Alterations in development of hippocampal and cortical memory mechanisms following very preterm birth

Deficits in memory function have been described in children and adolescents who were born very preterm (VPT), which can have profound effects on their school achievement and everyday life. However, to date, little is known about the development of the neuroanatomical substrates of memory following VPT birth. Here we focus on episodic and working memory and highlight key recent functional and structural magnetic resonance imaging (MRI) studies that have advanced our understanding of the relationship between alterations seen in the VPT brain and typical neurodevelopment of networks supporting these memory functions. We contrast evidence from the episodic and working memory literatures and suggest that knowledge gained from these functional and neuroanatomical studies may point to specific time windows in which working memory interventions may be most effective.

Individuals who were born very preterm (VPT; <32wks' gestation) are more likely to exhibit memory deficits compared with term-born controls, which could impact their school performance and everyday life. 1 In this paper we will focus on two different types of memory, episodic and working memory. Episodic memory could be described as the ability to encode, store, and retrieve information about an event, including information about the context in which the event took place. In contrast, working memory refers to short-term storage and 'online' manipulation of information. Episodic and working memory are important components of executive control functions and are closely associated with each other. Working memory capacity has in fact been found to predict successful episodic remembering. 2 These operations may interact via shared neuroanatomical mechanisms related to common control processes, as overlapping activation patternspredominantly in prefrontal cortexhave been described for both episodic and working memory. 3 However, distinct neural networks and activation patterns have also been proposed for episodic and working memory processing, and in this paper we will describe how these are organized in the typically developing brain. At a behavioural level, the capacity to form episodic memories is believed to emerge at around 4 to 5 years of age, 4 and episodic memory performance improves during childhood until adolescence, with older children being more successful at integrating information with its contextual details. 5 Such improvements may parallel the development of the brain's connectional architecture, which we describe in detail later in this review.
Working memory abilities also emerge in early childhood and develop rapidly between the age of 4 years and 10 years. Much of the improvement in working memory capacity appears be closely associated with increases in processing speed, and with the development of subvocal rehearsal strategies. 6 After summarizing the organization of episodic and working memory in the typically developing brain, we will examine the structural alterations and injuries that are likely to impact specific memory components in VPT samples. Thirdly, we will describe the mechanisms by which the brains of VPT individuals may attempt to overcome such structural alterations. Lastly, we will contrast the findings from episodic and working memory, and discuss the possible implications these may have for cognitive interventions that aim to ameliorate the cognitive profile of VPT children.
Research on the structural and functional memory correlates in VPT samples is presented systematically. Studies were identified using PubMed with the search terms ('preterm OR VPT' AND 'MRI OR magnetic resonance imaging' OR 'fMRI OR functional magnetic resonance imaging' AND 'memory OR learning'). There were no restrictions on the year the studies took place. Studies in any language were considered. This search was supplemented by hand searching of reference lists of published articles. Studies were included if a group of individuals born preterm (including very or extremely preterm) were compared with a group of controls using quantitative structural MRI, and quantitative structural measures were correlated with memory outcome measures; or, for fMRI studies, if a group of individuals born preterm (including very or extremely preterm) were required to perform an fMRI memory task and activation was compared with a control group. The PubMed search produced 90 articles; 26 of these met the inclusion criteria. One additional article that was found through manually searching the references met the inclusion criteria. The outcomes of these searches were then divided into 'structural' (n=16) and (primarily) 'functional' (n=11) studies, and are shown in Tables I and II respectively.

