Borbála Brosig, Endre Csikós, Boróka Ferencz, Alex Ilyés, Hunor Kis (as of 02/2026: Erasmus Scholarship, MPI for Human Development), Zsuzsanna Nemecz (as of 02/2026: University of Regensburg, Germany)
Affiliate Researchers
Ulman Lindenberger (Center for Lifespan Psychology, MPI for Human Development), Zoe Ngo (Emmy Noether Group RAVEN, MPI for Human Development), Markus Werkle-Bergner (Center for Lifespan Psychology, MPI for Human Development)
The Hippocampal Circuit and Code for Cognition Lab (HCCCL) was established at the Brain Imaging Centre of the HUN-REN Research Centre for Natural Sciences in Budapest, Hungary, in 2021 and ends its term as a Max Planck Partner Group in 2026. It contributes to the broader scientific effort to understand how memories are represented in the human brain and how these representations are manipulated by neurocognitive operations. Specifically, HCCCL investigates (1) the role of hippocampal subfields and medial temporal cortical regions, as well as their networks, in memory formation, and (2) how structural and functional changes in these networks across the lifespan shape memory representations and operations. To this end, the group conducts longitudinal studies with participants across the lifespan using high-resolution magnetic resonance imaging (MRI) and behavioral methods. In addition, HCCCL develops measurement tools that support valid, reliable, and comparable delineation of hippocampal subfields on structural MRI images. HCCCL closely collaborates with several projects within the Center for Lifespan Psychology at the MPI for Human Development, most notably the Lifespan Rhythms of Memory and Cognition (RHYME) project, as well as the Emmy Noether Group RAVEN | Real-world Applications of Variability in Episodic memory and Neurodevelopment.
During the reporting period, HCCCL has primarily been involved in implementing two longitudinal studies. The MindMap study (2020–2026) explores age-related changes in two neural pathways within the medial temporal cortical circuitry supporting shared operations on distinct representations, and examines how these changes relate to memory outcomes. MindMap follows 100 adults aged 65–85 years over four years, with yearly behavioral assessment of a variety of mnemonic operations targeting distinct memory representations and biannual imaging to assess hippocampal subfield and medial temporal cortical volumetry, task-related activation and functional connectivity, resting-state connectivity, as well as white matter micro- and macrostructure.
Based on computational models, rodent studies, and human work (Keresztes et al., 2017), we hypothesized that the protracted maturation of the human ability to form and retrieve highly specific memories is tied to the protracted maturation of the hippocampal circuitry implementing neural computations that sparsify and orthogonalize overlapping memory representations (Keresztes et al., 2018). To test this hypothesis, the Neural Underpinnings of Memory Ontogeny (NeMO) study (2023–2027), running in collaboration with the MPI for Human Development, follows 130 children initially aged four for three years. NeMO assesses memory performance assumed to depend on neural computations that decrease representational overlap as well as changes of volumes and of micro- and macrostructure of subfields within the medial temporal networks.
Both MindMap and NeMO are long-term projects with primary research outputs emerging after the current reporting period (MindMap: Dissertations Zsuzsanna Nemecz and Endre Csikós; NeMO: Dissertations Hunor Kis and Boróka Ferencz).
Key References
Keresztes, A., Bender, A. R., Bodammer, N. C., Lindenberger, U., Shing, Y. L., & Werkle-Bergner, M. (2017). Hippocampal maturity promotes memory distinctiveness in childhood and adolescence. Proceedings of the National Academy of Sciences of the United States of America, 114(34), 9212–9217. https://doi.org/10.1073/pnas.1710654114
[Y.L.S. and M.W.-B. contributed equally to this work.].
Keresztes, A., Ngo, C. T., Lindenberger, U., Werkle-Bergner, M., & Newcombe, N. S. (2018). Hippocampal maturation drives memory from generalization to specificity. Trends in Cognitive Sciences, 22(8), 676–686. https://doi.org/10.1016/j.tics.2018.05.004
Computational theories of memory propose that a specific hippocampal subfield—the dentate gyrus—serves as the site of mnemonic interference reduction, where similar representations are transformed into sparse, non-overlapping codes through a process known as pattern separation.
