Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Mixed selectivity morphs population codes in prefrontal cortex

Nature Neuroscience volume 20, pages 1770–1779 (2017)Cite this article

Subjects

Abstract

The prefrontal cortex maintains working memory information in the presence of distracting stimuli. It has long been thought that sustained activity in individual neurons or groups of neurons was responsible for maintaining information in the form of a persistent, stable code. Here we show that, upon the presentation of a distractor, information in the lateral prefrontal cortex was reorganized into a different pattern of activity to create a morphed stable code without losing information. In contrast, the code in the frontal eye fields persisted across different delay periods but exhibited substantial instability and information loss after the presentation of a distractor. We found that neurons with mixed-selective responses were necessary and sufficient for the morphing of code and that these neurons were more abundant in the lateral prefrontal cortex than the frontal eye fields. This suggests that mixed selectivity provides populations with code-morphing capability, a property that may underlie cognitive flexibility.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on Springer Link
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Experimental design.
Fig. 2: Population decoding in the LPFC and FEF.
Fig. 3: Principal components analysis (PCA) of the population code.
Fig. 4: Decoding of error trials.
Fig. 5: Mixed selective responses.
Fig. 6: Roles of different types of neurons in code morphing.

Similar content being viewed by others

References

  1. Fuster, J. M. & Alexander, G. E. Neuron activity related to short-term memory. Science 173, 652–654 (1971).

    Article  CAS  PubMed  Google Scholar 

  2. Funahashi, S. Functions of delay-period activity in the prefrontal cortex and mnemonic scotomas revisited. Front. Syst. Neurosci 9, 2 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  3. Suzuki, M. & Gottlieb, J. Distinct neural mechanisms of distractor suppression in the frontal and parietal lobe. Nat. Neurosci. 16, 98–104 (2013).

    Article  CAS  PubMed  Google Scholar 

  4. Stamm, J. S. Electrical stimulation of monkeys’ prefrontal cortex during delayed-response performance. J. Comp. Physiol. Psychol. 67, 535–546 (1969).

    Article  CAS  PubMed  Google Scholar 

  5. Wegener, S. P., Johnston, K. & Everling, S. Microstimulation of monkey dorsolateral prefrontal cortex impairs antisaccade performance. Exp. Brain Res. 190, 463–473 (2008).

    Article  PubMed  Google Scholar 

  6. Cohen, J. D. et al. Activation of the prefrontal cortex in a nonspatial working memory task with functional MRI. Hum. Brain Mapp. 1, 293–304 (1994).

    Article  CAS  PubMed  Google Scholar 

  7. Everling, S., Tinsley, C. J., Gaffan, D. & Duncan, J. Filtering of neural signals by focused attention in the monkey prefrontal cortex. Nat. Neurosci. 5, 671–676 (2002).

    Article  CAS  PubMed  Google Scholar 

  8. Sakai, K., Rowe, J. B. & Passingham, R. E. Active maintenance in prefrontal area 46 creates distractor-resistant memory. Nat. Neurosci. 5, 479–484 (2002).

    Article  CAS  PubMed  Google Scholar 

  9. Romo, R., Brody, C. D., Hernández, A. & Lemus, L. Neuronal correlates of parametric working memory in the prefrontal cortex. Nature 399, 470–473 (1999).

    Article  CAS  PubMed  Google Scholar 

  10. Miller, E. K., Erickson, C. A. & Desimone, R. Neural mechanisms of visual working memory in prefrontal cortex of the macaque. J. Neurosci. 16, 5154–5167 (1996).

    CAS  PubMed  Google Scholar 

  11. Stokes, M. G. et al. Dynamic coding for cognitive control in prefrontal cortex. Neuron 78, 364–375 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Asaad, W. F., Rainer, G. & Miller, E. K. Neural activity in the primate prefrontal cortex during associative learning. Neuron 21, 1399–1407 (1998).