EPISODIC MEMORY Development of the episodic memory system
The hippocampus is a critical structure in the episodic memory system. It lies at the end of convergent processing streams of visual, auditory (from temporal visual and auditory areas, via entorhinal and perirhinal cortex), and spatial information (from parietal areas, including posterior cingulate, via parahippocampal cortex). The hippocampus is thus in an ideal position to bind object information (from temporal cortex) with spatial information (from parietal cortex), to form memories of complete episodes. 7 The hippocampus develops rapidly during the first 2 years of life, and then more gradually until it reaches its peak volume between the ages of 9 years and 11 years. 8 However, gains in episodic memory function continue after this point and are associated with increased specialization of hippocampal subdivisions, and with the structural development of hippocampal-cortical networks. 9 This reflects the fact that while the hippocampus is the critical hub, it is just one part of a greater episodic memory system, which includes the mammillary bodies, anterior thalamic nuclei, posterior cingulate, and parahippocampal cortex. This functional loop is complemented by more direct hippocampal-cortical connections to medial temporal, posterior cingulate, and prefrontal areas, which are likely to be involved in the recall of information from the hippocampus (Fig. 1). 7 The increased reliance on cortical support for episodic memory occurs in parallel with the maturation of the brain's connectional architecture. The fornixthe main hippocampal connection to other subcortical structuresreaches maturity early, at about 5 years of age. 10 The cingulum and uncinate fasciculus, which connect the hippocampus to the posterior cingulate and prefrontal cortex, undergo a more protracted development, 10 which is mirrored by increased fronto-hippocampal functional connectivity and better performance on episodic memory tasks involving higher-order functions. 9 Structural alterations to the episodic memory system following VPT birth The periventricular location of the hippocampus makes it particularly vulnerable to damage following VPT birth. Reductions in volume in the hippocampus, as well as the interconnected thalamus and posterior cingulate cortex, are already present at term-equivalent age and are related to the degree of prematurity. 11 Smaller bilateral hippocampal volume in VPT samples compared with controls has been described at school-age (~9% volume reduction), 12,13 adolescence (~14% volume reduction), 14,15 and adulthood 16,17 although some studies have found no hippocampal volume differences when correcting for factors such as intracranial volume. 18,19 Interestingly, hippocampal volume at term, but not at 7 years of age (or the amount of growth between these time-points), has been associated with episodic memory abilities at age seven. 12,18,20 This suggests that children with early hippocampal damage may have a limited capacity for the development of episodic memory functions. Structural and functional alterations in the fornix, corpus callosum, and the parahippocampal, entorhinal, and perirhinal cortices following VPT birth persist until adolescence 21,22 and early adulthood [23][24][25][26] and are correlated with memory ability. 25 Table I summarizes studies investigating the structural brain correlates of memory functions in VPT samples.

Intrinsic coping strategies and functional adaptation
As the brain structures centrally implicated in episodic memory do not appear to catch-up developmentally, even by early adulthood, individuals born VPT may find alternative means of successfully completing tasks requiring episodic memory processing. Various studies have focused on how VPT individuals can bind information to form memories and later recall that information despite structural damage to the core circuit. In an early combined structural/functional MRI study Gim enez et al. showed evidence that lateralized structural deficits to the hippocampus in a heterogeneous sample of VPT adolescents can be partially compensated by an increase in functional activation of the contralateral hippocampus. 27 A similar study by Lawrence et al. found that while hippocampal volumetric reductions were present in the VPT group compared with controls, there was an increase in the volume of the parahippocampal cortex, with this increase being correlated with an increase in activation in this area. 28 Kalpakidou et al. showed that increased levels of neonatal brain injury following VPT birth were associated with reduced access to key nodes of the episodic memory network, namely the posterior cingulate and lateral prefrontal cortex. 29 In two recent studies, we analysed the dynamics of how VPT individuals and controls learn memory associations and showed that 'pure' memory deficits can be understood in the context of neuroanatomical alterations occurring during learning. Salvan et al. used a task in which partici-

What this paper adds
This article highlights how typical brain injuries associated with very preterm birth can interact with the typical development of episodic and working memory networks.
By drawing on the lessons of 15 years of neuroimaging studies, suggested targets for cognitive interventions are proposed. pants were required to learn associations between two words across repeated experimental blocks, and were later asked to remember them during a cued recall task. 24 By looking for patterns of adaptive activation we were able to detect learning signals in the brain, and crucially, see how these patterns differed between VPT and control participants. VPT individuals made increasing use of domaingeneral cognitive regions (caudate and anterior cingulate cortex) during successive encoding trials, which was in sharp contrast to the typical habituation signal seen in control participants. While controls increased activation in the hippocampal-parahippocampal-thalamic section of the episodic memory circuit during recall trials, VPT participants reduced activation in these areas, possibly reflecting a search for alternative strategies due to suboptimal engagement of the core episodic memory network. This hypothesis was supported by the finding that such functional adaptations were linked to structural alterations in pathways linking these structures, including the fornix. 24 A related study of the dynamic formation of visual memory associations by Brittain et al. 30 revealed reduced recruitment of the hippocampus, parahippocampal, and posterior cingulate cortices in VPT adults during learning of visual paired associates. This study also shed light on how such memory tasks can affect other networks, with similar reductions in activation seen in dopaminergic regions, such as the substantia nigra, which is thought to enhance hippocampus-dependent memory formations through rewardlearning related mechanisms. 31 Figure 1 shows a visual summary of functional alterations to the typical episodic memory network following VPT birth.