While the dentate gyrus is often viewed as a domain-general, universal pattern separator, recent work suggests that cortical regions (e.g., the parahippocampal and perirhinal cortices) may play a role in representational domain-specific (e.g., spatial or object-related) interference reduction. Moreover, evidence for a universal pattern separator needs to meet the criterion of content-independent pattern separation in the dentate gyrus. Using high-resolution functional imaging of the medial temporal lobe in a task assessing mnemonic interference resolution for non-meaningful fractals, Nemecz et al. (2026) showed that the parahippocampal cortex contributes to interference reduction in the spatial domain, whereas the perirhinal cortex contributes to interference reduction in the object domain (Figure 1). The fronto-parietal network additionally supported the encoding of interfering memories. Contrary to expectations, this study did not provide evidence for dentate gyrus involvement in pattern separation for memories of non-meaningful stimuli. While these findings challenge the view of the dentate gyrus as a universal pattern separator, they contribute to more nuanced mechanistic understanding of memory representations and operations within medial temporal and broader cortical networks.
Figure 1. Neural proxies of pattern separation in medial temporal pathways of interference resolution. (a) Design of object and location encoding tasks. Participants incidentally encoded fractal images presented serially, with some fractals repeated exactly as they were presented earlier, whereas some novel fractals were similar to fractals presented earlier either in their features (Object Task), or location (Location Task). First presentation trials were divided into first presentations of later similar trials and later repeats. (b) Expected BOLD responses during encoding reflecting pattern separation, with reduced repetition suppression for similar versus repeat trials. (c–d) Neural response patterns in the Object Task (c) and the Location Task (d) suggesting differentiation between changed and repeat trials are shaded yellow. Significant differences (unadjusted) are denoted with a symbol: *p < .05, **p < .01, ***p < .001. The comparison between repeat and similar trials in the left perirhinal cortex (panel d, Location Task, marked with a gray *) was not significant after correcting for multiple comparisons. CA3DG: CA3 and dentate gyrus, PHC: parahippocampal cortex, PRC: perirhinal cortex. Error bars represent 95% confidence intervals.
Figure 1. Neural proxies of pattern separation in medial temporal pathways of interference resolution. (a) Design of object and location encoding tasks. Participants incidentally encoded fractal images presented serially, with some fractals repeated exactly as they were presented earlier, whereas some novel fractals were similar to fractals presented earlier either in their features (Object Task), or location (Location Task). First presentation trials were divided into first presentations of later similar trials and later repeats. (b) Expected BOLD responses during encoding reflecting pattern separation, with reduced repetition suppression for similar versus repeat trials. (c–d) Neural response patterns in the Object Task (c) and the Location Task (d) suggesting differentiation between changed and repeat trials are shaded yellow. Significant differences (unadjusted) are denoted with a symbol: *p < .05, **p < .01, ***p < .001. The comparison between repeat and similar trials in the left perirhinal cortex (panel d, Location Task, marked with a gray *) was not significant after correcting for multiple comparisons. CA3DG: CA3 and dentate gyrus, PHC: parahippocampal cortex, PRC: perirhinal cortex. Error bars represent 95% confidence intervals.
To further examine pattern separation across representational domains, another study (Ilyés et al., 2024) demonstrated that older adults are less sensitive to similarity during mnemonic discrimination of verbal memory representations compared to young adults (Figure 2). These age differences are potentially driven by differences in pattern separation ability, as suggested by recent evidence (Ilyés et al., 2026) indicating a critical role of the dentate gyrus in mnemonic discrimination in the task used in Ilyés et al. (2024). Together, these findings strengthen the notion that pattern separation contributes to memory specificity across representational domains when representations are anchored in conceptual knowledge.