    Article  CAS  PubMed  Google Scholar 

  13. Mansouri, F. A., Matsumoto, K. & Tanaka, K. Prefrontal cell activities related to monkeys’ success and failure in adapting to rule changes in a Wisconsin Card Sorting Test analog. J. Neurosci. 26, 2745–2756 (2006).

    Article  CAS  PubMed  Google Scholar 

  14. Hussar, C. R. & Pasternak, T. Flexibility of sensory representations in prefrontal cortex depends on cell type. Neuron 64, 730–743 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Warden, M. R. & Miller, E. K. Task-dependent changes in short-term memory in the prefrontal cortex. J. Neurosci. 30, 15801–15810 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  16. Noudoost, B. & Moore, T. The role of neuromodulators in selective attention. Trends Cogn. Sci. 15, 585–591 (2011).

    Article  PubMed  PubMed Central  Google Scholar 

  17. Liebe, S., Hoerzer, G. M., Logothetis, N. K. & Rainer, G. Theta coupling between V4 and prefrontal cortex predicts visual short-term memory performance. Nat. Neurosci. 15, 456–462 (2012). S1–S2.

    Article  CAS  PubMed  Google Scholar 

  18. Rigotti, M. et al. The importance of mixed selectivity in complex cognitive tasks. Nature 497, 585–590 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  19. Sarma, A., Masse, N. Y., Wang, X.-J. & Freedman, D. J. Task-specific versus generalized mnemonic representations in parietal and prefrontal cortices. Nat. Neurosci. 19, 143–149 (2016).

    Article  CAS  PubMed  Google Scholar 

  20. Fusi, S., Miller, E. K. & Rigotti, M. Why neurons mix: high dimensionality for higher cognition. Curr. Opin. Neurobiol. 37, 66–74 (2016).

    Article  CAS  PubMed  Google Scholar 

  21. Jacob, S. N. & Nieder, A. Complementary roles for primate frontal and parietal cortex in guarding working memory from distractor stimuli. Neuron 83, 226–237 (2014).

    Article  CAS  PubMed  Google Scholar 

  22. Murray, J. D. et al. Stable population coding for working memory coexists with heterogeneous neural dynamics in prefrontal cortex. Proc. Natl. Acad. Sci. USA 114, 394–399 (2017).

    Article  CAS  PubMed  Google Scholar 

  23. Enel, P., Procyk, E., Quilodran, R. & Dominey, P. F. Reservoir computing properties of neural dynamics in prefrontal cortex. PLOS Comput. Biol. 12, e1004967 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  24. Miller, E. K. & Fusi, S. Limber neurons for a nimble mind. Neuron 78, 211–213 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  25. Sreenivasan, K. K., Curtis, C. E. & D’Esposito, M. Revisiting the role of persistent neural activity during working memory. Trends Cogn. Sci. 18, 82–89 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  26. Stokes, M. G. ‘Activity-silent’ working memory in prefrontal cortex: a dynamic coding framework. Trends Cogn. Sci. 19, 394–405 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  27. Duncan, J. An adaptive coding model of neural function in prefrontal cortex. Nat. Rev. Neurosci. 2, 820–829 (2001).

    Article  CAS  PubMed  Google Scholar 

  28. Mante, V., Sussillo, D., Shenoy, K. V. & Newsome, W. T. Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature 503, 78–84 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  29. Meyers, E. M., Freedman, D. J., Kreiman, G., Miller, E. K. & Poggio, T. Dynamic population coding of category information in inferior temporal and prefrontal cortex. J. Neurophysiol. 100, 1407–1419 (2008).

    Article  PubMed  PubMed Central  Google Scholar 

  30. Buschman, T. J. & Miller, E. K. Top-down versus bottom-up control of attention in the prefrontal and posterior parietal cortices. Science 315, 1860–1862 (2007).

    Article  CAS  PubMed  Google Scholar 

  31. Qi, X.-L. et al. Comparison of neural activity related to working memory in primate dorsolateral prefrontal and posterior parietal cortex. Front. Syst. Neurosci 4, 12 (2010).