WORKING MEMORY Development of the working memory system
Current working memory capacity depends primarily on a fronto-parietal system that involves dorsomedial, dorsolateral, and ventrolateral prefrontal cortex, as well as medial and lateral portions of the posterior parietal cortex. 32 These regions are interconnected laterally by the threepronged superior longitudinal fasciculi and medially by the dorsal portion of the cingulum bundle (Fig. 2). Maturation of the working memory network is associated with functional activation increases in the frontal and parietal cortical areas 33 and with cortical thickness decreases in prefrontal 34 and posterior parietal cortex. 34,35 Concurrent micro-and macro-structural development of the frontoparietal tracts is also seen and is associated with age-related improvements in working memory capabilities. 35 Furthermore, microstructure of the fronto-striatal tracts and activation of the caudate nucleus during fMRI tasks have been shown to be predictive of future working memory capacity. 35 The involvement of the hippocampus in working memory is strongly disputed, although it appears that dependence on this structure is seen mainly in childhood, 36 possibly reflecting the comparatively early hip- Verbal paired associates [20][21][22][23][24] No between group performance differences. Encoding: PVH+VD and UPVH activated the right middle frontal gyrus, which was mildly deactivated in the VPTN group and strongly deactivated in the controls.
Recall: PVH+VD strongly activated the posterior cingulate, which was also activated, but to a smaller extent in the UPVH group, showed little response in the VPTN group and was deactivated in the controls.
A linear trend analysis of the data revealed a linear trend in the grey matter volume PVH+VD<UPVH<VPTN<Control in the right superior temporal gyrus, right cerebellum, left middle temporal gyrus, right globus pallidus, and right medial frontal cortex. Structure-function relationships not significant No between group performance differences. Encoding: Controls showed a progressive decrease in activation in the anterior cingulate/caudate nucleus as the task repeated. VPT showed a progressive increase in activation in the same areas.
Recall: Controls showed a progressive increase in activation in the hippocampus/parahippocampal cortex/thalamus across as they were asked to repeatedly recall learned paired associates. VPT showed a progressive decrease in activation in the same area.
VPT exhibited decreased fractional anisotropy in white matter tracts connecting these memory-related areas, including the fornix, the inferior longitudinal fasiculus, inferior fronto-occipital fasciculus and the splenium of the corpus callosum Murner-Lavanchy et al. 46 Whole brain and ROI 41 (<32wks gestational age)

36
Dot location visuospatial shortterm/working memory task+out of scanner shape location task.

7-12
No between group performance differences. Whole brain: reduced activation in VPT in posterior middle frontal gyrus/precentral gyrus ROI analysis: no activation differences in the middle frontal gyri, and increased activation in the superior frontal gyri in the VPT group Brittain et al. 30 Whole brain (learning linear trend analysis) 24 (<33wks gestation) 22 Visual paired associates 20 No between group performance differences. Encoding: Controls increased activation in the parahippocampal cortex, substantia nigra, left inferior frontal gyrus and right anterior cingulate, while VPT individuals decreased activation in these areas. Controls decreased activation in the cerebellum, right middle temporal gyrus and medial frontal cortex, while VPT individuals increased activation in these areas. Recognition: Controls progressively increased activation in the insula/claustrum/putamen, while VPT individuals reduced activation in this area. In contrast, VPT increased activation in the left cerebellum, which showed a progressive reduction in activation in the control group Daamen et al. 47 Whole brain 73 (<32wks gestation or <1500g birthweight)

73
One-and two-back (from the n-back working memory task).

26.5
No between group performance differences. Reaction times higher in the VPT group, but there was no task level x group interaction. VPT showed stronger load-dependent deactivation of the right precuneus than controls. A similar pattern was found in the right cerebellum/parahippocampal gyrus in post-hoc tests.
Deactivation in the cerebellum/parahippocampal gyrus was associated with reaction time in the VPT group, but not controls (although the correlations were not statistically compared) DMS, delayed matching to sample; DNMS, delayed non-match to sample; PVH+VD, periventricular haemorrhage with ventricular dilatation; RAVLT, Rey Auditory Verbal Learning Test; ROI, region of interest; UPVH, uncomplicated periventricular haemorrhage; VBM, voxel-based morphometry; VPTN, very preterm, no periventricular haemorrhage.

Review 41
pocampal development compared with the frontal and parietal areas that dominate the adult working memory system.
Structural alterations to the working memory system following VPT birth Working memory deficits have been described in childhood following VPT birth (see Anderson 37 for a review). Prematurity-related hippocampal damage at term-equivalent age has been associated with working memory deficits at age 2 years. 38 It remains to be seen whether these deficits are temporary and reflect the reliance of working memory function on the hippocampus that is seen early in development, or if early hippocampal damage disrupts subsequent neurodevelopment in brain areas sub-serving working memory processing that can then result in permanent functional deficits. Interestingly, one recent study found a relationship between the volume of the right hippocampus and working memory ability in young adulthood, 16 although to date this appears to be the only study to find such a specific link in adulthood. Furthermore, animal data suggest that early anatomical damage to the hippocampus may be sufficient to disrupt typical maturation of the prefrontal cortex and cause a permanent reduction in working memory abilities. 39 In line with the possible role of striatal structures in the acquisition of working memory abilities in typical development described above, neonatal striatal and thalamic injury is predictive of working memory abilities at age 7 years in VPT children. 40 In late adolescence, however, evidence suggests that working memory deficits following preterm birth are linked to the structural abnormalities of the mature working memory system observed in normative samples, such as cortical thickness alterations of the medial and superior parietal cortex 41 and widespread cortical surface area alterations. 42 Our recent work suggests that delays in fronto-parietal cortical thickness maturation substantially diminish by early adulthood, 43 although the functional correlates of such structural 'catch-up' remain to be investigated in detail.
A summary of structural MRI studies in relation to memory functions in VPT samples is provided in Table I.

Intrinsic coping strategies and functional adaptation
Despite considerable neuropsychological evidence of working memory deficits in children born VPT, there have been surprisingly few fMRI studies investigating their functional neuroanatomical underpinnings. Taylor et al. used an fMRI task requiring basic visual working memory processing, and showed evidence of decreased activation in preterm children versus controls in regions belonging to both early and mature working memory systems, namely the right medial temporal and left medial parietal cortices respectively. 44 These results suggest this phase of maturation (7-9y) may represent a key transitional stage of working memory development. Support for this argument comes from a study of 11-year-old children born extremely preterm by Griffiths et al., which shows activation reductions compared with controls in frontal working memory areas, which belong to mature (as opposed to early) working memory networks. 45 Preliminary evidence of a catchup of activation patterns comes from a study by Murner-Lavanchy et al. 46 who showed that VPT children (7-12y) may have increased activation in superior frontal cortex to compensate for persistent functional deficits in the adjacent middle frontal gyrus. 46 The most recent, and to date the largest, fMRI working memory study of VPT individuals found no working memory deficits in adulthood. 47 Daamen et al., however, observed altered deactivations of the medial parietal cortex and a cerebellar node in their VPT group in the absence of 'positive' activation differences, which they hypothesized could be representative of compensatory adaptation. It is worth noting that the task used in the study by Daamen et al. was a relatively easy variant of the n-back working memory task, and activation (or deactivation) patterns did not correlate with task performance. It remains to be seen what adaptations to the working memory system could lead to a catch-up of function, and whether neonatal and environmental factors may modify activation and developmental patterns within VPT samples. We recently suggested that individuals born VPT with neonatal periventricular injuries may exhibit reduced prefrontal activation during working memory performance in adulthood compared with those without neonatal injury, despite having comparable activation on other tasks of executive function. 48 Figure 2 shows a visual summary of functional alterations to the typical working memory network following VPT birth.
A summary of functional MRI studies in relation to memory functions in VPT samples is provided in Table II.

CONCLUSION
To date, studies of memory deficits in VPT samples have focused on episodic and working memory. The available evidence suggests that structural deficits to the hippocampus and connected structures following VPT birth may be long-lasting, affecting episodic memory abilities from childhood to early adulthood. Functional and structural alterations to hippocampal-cortical networks are also associated with working memory deficits in VPT children. The lack of concluding evidence for an association between hippocampal-related deficits and working memory function in adolescents or adults born VPT (with the exception of Aanes et al.'s study 16 ) gives hope that the apparent transfer of working memory function from subcortical to fronto-parietal cortex may result in neurodevelopmental adaptation. Functional neural compensation hypotheses are yet to be investigated in relation to memory processing, but promising results are provided by studies of other cognitive functions, such as language, which have suggested the existence of alternative task-specific neural pathways in the developing preterm brain. 49 Multimodal studies covering a greater range of ages and ultimately long-term longitudinal studies need to be undertaken to provide a more comprehensive overview of the dynamics of development of memory function following VPT birth.
Although more research is needed, current evidence suggests that working memory may be a more promising target than episodic memory for cognitive training in VPT individuals. Moreover, interventions may be more effective if applied at an age when frontal and parietal cortices are sufficiently developed to be able to undertake such tasks, yet still maturing, so that focused interventions could exploit their residual neuroplastic capacity. Lastly, future research should include the identification of areas outside the typical memory-specific networks that may be engaged in VPT samples as a result of neural adaptation, to successfully perform specific cognitive tasks. working memory using neurosynth.org). 50 White matter tracts that connect the fronto-parietal structures of the working memory system shown beneath the cortical surface (SLF [superior longitudinal fasciculus]; SLF1, red; SLF2, yellow; SLF3, green; dorsal cingulum bundle, purple). The hippocampus is shown in yellow due to its apparent importance to childhood working memory deficits in VPT individuals. Working memory structures that most commonly show reduced activation during functional magnetic resonance imaging (fMRI) tasks of working memory in VPT individuals compared with controls are marked with a green circle. Structures that show increased activation in VPT individuals compared with controls are marked with a pink circle. The size of each circle represents the number of fMRI studies that have reported over-/under-activation of the structure relative to controls.