Patterns across multiple memories are identified over time

Subjects

Abstract

Memories are not static but continue to be processed after encoding. This is thought to allow the integration of related episodes via the identification of patterns. Although this idea lies at the heart of contemporary theories of systems consolidation, it has yet to be demonstrated experimentally. Using a modified water-maze paradigm in which platforms are drawn stochastically from a spatial distribution, we found that mice were better at matching platform distributions 30 d compared to 1 d after training. Post-training time-dependent improvements in pattern matching were associated with increased sensitivity to new platforms that conflicted with the pattern. Increased sensitivity to pattern conflict was reduced by pharmacogenetic inhibition of the medial prefrontal cortex (mPFC). These results indicate that pattern identification occurs over time, which can lead to conflicts between new information and existing knowledge that must be resolved, in part, by computations carried out in the mPFC.

Access options

Rent or Buy article

Get time limited or full article access on ReadCube.

from$8.99

All prices are NET prices.

Figure 1: Passage of time improves behavioral match to pattern.
Figure 2: Pattern matching improves over time with alternate distribution and measures.
Figure 3: Sensitivity to difference between pattern consistency versus conflict increases over time.
Figure 4: Time-dependent sensitivity to pattern conflict with alternate distribution.
Figure 5: CNO inhibits spiking activity in hM4Di infected neurons in the mPFC.
Figure 6: mPFC inhibition impairs learning of conflicting platform after 30-d delay.

References

  1. 1

    Bartlett, F.C. Remembering: A Study in Experimental and Social Psychology (Cambridge University Press, 1932).

  2. 2

    Piaget, J. & Cook, M. The Origins of Intelligence in Children (International Universities Press, 1952).

  3. 3

    Posner, M.I. & Keele, S.W. On the genesis of abstract ideas. J. Exp. Psychol. 77, 353–363 (1968).

    CAS  Article  Google Scholar 

  4. 4

    Knowlton, B. & Squire, L. The learning of categories: parallel brain systems for item memory and category knowledge. Science 262, 1747–1749 (1993).

    CAS  Article  Google Scholar 

  5. 5

    Squire, L.R. & Knowlton, B.J. Learning about categories in the absence of memory. Proc. Natl. Acad. Sci. USA 92, 12470–12474 (1995).

    CAS  Article  Google Scholar 

  6. 6

    Tse, D. et al. Schema-dependent gene activation and memory encoding in neocortex. Science 333, 891–895 (2011).

    CAS  Article  Google Scholar 

  7. 7

    Wang, S.-H., Tse, D. & Morris, R.G.M. Anterior cingulate cortex in schema assimilation and expression. Learn. Mem. 19, 315–318 (2012).

    Article  Google Scholar 

  8. 8

    Winocur, G., Moscovitch, M., Fogel, S., Rosenbaum, R.S. & Sekeres, M. Preserved spatial memory after hippocampal lesions: effects of extensive experience in a complex environment. Nat. Neurosci. 8, 273–275 (2005).

    CAS  Article  Google Scholar 

  9. 9

    Van Kesteren, M.T.R., Fernández, G., Norris, D.G. & Hermans, E.J. Persistent schema-dependent hippocampal-neocortical connectivity during memory encoding and postencoding rest in humans. Proc. Natl. Acad. Sci. USA 107, 7550–7555 (2010).

    CAS  Article  Google Scholar 

  10. 10

    Tse, D. et al. Schemas and memory consolidation. Science 316, 76–82 (2007).

    CAS  Article  Google Scholar 

  11. 11

    Preston, A.R. & Eichenbaum, H. Interplay of hippocampus and prefrontal cortex in memory. Curr. Biol. 23, R764–R773 (2013).

    CAS  Article  Google Scholar 

  12. 12

    Rumelhart, D.E., Smolensky, P., McClelland, J.L. & Hinton, G.E. Schemata and sequential thought processes in PDP models. Parallel Distrib. Process. Explor. Microstruct. Cogn. 2, 7–57 (1986).

    Google Scholar 

  13. 13

    Smolensky, P. Information processing in dynamical systems: foundations of harmony theory. Parallel Distrib. Process. Explor. Microstruct. Cogn. 1, 194–281 (1986).