    PubMed  PubMed Central  Google Scholar 

  32. Salazar, R. F., Dotson, N. M., Bressler, S. L. & Gray, C. M. Content-specific fronto-parietal synchronization during visual working memory. Science 338, 1097–1100 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  33. Qi, X. L., Elworthy, A. C., Lambert, B. C. & Constantinidis, C. Representation of remembered stimuli and task information in the monkey dorsolateral prefrontal and posterior parietal cortex. J. Neurophysiol. 113, 44–57 (2015).

    Article  PubMed  Google Scholar 

  34. Spaak, E., Watanabe, K., Funahashi, S. & Stokes, M. G. Stable and dynamic coding for working memory in primate prefrontal cortex. J. Neurosci. 37, 6503–6516 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  35. Herbst, J. A., Gammeter, S., Ferrero, D. & Hahnloser, R. H. Spike sorting with hidden Markov models. J. Neurosci. Methods 174, 126–134 (2008).

    Article  PubMed  Google Scholar 

  36. Brainard, D. H. The psychophysics toolbox. Spat. Vis 10, 433–436 (1997).

    Article  CAS  PubMed  Google Scholar 

  37. Bruce, C. J., Goldberg, M. E., Bushnell, M. C. & Stanton, G. B. Primate frontal eye fields. II. Physiological and anatomical correlates of electrically evoked eye movements. J. Neurophysiol 54, 714–734 (1985).

    Article  CAS  PubMed  Google Scholar 

  38. Buschman, T. J., Siegel, M., Roy, J. E. & Miller, E. K. Neural substrates of cognitive capacity limitations. Proc. Natl. Acad. Sci. USA 108, 11252–11255 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  39. Ojala, M. & Garriga, G. C. Permutation tests for studying classifier performance. J. Mach. Learn. Res. 11, 1833–1863 (2010).

    Google Scholar 

  40. Hedges, L. V. Distribution theory for Glass’s estimator of effect size and related estimators. J. Educ. Behav. Stat. 6, 107–128 (1981).

    Article  Google Scholar 

  41. Fisher, R. A. The use of multiple measurements in taxonomic problems. Ann. Hum. Genet 7, 179–188 (1936).

    Google Scholar 

Download references

Acknowledgements

We thank A. Tan for comments and suggestions on an earlier version of this manuscript. We thank C. Lim, M.N. Lynn, L. Chan, K. Chng, and E.M. Peña for help with animal training, surgery, and care. This work was supported by startup grants from the Ministry of Education Tier 1 Academic Research Fund and SINAPSE to C.L., a grant from the NUS-NUHS Memory Networks Program to S.-C.Y., and a grant from the Ministry of Education Tier 2 Academic Research Fund to C.L. and S.-C.Y. (MOE2016-T2-2-117).

Author information

Author notes

  1. Aishwarya Parthasarathy

    Present address: Institute of Molecular and Cell Biology, A*STAR, Singapore, Singapore

  2. Camilo Libedinsky and Shih-Cheng Yen contributed equally to this work.

Authors and Affiliations

  1. NUS Graduate School of Integrative Science and Engineering, National University of Singapore (NUS), Singapore, Singapore

    Aishwarya Parthasarathy & Shih-Cheng Yen

  2. Department of Electrical and Computer Engineering, NUS, Singapore, Singapore

    Roger Herikstad, Jit Hon Bong & Shih-Cheng Yen

  3. Department of Psychology, NUS, Singapore, Singapore

    Felipe Salvador Medina & Camilo Libedinsky

  4. Singapore Institute for Neurotechnology, NUS, Singapore, Singapore

    Camilo Libedinsky

  5. Institute of Molecular and Cell Biology, A*STAR, Singapore, Singapore

    Camilo Libedinsky

Authors
  1. Aishwarya Parthasarathy
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Roger Herikstad
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Jit Hon Bong
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Felipe Salvador Medina
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Camilo Libedinsky
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Shih-Cheng Yen
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

C.L., A.P., R.H., J.H.B., and S.-C.Y. designed the experiments. A.P., R.H., and J.H.B. collected behavioral and electrophysiological data. F.S.M. and A.P. performed the microstimulation verification experiments. A.P. and R.H. analyzed behavioral and electrophysiological data. S.-C.Y. and C.L. guided the data analysis. All authors discussed the results, and A.P., C.L., and S.-C.Y. wrote the manuscript.