Building on this line of research, HCCCL is currently exploring the role of hippocampal subfield computations in conceptual knowledge formation in early middle childhood (Dissertation Borbála Brosig) and the age-specific mechanism of mnemonic interference resolution in older adults (Nemecz et al., in press).
Figure 2. Generalized linear mixed-effects modeling of age-differences in sensitivity to similarity of conceptual mnemonic representations. Participants incidentally encoded 200 adjective-noun phrases (e.g., practical lesson) then made an old-new recognition judgment on four different trial types: target (practical lesson), close lure (theoretical lesson), distant lure (instrumental lesson), and foils (exotic zoo). Cosine similarity between targets and lures was calculated from a Hungarian word2vec (Takács & Gábor, 2015) and used to determine distance of lures from targets. To assess sensitivity to similarity during mnemonic discrimination, we used linear mixed-effects modelling. (a) Exponentiated estimates, standard errors, and p-value of the fixed effects in a General Linear Mixed-Effects Model (GLMEM) predicting "old" responses at recognition, based on the fixed and interaction terms of age group (young adults, old adults) and condition at recognition (targets, close lures, distant lures, foils). The intercept is foils and young adults. Estimates lower than 1 correspond to a negative effect, while estimates above 1 correspond to a positive effect on the probability of "old" responses. (b) Exponentiated estimates, standard errors, and p-value of the fixed effects in a GLMEM including various covariates. The model predicts "old" responses at recognition, based on the fixed and interaction terms of age group (young adults, old adults) and condition at recognition (targets, close lures, distant lures, foils), plus the interaction of all the covariates with the respective fixed effect (participant-level covariates with age group, item-level covariates with condition). The intercept is foil and young adults. Only significant estimates of covariate interactions are shown. Estimates lower than 1 correspond to a negative effect, while estimates above 1 correspond to a positive effect on the probability of "old" responses. (c) Percentage of "old" responses for targets, close lures, and distant lures as a function of cosine similarity measured with the word2vec model with red line showing young adults and green line showing old adults. Young adults compared to old adults show a steeper increase in false alarm rate with increasing similarity.
Figure 2. Generalized linear mixed-effects modeling of age-differences in sensitivity to similarity of conceptual mnemonic representations. Participants incidentally encoded 200 adjective-noun phrases (e.g., practical lesson) then made an old-new recognition judgment on four different trial types: target (practical lesson), close lure (theoretical lesson), distant lure (instrumental lesson), and foils (exotic zoo). Cosine similarity between targets and lures was calculated from a Hungarian word2vec (Takács & Gábor, 2015) and used to determine distance of lures from targets. To assess sensitivity to similarity during mnemonic discrimination, we used linear mixed-effects modelling. (a) Exponentiated estimates, standard errors, and p-value of the fixed effects in a General Linear Mixed-Effects Model (GLMEM) predicting "old" responses at recognition, based on the fixed and interaction terms of age group (young adults, old adults) and condition at recognition (targets, close lures, distant lures, foils). The intercept is foils and young adults. Estimates lower than 1 correspond to a negative effect, while estimates above 1 correspond to a positive effect on the probability of "old" responses. (b) Exponentiated estimates, standard errors, and p-value of the fixed effects in a GLMEM including various covariates. The model predicts "old" responses at recognition, based on the fixed and interaction terms of age group (young adults, old adults) and condition at recognition (targets, close lures, distant lures, foils), plus the interaction of all the covariates with the respective fixed effect (participant-level covariates with age group, item-level covariates with condition). The intercept is foil and young adults. Only significant estimates of covariate interactions are shown. Estimates lower than 1 correspond to a negative effect, while estimates above 1 correspond to a positive effect on the probability of "old" responses. (c) Percentage of "old" responses for targets, close lures, and distant lures as a function of cosine similarity measured with the word2vec model with red line showing young adults and green line showing old adults. Young adults compared to old adults show a steeper increase in false alarm rate with increasing similarity.