    Google Scholar 

  14. 14

    Turk-Browne, N.B., Scholl, B.J., Chun, M.M. & Johnson, M.K. Neural evidence of statistical learning: efficient detection of visual regularities without awareness. J. Cogn. Neurosci. 21, 1934–1945 (2009).

    Article  Google Scholar 

  15. 15

    Durrant, S.J., Taylor, C., Cairney, S. & Lewis, P.A. Sleep-dependent consolidation of statistical learning. Neuropsychologia 49, 1322–1331 (2011).

    Article  Google Scholar 

  16. 16

    Lewis, P.A. & Durrant, S.J. Overlapping memory replay during sleep builds cognitive schemata. Trends Cogn. Sci. 15, 343–351 (2011).

    Article  Google Scholar 

  17. 17

    Durrant, S.J., Cairney, S.A. & Lewis, P.A. Overnight consolidation aids the transfer of statistical knowledge from the medial temporal lobe to the striatum. Cereb. Cortex 23, 2467–2478 (2013).

    Article  Google Scholar 

  18. 18

    Djonlagic, I. et al. Sleep enhances category learning. Learn. Mem. 16, 751–755 (2009).

    Article  Google Scholar 

  19. 19

    Marr, D. A theory for cerebral neocortex. Proc. R. Soc. Lond. B Biol. Sci. 176, 161–234 (1970).

    CAS  Article  Google Scholar 

  20. 20

    Marr, D. Simple memory: a theory for archicortex. Phil. Trans. R. Soc. Lond. B 262, 23–81 (1971).

    CAS  Article  Google Scholar 

  21. 21

    Nadel, L. & Moscovitch, M. Memory consolidation, retrograde amnesia and the hippocampal complex. Curr. Opin. Neurobiol. 7, 217–227 (1997).

    CAS  Article  Google Scholar 

  22. 22

    Moscovitch, M., Nadel, L., Winocur, G., Gilboa, A. & Rosenbaum, R.S. The cognitive neuroscience of remote episodic, semantic and spatial memory. Curr. Opin. Neurobiol. 16, 179–190 (2006).

    CAS  Article  Google Scholar 

  23. 23

    Winocur, G., Moscovitch, M. & Bontempi, B. Memory formation and long-term retention in humans and animals: Convergence towards a transformation account of hippocampal–neocortical interactions. Neuropsychologia 48, 2339–2356 (2010).

    Article  Google Scholar 

  24. 24

    McClelland, J.L., McNaughton, B.L. & O'Reilly, R.C. Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychol. Rev. 102, 419–457 (1995).

    Article  Google Scholar 

  25. 25

    McClelland, J.L. Incorporating rapid neocortical learning of new schema-consistent information into complementary learning systems theory. J. Exp. Psychol. Gen. 142, 1190–1210 (2013).

    Article  Google Scholar 

  26. 26

    Winocur, G., Moscovitch, M. & Sekeres, M. Memory consolidation or transformation: context manipulation and hippocampal representations of memory. Nat. Neurosci. 10, 555–557 (2007).

    CAS  Article  Google Scholar 

  27. 27

    Wiltgen, B.J. & Silva, A.J. Memory for context becomes less specific with time. Learn. Mem. 14, 313–317 (2007).

    Article  Google Scholar 

  28. 28

    Rosenbaum, R.S. et al. Remote spatial memory in an amnesic person with extensive bilateral hippocampal lesions. Nat. Neurosci. 3, 1044–1048 (2000).

    CAS  Article  Google Scholar 

  29. 29

    Steele, R.J. & Morris, R.G.M. Delay-dependent impairment of a matching-to-place task with chronic and intrahippocampal infusion of the NMDA-antagonist D-AP5. Hippocampus 9, 118–136 (1999).

    CAS  Article  Google Scholar 

  30. 30

    Kullback, S. & Leibler, R.A. On information and sufficiency. Ann. Math. Stat. 22, 79–86 (1951).

    Article  Google Scholar 

  31. 31

    Schultz, W., Dayan, P. & Montague, P.R. A neural substrate of prediction and reward. Science 275, 1593–1599 (1997).

