levels of activation or a less focused cortical activity state, thus responding to demands less efficiently [
92,
93]. Ragland
et al. report that, while healthy participants respond to increased processing demands through an increase in DLPFC activity, patients with schizophrenia did not show higher DLPFC activity, but diffusely engaged a number of cortical and subcortical regions to meet the same demands [
94]. In addition to reduced activation of the right DLPFC, an abnormal correlation between the left DLPFC and the left hippocampal formation that was not present in healthy controls was seen in a group of unmedicated patients, again underscoring widespread neural abnormalities [
95]. In a quantitative meta-analysis of twelve studies using a working memory paradigm, Glahn
et al. report clear support for hypofrontality, but also consistently increased activation of the anterior cingulate cortex and frontal pole in patients [
67]. Consistent with this, aberrant frontotemporal functional connectivity patterns, with dorsal prefrontal/anterior cingulate hypoactivity were described in working memory tasks in patients compared to healthy controls [
96]. Interestingly, an increase in functional connectivity between the DLPFC and hippocampus was found in patients [
97]. Taken together, these findings could be suggestive of a disruption in frontal-based top-down cognitive control function and resulting compensatory response to support alternative strategies in task performance in patients with schizophrenia. In a meta-analysis of functional imaging studies of executive function in schizophrenia that also included a subset of studies using working memory tasks, Minzenberg
et al. report widespread abnormalities of activation patterns, including deficits in the middle frontal gyrus, anterior cingulate cortex and thalamus in patients with schizophrenia compared to healthy controls. To maintain performance, patients may increase other processes including mnemonic, attentional and performance monitoring functions that manifest as relative hyperactivation in aforementioned regions;
i.e., less focused cortical activity [
98].
Given the heritability of working memory deficits, several studies investigated the association between processing efficiency and catechol-O-methyltransferase (COMT), a known schizophrenia risk gene, during a working memory task. The authors did find the Val allele to be associated with a reduced level of performance and reduced efficiency of the physiological response in the dorsolateral prefrontal cortex, a finding that presumably is mediated by reducing signal to noise through compromising postsynaptic impact of evoked dopamine response [
93,
99,
100,
101,
102,
103]. When dopamine D1 receptor availability in the DLPFC was assessed in unmedicated patients, a relationship between increased receptor availability and worse working memory performance was identified, further substantiating the role of dopamine abnormalities in working memory deficits [
104].
Several studies investigated effects of antipsychotic medications, which act by blocking dopamine receptors, on brain activation patterns in patients with schizophrenia. In a longitudinal study, Wolf
et al. do report enhanced bilateral frontotemporal function after 7–8 weeks of multimodal antipsychotic treatment that was associated with improved accuracy in a verbal working memory task and improvement of psychotic symptoms [
105]. Schlagenauf
et al. report activation deficits during a working memory task in the DLPFC, but did not find significant changes in activation after switching patients from a typical antipsychotic to olanzapine. Surprisingly, they do report a normalization of frontal lobe activity when switching to aripiprazole, a partial dopamine agonist [
106,
107]. In another group of first-episode, medication naive patients, the baseline activation deficits in the DLPFC were unchanged after ten weeks of treatment with different second generation antipsychotics, an effect that was driven by medication non-responders. Patients who did respond to medications had DLPFC activation levels that were much more similar to that of healthy controls [
108,
109]. Other studies report improvement of baseline activation deficit in the DLPFC after a twelve week course of quetiapine and after 6–8 weeks of risperidone or olanzapine [
81,
110]. Consistent with this, Honey
et al. did report improvement of functional activation by a working memory task in the right DLPFC within six weeks of switching from typical antipsychotics to risperidone and hypothesized that this finding may be due to reduced dopamine D2 receptor antagonism in the nigrostriatal system, which might enhance frontal cortical activity by reducing inhibitory outputs [
111]. Contrastingly, regional cerebral blood flow abnormalities in the anterior cingulate cortex were corrected in patients treated with clozapine, but not haloperidol, suggesting differential mechanisms in mediating neural activation between these drugs [
112].
2.2. Episodic Memory Deficits
A substantially smaller number of functional neuroimaging studies have been conducted with the goal to elucidate neural correlates of episodic memory deficits in schizophrenia. In the prefrontal cortex, activation deficits have most commonly been reported [
73,
113,
114,
115,
116], but also increased activation, especially in the frontal pole, has been reported [
113,
116]. In addition to prefrontal abnormalities, some studies have also reported reduced activation of the MTL [
73,
115], with others reporting no change [
87], or increased activation in this area [
114,
116].