Corresponding authors

Correspondence to Camilo Libedinsky or Shih-Cheng Yen.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Integrated supplementary information

Supplementary Figure 1 Single-neuron and population decoding responses in the LPFC and FEF.

(a). Peristimulus time histograms (PSTHs) of the response of an example LPFC neuron to targets presented at each of 8 locations. (b) The average percentage of explained variance (PEV) for selective neurons in the LPFC to the target (red, also shown in red in Fig. 1d) and distractor (green) location are plotted as a function of time in the trial. The shaded regions around the line plots indicate the 95th percentile range (two-sided). (c) The heat map shows the cross-temporal population decoding performance for the distractor location in the LPFC. (d – f) Same plots as (a – c) but for FEF.

Supplementary Figure 2 Cross-temporal decoding in the LPFC and FEF.

(a, b) The cross-temporal decoding performance, LP 11 (50.5 ± 0.9%), is shown for classifiers trained and tested with data from Delay 1 (800 – 1300 ms after target onset, highlighted by the red bar in the bottom plot, which is the average performance in the square region labelled LP 11 in Fig. 2a). When these classifiers were tested in Delay 2 (2100 – 2600 ms after target onset), the performance, LP 12 , was 32.4 ± 1.0% (the magenta bar in the bottom plot). This was still above chance (which was 1 out 7 = 14.3%), but significantly lower than LP 11 . From these results, it was clear that the distractor caused a significant drop in the performance of the classifiers trained in Delay 1 (shown in the middle box plot in (b), p < 0.001). However, it was not clear if there was an overall decrease in target information in Delay 2 after the presentation of the distractor, or if the target information was now encoded differently. To answer this question, we looked at classifiers trained and tested in the same 500 ms window in Delay 2. We found LP 22 to be 47.9 ± 0.9% (shown in the red bar in the top plot), which was not significantly different from LP 11 (the difference is shown in the left box plot in (b), p ~ 0.07). This showed that there was very little loss in information in Delay 2 after the presentation of the distractor. This meant that the reason for the drop in performance for the Delay 1 classifiers when tested in Delay 2, was a morphing in the population code in Delay 2 that encoded target information differently from Delay 1, and not due to an overall decrease in target information in Delay 2. Interestingly, if the classifiers trained in Delay 2 were tested on the data in Delay 1, the performance decreased significantly again (LP 21 = 37.2 ± 0.1%, shown in the magenta bar in the top plot). The change in performance is again plotted in (b) (right box plot, p < 0.001). This meant that the morphed code in Delay 2 incorporated very little of the code in Delay 1, resulting in significantly poorer performance when tested on the responses in Delay 1, suggesting a fairly large transformation occurred. The error bars in (b) indicate the 95th percentile range (two sided) obtained through 1000 measurements of the performance. (c) The shift in the cluster centers from Delay 1 to Delay 2 (i.e. the blue to the red dots in Fig. 3a & b) for each cluster is plotted here. All the shifts found in the Delay 1 space and Delay 2 space were significantly larger than chance (the 97.5th percentile of the shifts obtained by chance is denoted by the dashed line, p < 0.001 for all locations, from left to right. (d – e) Same plots as (a, b), but for the FEF. The performance of classifiers trained in Delay 1, went from FP 11 = 39.0 ± 0.9% when tested in Delay 1, to FP 12 = 28.8 ± 0.7% when tested in Delay 2 (the difference is shown in the middle box plot in (e), p < 0.001). This appeared to be similar to what we observed in the LPFC. However, when we trained and tested classifiers in Delay 2, the performance was found to be FP 22 = 31.7 ± 0.7% (the difference is shown in the left box plot in (e), p < 0.001). These results suggest that there was an overall loss in information in Delay 2, consistent with a degraded code, rather than a change in the encoding of the information. Surprisingly, the classifiers trained in Delay 2 when tested in Delay 1 produced a performance of FP 21 = 30.6 ± 0.1% in Delay 1, which was not significantly different from the performance found in Delay 2 (the difference is shown in the right box plot in (e), p ~ 0.06). This was quite different from what happened in the LPFC, where we observed a significant drop from LP 22 to LP 21 . This appears to be an interesting asymmetry in the performance that warrants further study. (f) The shift in the cluster centers for each cluster is plotted here. All the shifts were not significantly larger than chance (p ~ 0.75, ~ 0.98, ~ 0.51, ~ 0.81, ~ 0.94, ~ 0.30, ~ 0.97, and ~ 0.67 for Delay 1; ~ 0.99, ~ 0.12, ~ 0.48, ~ 0.34, ~ 0.47, and ~ 0.16 for Delay 2). n.s. - not significant. ** - indicates that the 2.5th and the 97.5th percentile range did not overlap with 0 (in b and e) or with the 95th percentile range obtained from a chance distribution. The horizontal red line in the box plots represents the median of the distribution.