Ilyés, A., Brosig, B. M., Mező, G., & Keresztes, A. (2026). The human hippocampus can pattern separate memories by meaning. Proceedings of the National Academy of Sciences of the United States of America, 123(31), Article e2603114123. https://doi.org/10.1073/pnas.2603114123
Nemecz, Z., Homolya, I., Ilyés, A., Kis, H., Mező, G., Varga, V. A., Werkle-Bergner, M., & Keresztes, A. (2026). fMRI signals of pattern separation in the neocortex and hippocampus to non-meaningful objects and their spatial location. Imaging Neuroscience, 4, Article IMAG.a.1088. https://doi.org/10.1162/IMAG.a.1088
Nemecz, Z., Ilyés, A., Kerekes, L., Kis, H., Werkle-Bergner, M., & Keresztes, A. (in press). Different object features shape mnemonic discrimination in younger and older adults. Psychology and Aging.
HCCCL has been contributing to international efforts to improve the reliability and comparability of hippocampal subfield measurements. Specifically, the group participated in the development of a consensus-based, histologically grounded harmonized segmentation protocol within the Hippocampal Subfields Group (Daugherty et al., 2026), and in the development of recommendations for quality control procedures and transparent reporting for both manual and automated segmentations to support reproducible large-scale neuroimaging studies of hippocampal subfields (Canada et al., 2024). Building on these advances, HCCCL is currently building an atlas to support automatic segmentation of hippocampal subfields in 4–8-year-old children. In parallel, the group is pursuing two lines of research aimed at improving the measurement of behavioral outcomes of hippocampal computations. First, HCCCL is developing a task battery to assess memory processes relying on hippocampal computations in children, employing a latent variable approach (Dissertation Hunor Kis). Second, HCCCL evaluated the reliability of the mnemonic similarity task—the most widely used behavioral tool to assess pattern separation—and formulated recommendations for the field for its efficient use in longitudinal analyses (Dissertation Zsuzsanna Nemecz).
Key References
Canada, K. L., Mazloum-Farzaghi, N., Rådman, G., Adams, J. N., Bakker, A., Baumeister, H., Berron, D., Bocchetta, M., Carr, V. A., Dalton, M. A., de Flores, R., Keresztes, A., La Joie, R., Mueller, S. G., Raz, N., Santini, T., Shaw, T., Stark, C. E. L., Tran, T. T., Wang, L., Wisse, L. E. M., Wuestefeld, A., Yushkevich, P. A., Olsen, R. K., Daugherty, A. M., & on behalf of the Hippocampal Subfields Group. (2024). A (sub)field guide to quality control in hippocampal subfield segmentation on high-resolution T2-weighted MRI. Human Brain Mapping, 45(15), Article e70004. https://doi.org/10.1002/hbm.70004
Daugherty, A. M., Carr, V., Canada, K. L., Rådman, G., Brown, T., Augustinack, J., Amunts, K., Bakker, A., Berron, D., Burggren, A., Chetelat, G., de Flores, R., Ding, S.-L., Huang, Y., Johnson, E., Kanel, P., Keresztes, A., Kedo, O., Kennedy, K. M., Lee, J., Malykhin, N., Martinez, A., Mueller, S., Mulligan, E., Ofen, N., Palombo, D., Pasquini, L., Pluta, J., Raz, N., Riggins, T., Rodrigue, K. M., Saifullah, S., Schlichting, M. L., Stark, C., Wang, L., Yushkevich, P., La Joie, R., Wisse, L., Olsen, R., Alzheimer’s Disease Neuroimaging Initiative, & Hippocampal Subfields Group. (2026). Harmonized protocol for subfield segmentation in the hippocampal body on high-resolution in vivo MRI from the Hippocampal Subfields Group (HSG). Hippocampus, 36(2), Article e70073. https://doi.org/10.1002/hipo.70073
HCCCL is strongly committed to open science principles. For all publications from 2024 onward, data and code are made freely available whenever ethical constraints permit.