Given that abnormalities appear to involve both the prefrontal cortex and MTL, a disruption in the frontotemporal network has been proposed [
22]. In 14 patients with schizophrenia, Ragland
et al. found evidence of this, reporting prefrontal activation deficits and parahippocampal over activation during encoding [
114]. Evidence of reduced left inferior frontal activation during encoding and reduced anterior cingulate cortex and temporal lobe activation, but no reduction of hippocampal activation was later reported in a group of stable, medicated patients [
117]. Similar to this, Hofer
et al. report an activation deficit in the DLPFC and anterior cingulate cortex in encoding and impairments in DLPFC activation during retrieval. When encoding and recognition were contrasted, attenuated frontotemporal activation in patients with schizophrenia was apparent, despite intact behavioral performance [
118]. Later, they also reported activation failure in the frontal, posterior cingulate and retrosplenial regions during encoding and reduced activation in the DLPFC and paralimbic regions during retrieval in acutely psychotic patients [
119]. When patients were given organizational strategies through levels-of processing paradigms, patients had similar behavioral effects compared to healthy controls in a word encode and retrieve task that was accompanied by increased ventrolateral prefrontal activation during encoding in both groups. Overactivation in the thalamus, hippocampus and lingual cortex were present during encoding and overactivation of the left frontal pole with decreased activation of the right prefrontal cortex during retrieval were observed in patients with schizophrenia [
116]. Hippocampal activity during deep encoding along with reduced anterior cingulate cortex and dorsomedial prefrontal cortex activation led to the hypothesis that an activation deficit in the anterior cingulate could result in insufficient top-down modulation of attention, resulting in impaired encoding performance [
120]. Given the association between increased hippocampal activity, positive symptoms, and impaired memory performance, it is conceivable that increased neural activity is necessary to compensate for dysfunctional mesolimbic circuitry that results in a hyperdopaminergic state [
121,
122,
123]. Both experimental and computational evidence suggests that excitation of the hippocampus can lead to excitation of dopaminergic neurons in the ventral tegmental area and, in turn, leads to increased release of dopamine in the hippocampus, possibly resulting in positive symptoms and memory impairment [
121,
124,
125].
Meta-analytic evidence of episodic memory studies initially suggested reduced right hippocampal activation along with activation deficits of the left inferior frontal cortex and medial prefrontal cortex during encoding in patients with schizophrenia and a deactivation of the hippocampus, but hyperactivation of the parahippocampal gyrus with activation deficits of the anterior cingulate cortex, the inferior frontal cortex, middle frontal cortex and medial prefrontal cortex during retrieval [
126]. However, a later meta-analysis investigating neuroimaging correlates of episodic memory deficits in schizophrenia did report less prefrontal activation in the frontal pole, DLPFC and ventrolateral prefrontal cortices during encoding and less DLPFC and ventrolateral prefrontal cortex during retrieval, but did not confirm reduced hippocampal or surrounding medial temporal lobe activation in patients during encoding or retrieval. They did find a relative increase in activation in the parahippocampal gyrus during encoding and retrieval, which may reflect a compensatory mechanism, with overall greater distribution in activation abnormalities in retrieval than encoding [
127]. Discrepancies in findings of these meta-analyses may be associated with the difference in study inclusion criteria; with the latter excluding region-of-interest studies that the former had included.
Genetic contributions to functional abnormalities with episodic memory tasks have been confirmed in a study of patients with schizophrenia and healthy siblings who demonstrated similar patterns of reduced hippocampal-parahippocampal activation with a task, suggesting a susceptibility-related phenotype [
128]. In a similar approach, using a visual task, incremental increase in BOLD responses in schizophrenia compared with first-degree relatives and healthy controls in prefrontal regions, thalamus and insula were reported during retrieval, but no activation differences were noted during encoding, again highlighting a possible genetic component [
129]. When directly testing effects of neuregulin, neurogranin and dystrobrevin-binding protein 1, all genes previously identified as increasing risk for schizophrenia have been demonstrated to modulate brain activation during episodic memory processing in healthy controls [
130,
131,
132]. However, these links remain to be confirmed in patients with schizophrenia. An association that has been suggested to be relevant is the schizophrenia risk gene, COMT, that also been implicated in BOLD signal activation during working memory tasks. A genotype by diagnosis interaction has been demonstrated in investigating parahippocampal activation during encoding in patients with schizophrenia and healthy controls [
133]. This finding again implies an alteration in dopamine signaling that could be related to observed abnormalities in brain activation patterns.
Given this and the above discussed, evidence, albeit, not undisputed, of antipsychotic medications possibly affecting brain activation patterns in working memory, it is conceivable that these medications may also have modulating effects on activation during episodic memory tasks. While fMRI studies have enrolled subjects who were medicated [
114,
115,
116,
117,
118] and unmedicated [
119], there is a paucity of longitudinal investigations with the goal to elucidate changes of activation patterns induced by antipsychotic medications.
2.3. Shared Abnormalities in Working Memory and Episodic Memory
Much of the work investigating memory deficits in schizophrenia has conceptualized working memory and episodic memory as two independent constructs and studied them individually. Given the shared neural substrates of both these memory systems in healthy controls and the evidence that prefrontal cortical abnormalities contribute to both working memory and episodic memory deficits, Barch
et al. chose to test the hypothesis that abnormal prefrontal cortex activation contribute to deficits in both memory systems. Indeed, the authors found impaired activation in the right DLPFC in patients with schizophrenia in both tasks. Interestingly, all functional abnormalities observed in the working memory task were also present in the episodic memory task [
73]. Congruent with this, working memory deficits in patients with schizophrenia were found to account for deficits in long-term recall in a behavioral study [
134], but others report that these are differential core deficits in schizophrenia [
135]. In a study investigating the relationship between neural processing related to working memory and long-term memory, Ragland
et al. found that the DLPFC was activated during a working memory task, but that activation was less focal in patients compared to controls and that this activation did not translate to higher success in long-term memory processes. The authors further report that patients showed disproportional impairment in recognizing familiarity for items compared to general long-term memory deficits, emphasizing the processing of relational information in working memory [
94]. It is conceivable that abnormal prefrontal cortex and MTL activation patterns and aberrant functional connectivity between these regions could be related to DLPFC dysfunction with subsequent compensatory mechanisms that result in difficulties both working memory and long-term memory deficits [
136,
137]. However, more investigations need to be conducted to clarify if both memory systems share a common neural substrate.