Supplementary Figure 3 Cross-temporal decoding performance in the FEF before saccade onset.

Cross-temporal decoding performance for populations in the FEF leading up to the saccade onset. To underscore the role of the FEF in controlling eye movements, the decoding performance increases dramatically right before the saccade onset. This increase in information may be coming from the morphed activity obtained from the LPFC.

Supplementary Figure 4 Single-session cross-temporal decoding in the LPFC and FEF.

(a – h) Cross-temporal decoding results and shift between Delay 1 and Delay 2 responses for simultaneously recorded neurons in the LPFC. The number of neurons recorded is shown in parenthesis above the heat map. The numbers below the heat maps summarize the key decoding performance levels, the number of neurons with NMS, and the mean F-statistic for those neurons. The following are the results of statistical tests for Delay 1 space in plots f, g, h, and i, correspondingly – p < 0.001, g = 7.26; p < 0.001, g = 14.88; p ~ 0.05, g = 4.15; and p < 0.001, g = 10.61. The following are the results of statistical tests for Delay 2 space in plots f, g, h, and i, correspondingly – p < 0.001, g = 7.15; p < 0.001, g = 12.27; p < 0.001, g = 4.19; and p < 0.001, g = 8.85. The shift in cluster centers in all the sessions were significantly different from the distributions obtained by chance in Delay 1 and Delay 2 space for all the sessions. (e, j) The average heat map and shift obtained by pooling over 8 simultaneously recorded sessions. The shift in cluster centers shown in j were significantly different from those obtained by chance for Delay 1 (p < 0.001, g = 23.46) and Delay 2 space (p < 0.001, g = 23.61). (k – r) Results for simultaneously recorded neurons in the FEF. The following are the results of statistical tests for Delay 1 space in plots p, q, r, and s, correspondingly – p ~ 0.93; p ~ 0.8; p ~ 0.99, and p ~0.83. The following are the results of statistical tests for Delay 2 space in plots p, q, r, and s, correspondingly – p ~ 0.94; p ~ 0.94; p ~ 0.99, and p ~ 0.43. The shift in cluster centers were not significantly different from the distribution obtained by chance in Delay 1 and Delay 2 space for any of the sessions. (o, t) Same as for (e, j) but for FEF neurons. The shift in cluster centers shown in t were not significantly different from those obtained by chance for Delay 1 (p ~ 0.94) and Delay 2 space (p ~ 0.88). The error bars in the box plots indicate the 95th percentile range (two-sided) of the distribution. The horizontal dashed line in the box plot indicate the 97.5th percentile of the distribution obtained by chance. The horizontal red line in the box plots indicate the median of the distribution. n.s. – not significant. ** - indicates that the 95th percentile range (two-sided) do not overlap for the distributions being compared.