    CAS  Article  Google Scholar 

  32. 32

    Farrell, M.S. & Roth, B.L. Pharmacosynthetics: reimagining the pharmacogenetic approach. Brain Res. 1511, 6–20 (2013).

    CAS  Article  Google Scholar 

  33. 33

    Ferguson, S.M. et al. Transient neuronal inhibition reveals opposing roles of indirect and direct pathways in sensitization. Nat. Neurosci. 14, 22–24 (2011).

    CAS  Article  Google Scholar 

  34. 34

    Baddeley, A. & Hitch, G. The recency effect: Implicit learning with explicit retrieval? Mem. Cognit. 21, 146–155 (1993).

    CAS  Article  Google Scholar 

  35. 35

    Hinton, G.E. & Sejnowski, T.J. Learning and relearning in Boltzmann machines. Parallel Distrib. Process. Explor. Microstruct. Cogn. 1, 282–317 (1986).

    Google Scholar 

  36. 36

    Hinton, G.E., Dayan, P., Frey, B.J. & Neal, R.M. The 'wake-sleep' algorithm for unsupervised neural networks. Science 268, 1158–1161 (1995).

    CAS  Article  Google Scholar 

  37. 37

    Dayan, P. & Hinton, G.E. Varieties of Helmholtz machine. Neural Netw. 9, 1385–1403 (1996).

    Article  Google Scholar 

  38. 38

    Turk-Browne, N.B., Scholl, B.J., Johnson, M.K. & Chun, M.M. Implicit perceptual anticipation triggered by statistical learning. J. Neurosci. 30, 11177–11187 (2010).

    CAS  Article  Google Scholar 

  39. 39

    Berkes, P., Orbán, G., Lengyel, M. & Fiser, J. Spontaneous cortical activity reveals hallmarks of an optimal internal model of the environment. Science 331, 83–87 (2011).

    CAS  Article  Google Scholar 

  40. 40

    Makin, J.G., Fellows, M.R. & Sabes, P.N. Learning multisensory integration and coordinate transformation via density estimation. PLoS Comput. Biol. 9, e1003035 (2013).

    CAS  Article  Google Scholar 

  41. 41

    Kording, K.P. & Wolpert, D.M. Bayesian integration in sensorimotor learning. Nature 427, 244–247 (2004).

    Article  Google Scholar 

  42. 42

    Jaramillo, S. & Zador, A.M. The auditory cortex mediates the perceptual effects of acoustic temporal expectation. Nat. Neurosci. 14, 246–251 (2011).

    CAS  Article  Google Scholar 

  43. 43

    Kepecs, A., Uchida, N., Zariwala, H.A. & Mainen, Z.F. Neural correlates, computation and behavioural impact of decision confidence. Nature 455, 227–231 (2008).

    CAS  Article  Google Scholar 

  44. 44

    Heffner, H.E. & Heffner, R.S. Hearing ranges of laboratory animals. J. Am. Assoc. Lab. Anim. Sci. 46, 20–22 (2007).

    CAS  PubMed  Google Scholar 

  45. 45

    Paxinos, G. & Franklin, K.B.J. The Mouse Brain in Stereotaxic Coordinates Ed. 2nd edn. (Academic, 2001).

  46. 46

    Berens, P. CircStat: a matlab toolbox for circular statistics. J. Stat. Softw. 31, 1–21 (2009).

    Article  Google Scholar 

Download references

Acknowledgements

This work was supported by grants from the Canadian Institutes of Health Research to P.W.F. (MOP-77561), S.A.J. (MOP-74650) and M.A.W. (MOP-123466). B.A.R. received financial support from a Banting Postdoctoral Fellowship via the Natural Sciences and Engineering Research Council of Canada. B.A.R., F.X. and J.H. received financial support from the Canadian Institutes of Health Research Sleep and Biological Rhythms Training Program. A.S. received financial support from the Hospital for Sick Children. We thank G. Hinton and M. Moscovitch for comments on earlier drafts of this manuscript.

Author information

Affiliations

Authors

Contributions

B.A.R., S.A.J. and P.W.F. conceived and designed the experiments. B.A.R., F.X., A.S. and J.H. performed the behavioral experiments. B.A.R. and M.A.W. performed the patch clamp experiments. B.A.R. conducted the analyses. B.A.R., S.A.J. and P.W.F. wrote the paper.

Corresponding author

Correspondence to Paul W Frankland.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Integrated supplementary information

Supplementary Figure 1 Platform selection for training and testing.

(a) Platforms were selected using normal distributions over the platform distance from the center of the pool, d, and Von Mises distributions over the platform angle, θ. The platforms for the first set of experiments (Fig. 1) are shown here. (b) Platforms from the weighted bimodal distribution experiments (Fig. 2) are shown. (c) Platforms from the first set of Consistent vs. Conflict experiments (Fig. 3) are shown. (d) Platforms from the second set of Consistent vs. Conflict experiments (Fig. 4) and the mPFC inhibition experiments (Fig. 6) are shown.

Supplementary Figure 2 Behavioral experiment timelines.

(a) Mice were trained to locate each platform for 4 trials a day, with start locations randomly selected from the 4 cardinal coordinates (N, E, S, W). The start locations are shown. Mice were given a 1 d or 30 d delay after training. The time line for the first experiments (Fig. 1) is shown. (b) Time line for the weighted bimodal distribution experiments (Fig. 2) is shown. (c-d) Time lines for the Consistent vs. Conflict experiments (Fig. 3, Fig. 4 and Fig 6) are shown.

Supplementary Figure 3 Calculating the Kullback-Leibler divergence (DKL).

The platform distribution (Pplatform) was a kernel density estimate based on the training platform locations with a bandwidth selected by a likelihood cross-validation optimization algorithm. The search distributions, Psearch, were estimated via a kernel density estimate using the mouse's location in the pool at each time step with a 5 cm bandwidth. To calculate the DKL between Pplatform and Psearch the natural logarithm of the ratio between the two distributions was taken and multiplied by Pplatform at each location across the pool. The sum across the pool then gave the DKL value in nats. Two example paths and the resulting DKL values are shown here.

Supplementary Figure 4 Escape latencies during training in pattern-matching experiments.

Times to find the platform for (a) the first distribution (Fig. 1), (b) the single platform (Supplementary Fig. 5) and (c) the weighted bimodal distribution (Fig. 2) are shown. Data shown is mean ± s.e.m. Source data

Supplementary Figure 5 Reduction in DKL does not occur spontaneously.

(a) To determine whether mice might naturally gravitate towards improved matching to the distribution used in the first set of experiments (Fig. 1), we performed a set of control experiments. Mice were trained to locate a single platform for 9 days, located at the mean of the original distribution, then probed after a 1 d or 30 d delay. Example paths from a probe 1 d and 30 d later are shown. (b) The DKL between mouse search paths and the original distribution showed a trend towards an increase at the 30 d delay (Mann-Whitney U-test: 1 d n = 11, 30 d n = 12, ranksum = 118, P = 0.41). This was in the opposite direction to the first experiments (Fig. 1d), and indicates that the reduction in DKL depended on exposure to the original platform distribution and did not emerge spontaneously. (c) Mice exhibited decreased accuracy in their memory for the single platform location after a 30 d delay, as shown by the number of platform crossings during the probe (Mann-Whitney U-test: ranksum = 169.5, P = 0.02). (d) The decrease in accuracy did not reflect an increase in the spread of their search paths, as evidenced by the lack of difference in search path entropy (t-test: t21 = 1.68, P = 0.11). Likewise, animals trained on a distribution did not show an increase in entropy after a 30 d delay (t-test 1 d vs. 30 d: t90 = 1.35, P = 0.18). (e) To determine whether the observed decrease in accuracy of a single platform memory could account for a reduced DKL over time if extrapolated to a multiple-platform memory situation, we conducted a Monte Carlo simulation. Simulated probe search path distributions were constructed using kernel density estimation with the 9 platform locations from the first set of experiments (Supplementary Fig. 1a) used as the kernel centers. The kernels were circular Gaussians with standard deviations based on the search paths of the single platform mice (σ = 19 cm). The kernel locations were randomly shuffled in space to simulate memory error. The memory error was sampled from a normal distribution with mean 0 and different standard deviations. 100 samples were drawn for each level of standard deviation, and for each sample the DKL between the simulated path and the actual platform distribution was calculated. As the standard deviation in memory error increased, the DKL also increased. Therefore, an observed change in single platform memory accuracy would not account for a drop in DKL when applied to a multi-platform situation. Data shown is mean ± s.e.m. (*: P < 0.05) Source data

Supplementary Figure 6 Thigmotaxis cannot account for improved pattern matching over time.

(a) In the first set of experiments (Fig. 1), mice spent more time in the 10 cm outer ring of the pool during the probe test following a 30 d delay (Mann-Whitney U-test: ranksum = 2345, P = 9.7 x 10–5). (b) This did not account for the reduction in DKL observed in these groups, though, because there was no correlation between time spent in the outer 10 cm zone and DKL for either the 1 d (Pearson's: r = –0.16, P = 0.26) or 30 d (Pearson's: r = 0.15, P = 0.37) groups and there was no difference between the groups in this level of correlation (z-test: z = 0.31, P = 0.32). (c) Thigmotaxis was further dissociated from DKL by the fact that the mice trained on a single platform also showed increased time spent in the outer 10 cm ring of the pool (t-test: t21 = –3.67, P = 0.001), yet they did not show any reduction in DKL (Supplementary Fig. 5b). (d) Thigmotaxis could also not account for the results in the weighted bimodal distribution experiments (Fig. 2). While mice showed improved pattern matching at the 30 d delay, there was no significant difference in thigmotaxis in the probe tests 1 d vs. 30 d after training (Mann-Whitney U-test: ranksum = 125, P = 0.16). Data shown is individual mice or mean ± s.e.m. (**: P < 0.005) Source data

Supplementary Figure 7 Effects of platform recency on search path.

(a) In the first set of experiments (Fig. 1) there was no significant correlation between platform recency and the time the mice spent in the 10 cm zones surrounding the platforms during the probe test, for either the 1 d delay (Pearson's: r = 0.49, P = 0.18) or 30 d delay (Pearson's: r = 0.16, P = 0.68) groups. Furthermore, these correlations were not different between the 1 d delay and 30 d delay groups (z-test: z = 0.37, P = 0.65). (b) Both groups searched a fair amount in the last platform location, though, which corresponded to the mean of the distribution. (c) In the weighted bimodal distribution experiments (Fig. 2) there was also no significant correlation between platform recency and time spent in platform zone for either the 1 d delay (Pearson's: r = 0.52, P = 0.15) or 30 d delay (Pearson's: r = 0.03, P = 0.93) groups. There was no difference between the groups in this correlation either (z-test: z = 0.55, P = 0.5). However, in these experiments only the 1 d delay groups searched a great deal for the last platform, which was located in a lower probability zone within the distribution (Supplementary Fig. 1b). Data shown is mean ± s.e.m. Source data

Supplementary Figure 8 Escape latencies during training in new platform experiments.

Data shown is escape latencies (time to find the platform) for mice in the Consistent vs. Conflict new platform experiments, including (a) the second set of experiments (Fig. 4), (b) the 1 d delay groups in the mPFC inhibition experiments (Fig. 6a-c) and (c) the 30 d delay groups in the mPFC inhibition experiments (Fig. 6d-f). Data shown is mean ± s.e.m. Source data

Supplementary Figure 9 Complete timeline for mPFC inhibition experiments.

All animals were given at least 1 week to recover from viral microinfusion surgery. The time between the microinfusion and the new platform training (when animals received CNO injections) was kept constant (45 days) to ensure equivalent levels of virus expression in all groups.

Supplementary information

Source data

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Richards, B., Xia, F., Santoro, A. et al. Patterns across multiple memories are identified over time. Nat Neurosci 17, 981–986 (2014). https://doi.org/10.1038/nn.3736

Download citation

Further reading

Search

Sign up for the Nature Briefing newsletter for a daily update on COVID-19 science.
Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing