Strong Evidence for Pattern Separation in Human Dentate Gyrus

Abstract

The hippocampus is proposed to be critical in distinguishing between similar experiences by performing pattern separation computations that create orthogonalized representations for related episodes. Previous neuroimaging studies have provided indirect evidence that the dentate gyrus (DG) and CA3 hippocampal subregions support pattern separation by inferring the nature of underlying representations from the observation of novelty signals. Here, we use ultra-high-resolution fMRI at 7 T and multivariate pattern analysis to provide compelling evidence that the DG subregion specifically sustains representations of similar scenes that are less overlapping than in other hippocampal (e.g., CA3) and medial temporal lobe regions (e.g., entorhinal cortex). Further, we provide evidence that novelty signals within the DG are stimulus specific rather than generic in nature. Our study, in providing a mechanistic link between novelty signals and the underlying representations, constitutes the first demonstration that the human DG performs pattern separation.

Significance statement: A fundamental property of an episodic memory system is the ability to minimize interference between similar episodes. The dentate gyrus (DG) subregion of the hippocampus is widely viewed to realize this function through a computation referred to as pattern separation, which creates distinct nonoverlapping neural codes for individual events. Here, we leveraged 7 T fMRI to test the hypothesis that this region supports pattern separation. Our results demonstrate that the DG supports representations of similar scenes that are less overlapping than those in neighboring subregions. The current study therefore is the first to offer compelling evidence that the human DG supports pattern separation by obtaining critical empirical data at the representational level: the level where this computation is defined.

Keywords: 7 T fMRI; MVPA; hippocampal subfields; medial temporal lobe; memory; pattern separation.

Figure 1.

Experimental paradigm. Stimuli A and B were presented in short sequences (three to five stimuli; a–d show selected examples). Subjects were instructed to indicate the third occurrence of the first item seen in a sequence via button press. In half of the sequences, there was a repetition of identical stimuli in the beginning (repetition trials; a, b). In the other half, stimuli changed in the very beginning (lure trials; c, d). We included only the first and second stimulus in further analyses as indicated by the blue line. R indicates the appropriate response for each example sequence. We refer to the first stimulus in each sequence as “first presentation” although only the two scenes displayed in the figure were used in the entire experiment and therefore each scene was highly familiar. Note that the length and order of the sequences was counterbalanced between stimulus A and B repetitions and lures.

Figure 2.

Segmentation of MTL subregions and hippocampal subfields. Hippocampal ROIs were traced on high-resolution T2* images (A, B), whereas EC, PRC, and PHC were traced on the high-resolution T1 images. All ROIs were later coregistered and resliced to the mean functional image (C). sub, Subiculum.

Figure 3.

Univariate group results. A, Custer in CA1 (top row) shows significantly more activity in first presentations compared with repetitions. However, the cluster in the anterior DG (bottom row) shows significantly higher activation in lures compared with repetitions. Both clusters are in the right HC. Results are small-volume corrected for the bilateral HC (p_voxel-level < 0.005 and p_cluster < 0.05). Images on the right show a coronal view and those on the left show a sagittal view. B, Beta estimates extracted from clusters in DG (left) and CA1 (right). Error bars indicate SEM. C, Stimulus-specific novelty contrasts. Clusters of activity resulting from lure-related novelty contrasts of individual stimuli. A, Lure A > repetition A; B, lure B > repetition B. C, sagittal view. Results are illustrated on the group T1 template (p_voxel-level < 0.005, k ≥ 10 voxels).

Figure 4.

Beta estimates across MTL regions. Beta estimates were extracted for each MTL region and hippocampal subfield for first presentations, repetitions, and lures.

Figure 5.

Classification accuracies for first presentations, repetitions, and lures across hippocampal subfields and extrahippocampal MTL structures. A linear classifier (linear support vector machine) was trained to distinguish between the presentation of stimulus A and B. A, Although classification accuracies (in percentages) for hippocampal subfields are not significantly different from chance level (50%) in first presentations and repetitions, they exceed chance level significantly for lure-related activity in the DG (permutation testing: p = 0.001, t test: p < 0.001). B, Classification accuracies (in percentges) in extrahippocampal regions in the MTL are neither significantly different from chance level in first presentations, repetitions, and lures. The dashed line indicates chance level. Error bars indicate SEM. sub, Subiculum.

Figure 6.

Simulation of degree of representational overlap of the two stimuli in DG as a function of trial type. A, We assume that the underlying patterns of stimulus A and B are partially overlapping in DG, consistent with theoretical models (O'Reilly and McClelland, 1994; Knierim and Neunuebel, 2016). Although stimulus A activates units 1–5, stimulus B activates units 3–7. We further assume that first presentations activate these units maximally, whereas repetitions activate these units with decreased activity. For lure trials, we assume that the initially activated units (e.g., from stimulus A) are inactive by the time of the lure presentation ∼3.5 s later (e.g., stimulus B). B, Euclidean distance between the two patterns of activity (i.e., relating to stimulus A and B) as a function of trial type (first presentation, repetition, lure) and the amount of repetition suppression (i.e., the r parameter). See main text for details.