Supplementary Figure 5 Cross-correlation of target PEV.

(a) The heat map shows the correlation between all possible pairs of time-points of the target PEVs for 256 LPFC neurons. (b) Similar to Supplementary Figure 2(b), we found evidence (using PEV correlation) of 2 distinct and stable population codes that exhibited high correlation coefficients within the same delay period, but significantly lower correlation coefficients between delay periods. The errorbars indicate the 95th percentile range obtained using 1000 sets of bootstrapped trials. (c – d) Equivalent plots for FEF neurons. Using PEV correlation, there appeared to be no significant differences between the correlation coefficients in the different periods, suggesting that there was one population code that persisted throughout the 2 delay periods.

Supplementary Figure 6 Cross-content decoding.

(a) The heat maps show the LPFC target decoding performance for classifiers trained on trials with distractors at only one location (b) The boxplots indicate the mean target decoding performance for trials with distractors at the same location that was used for training (i.e. the within-condition, indicated by “w”), as well as for trials with distractors at different locations (i.e. cross-condition, indicated by “c”). (c) The decoding performance for within-condition trials are compared to the mean decoding performance for cross-condition trials (p-values from left to right: 0.9412, 0.8050, 0.8387,0.9105, 0.9135, 0.5465, 0.7623). (d – f) Equivalent plots for FEF neurons (p-values for (f): 0.8715, 0.8243, 0.9377, 0.9295, 0.9388, 0.9097, 0.6160).

Supplementary Figure 7 Distractor specific code morphing.

(a) The responses of the LPFC population in Delay 2 when projected onto the Delay 2 space are shown for 4 of the target locations. The responses were taken from one of the single-session recordings discussed in Supplementary Figure 4. The responses are color coded by distractor location, and it is clear that there is no clustering of responses by distractor location. (b) The mean cluster separation, a measure used to quantify how well separated clusters of points are from other points, for the points shown in (a) are plotted in (b) for each of the 4 target locations. The cluster separation of the same responses in Delay 1 (where there should be no cluster separation as the distractor has not been presented yet) are shown for comparison. The results show no significant clustering by distractor location in the responses (p-value from left to right: 0.9070, 0.9410, 0.8040, 0.9310, 0.8350, 0.9580, 0.7270, 0.9110). (c) The cluster separation values from another single-session recording (Session 1 in Supp. Fig. 4) are plotted here, and also did not show any significant clustering (p-value from left to right: 0.9290, 0.8930, 0.8380, 0.8200, 0.9220, 0.7330, 0.6580 and 0.8710) by distractor location. (d – f) Equivalent plots for FEF neurons (p-value in (e) from left to right: 0.8840, 0.9360, 0.9550, 0.9420, 0.9380, 0.9120, 0.9160, 0.9710; and p-value in (f) from left to right: 0.9030, 0.7470, 0.6500, 0.8700, 0.8370, 0.8360, 0.6840, 0.9870).

Supplementary Figure 8 Projections on principal components for neurons with different selectivities in the LPFC.

(a) The pie chart shows the distribution of different types of errors made by Monkey A. (b) The pie chart shows the distribution of different types of errors made by Monkey B.

Supplementary Figure 9 Activity from correct and error trials in the LPFC and FEF neurons projected onto the Delay 1 and Delay 2 response spaces.

(a) The responses of the LPFC population in Delay 1 when projected onto the first two principal components of the Delay 1 responses are shown for different target locations (correct and error trials are plotted with cyan and blue dots, respectively). The responses of the LPFC population in Delay 2 when projected onto the same components are also shown (correct and error trials are plotted with red and brown dots, respectively). (b) Equivalent plots to (a) but for Delay 2 space. (c, d) Equivalent plots to (a, b) but for FEF data.

Supplementary Figure 10 Projections on principal components for neurons with different selectivities in the LPFC.

We constructed a pseudo population of LPFC neurons consisting of equal numbers of neurons with CS, LMS, and NMS (27 neurons each, as in Fig. 6). (a) We performed PCA on the responses of the population without neurons with NMS, during Delay 1, and projected the responses of both Delay 1 (plotted in blue) and Delay 2 (plotted in red) onto the top two principal components. (b) Same as (a), but with responses projected onto the PCA space built with Delay 2 activity. (c – d) Same as (a – b), but with responses of the population comprised of only of neurons with NMS. (e – f) Same as (a – b), but with responses of the population without neurons with LMS. (g – h) Same as (a – b), but with responses of a pseudo population comprised only of neurons with LMS. (i – j) Same as (a – b), but with responses of a population without neurons with CS. (k – l) Same as (a – b), but with responses of a pseudo population comprised only of neurons with CS.

Supplementary Figure 11 Injection/removal of neurons with NMS.

(a) The heat map shows the cross-temporal decoding performance after neurons with NMS were added to the FEF population. (b, c) The results of the state space analysis are plotted as in the previous figure. (d) The average shift in the cluster centers between the Delay 1 and Delay 2 responses are shown in the Delay 1 (significant, p < 0.001, g = 9.53) and Delay 2 space (significant, p < 0.001, g = 10.16). (e) The heat map shows the cross-temporal decoding performance after the population of neurons with NMS in the LPFC were matched to those in the FEF in terms of effect sizes and numbers. (f, g) The results of the state space analysis are plotted as in the previous figure. (h) The average shift in the cluster centers between the Delay 1 and Delay 2 responses are shown in the Delay 1 (non-significant, p ~ 0.24 and Delay 2 space (non-significant, p ~ 0.13. By making the neurons with NMS in the LPFC more like those in the FEF, the code morphing in the LPFC was no longer significantly different from chance. n.s. - not significant. ** - indicates that the 2.5th and 97.5th percentile ranges did not overlap with the chance level. Horizontal dashed lines represent the 97.5th percentile of the shifts obtained by chance. The horizontal red line in box plots represents the median of the distribution.

Supplementary Figure 12 Excluding neurons without selectivity in both delay periods.

(a) The pie charts show the distribution of neurons with NMS, LMS, and CS after excluding neurons with NMS that exhibited selectivity in only one of the delay periods. (b) The effect sizes quantified by the F-statistic for the neurons with NMS those were selective in both delay periods. The plot shows that the mean F-statistic for FEF was significantly lower than LPFC (p < 0.001) (c – j) Equivalent to Figure 6, except with neurons with NMS that were selective in both delay periods. When we eliminated the neurons with NMS, the code morphing was eliminated, as seen by a lack of shift in cluster centers between Delay 1 (p ~ 0.53) and Delay 2 (p ~ 0.4). On the other hand, when we eliminated the neurons with LMS or CS, code morphing was preserved (Delay 1: p < 0.001 and g = 3.58 for the subpopulation without LMS neurons, p < 0.001 and g = 6.79 for the subpopulation without CS neurons, Delay 2: p < 0.001 and g = 3.03 for the subpopulation without LMS neurons, p < 0.001 and g = 6.61 for the subpopulation without CS neurons). When we carried out the same analysis separately on homogeneous populations of neurons with NMS, LMS, and CS, we found that neurons with NMS were responsible for a large change in code (Delay 1: p < 0.001, g = 5.13; Delay 2: p < 0.001, g = 7.64), neurons with LMS exhibited a small change in code (Delay 1: p < 0.001, g = 1.23; Delay 2: p < 0.001, g = 2.41), and neurons with CS had no role in changing the code (Delay 1: p ~ 0.98; Delay 2: p ~ 0.99). However, the change in code in LMS neurons were significantly lower than NMS neurons (Delay 1: p < 0.001, g = 1.47; Delay 2: p < 0.001, g = 2.91)

Supplementary Figure 13 Example of neurons with target–distractor nonlinear mixed selectivity (tdNMS).

(a) The plots show the firing rates of a neuron when the target was presented at each of the eight locations. The gray vertical shaded region indicates the target presentation period (time 0 was the onset of the target). The black dashed solid horizontal lines indicate the mean firing rate 300 ms prior to the target presentation. The red solid line and the pink shaded trace indicate the mean and standard error of the firing rates computed in windows of 50 ms time windows. This neuron exhibited almost no spatial tuning during the target presentation period. (b) These plots show the firing rates of the same neuron in (a) before and after the presentation of the distractor (time 0 was the onset of the distractor) at each of the eight locations. In this case, the neuron exhibited weak spatial tuning towards locations on the left side of the grid during the distractor presentation period. (c) The trials were sorted according to target-distractor pairs and the trial averaged firing rate of these combinations were plotted for the distractor presentation period and Delay 2 (time 0 was the distractor onset). Each tile in this plot represented the target location and the traces in the panel represented the average firing rates of trials with different distractor locations associated with the target. Each color represented one distractor location. The red horizontal bars below the plots denote time windows in which the firing rate significantly exceeded the pre-target period. This neuron displayed significant nonlinear activity for one target-distractor pair (target at bottom right corner and the distractor at bottom-center location) during the distractor presentation. (d) This heat-map summarizes the z-score activity presented in (c) for the distractor presentation period. The asterisk indicates the target-distractor pair that the neuron was selective to. (e – h) displays an example of a neuron with mixed selectivity exhibiting nonlinearity during Delay 2. (g – h) shows the nonlinear mixed selectivity during Delay 2 (550 – 750 ms after distractor onset). (i) represents the distribution of LPFC neurons exhibiting nonlinear mixed selectivity, while (j) illustrates the distribution of neurons that exhibit linear selectivity, neurons that exhibit both linear selectivity and nonlinear mixed selectivity at different time points, and neurons that only exhibit nonlinear mixed selectivity. (k – l) represents the same for FEF neurons.

Supplementary Figure 14 Decoding without neurons with target–distractor nonlinear mixed selectivity.

(a) The heat map shows the cross-temporal decoding performance after removing the neurons with tdNMS. (b – c) The results of the state space analysis after excluding the neurons with tdNMS are plotted as in Supplementary Figure 7. (d) The average shift in the cluster centers between the Delay 1 and Delay 2 responses are shown in the Delay 1 (significant, p < 0.001, g = 24.15) and Delay 2 space (significant, p < 0.001, g = 24.63). Excluding the neurons with tdNMS did not affect the code morphing observed in the Delay 1 space (compare plots with those shown in Fig. 3d). n.s. - not significant. ** - indicates that the 2.5th and 97.5th percentile ranges did not overlap with the chance level. Horizontal dashed lines represent the 97.5th percentile of the shifts obtained by chance. The horizontal red line in box plots represents the median of the distribution.

Supplementary Figure 15 Receptive field sizes in the LPFC and FEF.

(a – b) The distribution of receptive field sizes for LPFC and FEF neurons, respectively, in Delay 1. (c – d) The distribution of receptive field sizes for LPFC and FEF neurons, respectively, in Delay 2. (e – f) Box plots of the weighted average of the histograms obtained from bootstrapped populations of LPFC and the FEF responses in Delay 1 (p ~ 0.6) and Delay 2 (p ~ 0.54), respectively.

Supplementary information

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Parthasarathy, A., Herikstad, R., Bong, J.H. et al. Mixed selectivity morphs population codes in prefrontal cortex. Nat Neurosci 20, 1770–1779 (2017). https://doi.org/10.1038/s41593-017-0003-2

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41593-017-0003-2

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing