Abstract
In the brain, decision making is instantiated in dedicated neural circuits. However, there is considerable individual variability in decision-making behavior, particularly under uncertainty. The origins of decision variability within these conserved neural circuits are not known. Here we demonstrate in the rat medial frontal cortex (MFC) that individual variability is a consequence of altered stability in neuronal populations. In a sensory-guided choice task, rats trained on familiar stimuli were exposed to unfamiliar stimuli, resulting in variable choice responses across individuals. We created a recurrent network model to examine the source of variability in MFC neurons, and found that the landscape of neural population trajectories explained choice variability across different unfamiliar stimuli. We experimentally confirmed model predictions showing that trial-by-trial variability in neuronal activity indexes the landscape and predicts individual variation. These results show that neural stability is a critical component of the MFC neural dynamics that underpins individual variation in decision-making.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$209.00 per year
only $17.42 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data availability
The experimental data that support the findings in this study are available from the corresponding author upon reasonable request.
References
Gold, J. I. & Shadlen, M. N. The neural basis of decision making. Annu. Rev. Neurosci. 30, 535–574 (2007).
Machens, C. K., Romo, R. & Brody, C. D. Functional, but not anatomical, separation of ‘what’ and ‘when’ in prefrontal cortex. J. Neurosci. 30, 350–360 (2010).
Wang, X. Decision making in recurrent neuronal circuits. Neuron 60, 215–234 (2008).
Harvey, C. D., Coen, P. & Tank, D. W. Choice-specific sequences in parietal cortex during a virtual-navigation decision task. Nature 484, 62–68 (2012).
Fujisawa, S., Amarasingham, A., Harrison, M. T. & Buzsáki, G. Behavior-dependent short-term assembly dynamics in the medial prefrontal cortex. Nat. Neurosci. 11, 823–833 (2008).
Mazor, O. & Laurent, G. Transient dynamics versus fixed points in odor representations by locust antennal lobe projection neurons. Neuron 48, 661–673 (2005).
Handa, T., Takekawa, T., Harukuni, R., Isomura, Y. & Fukai, T. Medial frontal circuit dynamics represents probabilistic choices for unfamiliar sensory experience. Cereb. Cortex 27, 3818–3831 (2017).
Deco, G., Rolls, E. T. & Romo, R. Synaptic dynamics and decision making. Proc. Natl Acad. Sci. USA 107, 7545–7549 (2010).
Erlich, J. C., Bialek, M. & Brody, C. D. A cortical substrate for memory-guided orienting in the rat. Neuron 72, 330–343 (2011).
Narayanan, N. S., Cavanagh, J. F., Frank, M. J. & Laubach, M. Common medial frontal mechanisms of adaptive control in humans and rodents. Nat. Neurosci. 16, 1888–1895 (2013).
Maass, W., Natschläger, T. & Markram, H. Real-time computing without stable states: a new framework for neural computation based on perturbations. Neural Comput. 14, 2531–2560 (2002).
Jaeger, H. & Haas, H. Harnessing nonlinearity: predicting chaotic systems and saving energy in wireless communication. Science 304, 78–80 (2004).
Sussillo, D. & Abbott, L. F. Generating coherent patterns of activity from chaotic neural networks. Neuron 63, 544–557 (2009).
Laje, R. & Buonomano, D. V. Robust timing and motor patterns by taming chaos in recurrent neural networks. Nat. Neurosci. 16, 925–933 (2013).
Izhikevich, E. M. Solving the distal reward problem through linkage of STDP and dopamine signaling. Cereb. Cortex 17, 2443–2452 (2007).
Williams, R. J. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Mach. Learn. 8, 229–256 (1992).
Seung, H. S. Learning in spiking neural networks by reinforcement of stochastic synaptic transmission. Neuron 40, 1063–1073 (2003).
Condé, F., Maire-lepoivre, E., Audinat, E. & Crépel, F. Afferent connections of the medial frontal cortex of the rat. II. Cortical and subcortical afferents. J. Comp. Neurol. 352, 567–593 (1995).
Hoover, W. B. & Vertes, R. P. Anatomical analysis of afferent projections to the medial prefrontal cortex in the rat. Brain. Struct. Funct. 212, 149–179 (2007).
Reep, R. L., Corwin, J. V., Hashimoto, A. & Watson, R. T. Efferent connections of the rostral portion of medial agranular cortex in rats. Brain Res. Bull. 19, 203–221 (1987).
Teramae, J., Tsubo, Y. & Fukai, T. Optimal spike-based communication in excitable networks with strong-sparse and weak-dense links. Sci. Rep. 2, 485 (2012).
Ikegaya, Y. et al. Interpyramid spike transmission stabilizes the sparseness of recurrent network activity. Cereb. Cortex 23, 293–304 (2013).
Buzsáki, G. & Mizuseki, K. The log-dynamic brain: how skewed distributions affect network operations. Nat. Rev. Neurosci. 15, 264–278 (2014).
Mante, V., Sussillo, D., Shenoy, K. V. & Newsome, W. T. Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature 503, 78–84 (2013).
Sul, J. H., Jo, S., Lee, D. & Jung, M. W. Role of rodent secondary motor cortex in value-based action selection. Nat. Neurosci. 14, 1202–1210 (2011).
Siniscalchi, M. J., Phoumthipphavong, V., Ali, F., Lozano, M. & Kwan, A. C. Fast and slow transitions in frontal ensemble activity during flexible sensorimotor behavior. Nat. Neurosci. 19, 1234–1242 (2016).
Arieli, A., Sterkin, A., Grinvald, A. & Aertsen, A. Dynamics of ongoing activity: explanation of the large variability in evoked cortical responses. Science 273, 1868–1871 (1996).
Churchland, M. M. et al. Stimulus onset quenches neural variability: a widespread cortical phenomenon. Nat. Neurosci. 13, 369–378 (2010).
Churchland, A. K. et al. Variance as a signature of neural computations during decision making. Neuron 69, 818–831 (2011).
Kepecs, A., Uchida, N., Zariwala, H. A. & Mainen, Z. F. Neural correlates, computation and behavioural impact of decision confidence. Nature 455, 227–231 (2008).
Znamenskiy, P. & Zador, A. M. Corticostriatal neurons in auditory cortex drive decisions during auditory discrimination. Nature 497, 482–485 (2013).
Kopec, C. D., Erlich, J. C., Brunton, B. W., Deisseroth, K. & Brody, C. D. Cortical and subcortical contributions to short-term memory for orienting movements. Neuron 88, 367–377 (2015).
Kusev, P. et al. Understanding risky behavior: the influence of cognitive, emotional and hormonal factors on decision-making under risk. Front. Psychol. 8, 1–10 (2017).
Frank, M. J., Doll, B. B., Oas-Terpstra, J. & Moreno, F. Prefrontal and striatal dopaminergic genes predict individual differences in exploration and exploitation. Nat. Neurosci. 12, 1062–1068 (2009).
Finn, E. S. et al. Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity. Nat. Neurosci. 18, 1664–1671 (2015).
Wolff, M. J., Jochim, J., Akyürek, E. G. & Stokes, M. G. Dynamic hidden states underlying working-memory-guided behavior. Nat. Neurosci. 20, 864–871 (2017).
Carnevale, F., de Lafuente, V., Romo, R., Barak, O. & Parga, N. Dynamic control of response criterion in premotor cortex during perceptual detection under temporal uncertainty. Neuron 86, 1067–1077 (2015).
Chaisangmongkon, W., Swaminathan, S. K., Freedman, D. J. & Wang, X.-J. Computing by robust transience: how the fronto-parietal network performs sequential, category-based decisions. Neuron 93, 1504–1517 (2017).
Shenoy, K. V., Kaufman, M. T., Sahani, M. & Churchland, M. M. A dynamical systems view of motor preparation: implications for neural prosthetic system design. Prog. Brain. Res. 192, 33–58 (2011).
Kaufman, M. T., Churchland, M. M., Ryu, S. I. & Shenoy, K. V. Cortical activity in the null space : permitting preparation without movement. Nat. Neurosci. 17, 440–448 (2014).
Rajan, K., Abbott, L. F. & Sompolinsky, H. Stimulus-dependent suppression of chaos in recurrent neural networks. Phys. Rev. E 82, 1–5 (2010).
Curto, C., Sakata, S., Marguet, S., Itskov, V. & Harris, K. D. A simple model of cortical dynamics explains variability and state dependence of sensory responses in urethane-anesthetized auditory cortex. J. Neurosci. 29, 10600–10612 (2009).
Tavor, I. et al. Task-free MRI predicts individual differences in brain activity during task performance. Science 352, 216–220 (2016).
Luczak, A., Bartho, P. & Harris, K. D. Spontaneous events outline the realm of possible sensory responses in neocortical populations. Neuron 62, 413–425 (2009).
Wu, H. G., Miyamoto, Y. R., Gonzalez Castro, L. N., Ölveczky, B. P. & Smith, M. A. Temporal structure of motor variability is dynamically regulated and predicts motor learning ability. Nat. Neurosci. 17, 312–321 (2014).
Singh, P., Jana, S., Ghosal, A. & Murthy, A. Exploration of joint redundancy but not task space variability facilitates supervised motor learning. Proc. Natl Acad. Sci. USA 113, 14414–14419 (2016).
Shim, Y., Philippides, A., Staras, K. & Husbands, P. Unsupervised learning in an ensemble of spiking neural networks mediated by ITDP. PLoS Comput. Biol. 12, 1–41 (2016).
Colgin, L. L. Rhythms of the hippocampal network. Nat. Rev. Neurosci. 17, 239–249 (2016).
Yamamoto, J., Suh, J., Takeuchi, D. & Tonegawa, S. Successful execution of working memory linked to synchronized high-frequency gamma oscillations. Cell 157, 845–857 (2014).
Fries, P. Neuronal gamma-band synchronization as a fundamental process in cortical computation. Annu. Rev. Neurosci. 32, 209–224 (2009).
Takekawa, T., Isomura, Y. & Fukai, T. Spike sorting of heterogeneous neuron types by multimodality-weighted PCA and explicit robust variational Bayes. Front. Neuroinform. 6, 1–13 (2012).
Harris, K. D., Henze, D. A., Csicsvari, J., Hirase, H. & Buzsáki, G. Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements. J. Neurophysiol. 84, 473–478 (2000).
Isomura, Y., Harukuni, R., Takekawa, T., Aizawa, H. & Fukai, T. Microcircuitry coordination of cortical motor information in self-initiation of voluntary movements. Nat. Neurosci. 12, 1586–1593 (2009).
Narayanan, N. S., Horst, N. K. & Laubach, M. Reversible inactivations of rat medial prefrontal cortex impair the ability to wait for a stimulus. Neuroscience 139, 865–876 (2006).
Nakagawa, S. & Cuthill, I. C. Effect size, confidence interval and statistical significance: a practical guide for biologists. Biol. Rev. 82, 591–605 (2007).
Acknowledgements
We thank J. Johansen and C. Yakoyama for critical comments on the manuscript, and M. Tatsuno and K. Watanabe for discussions about the analysis of correlation. This work was partly supported by KAKENHI (grants 16H01289 and 17H06036 to T.F.) from MEXT.
Author information
Authors and Affiliations
Contributions
T.F. and T.K. designed the work and constructed the model. T.K. conducted numerical simulations and experimental and numerical data analyses. T. Handa performed behavioral and electrophysiological experiments and analyzed experimental data. T. Haga performed behavioral and lesion experiments, and R.H. performed behavioral training, surgery and histological staining. T.F, T.K., T. Handa and T. Haga wrote the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing 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 Fig. 1 Relationship between behavioral and neural metrics.
| A crucial hypothesis for the relationship between behavioral metric and neural metric is schematically illustrated. Left, Definition of behavioral sensitivity S. Fitting a tangent function to psychometric curves in rats and models gives behavioral sensitivity, which quantitatively characterizes individual differences. Higher the value of S, the more graded psychometric curve (decision-making is more sensitive to tone). Right, Definition of neural responsiveness χ of the model. Neural responsiveness is the averaged Euclidean distance between the non-perturbed trajectory and neural states generated by random perturbative stimuli applied to a given state of the network (Methods). The larger the value of χ is, the more sensitive to perturbation the neural dynamics are.
Supplementary Fig. 2 Robustness of parameter fitting for sensitivity.
| a, We resampled choices 50 times to generate 50 new psychometric curves (gray lines) for the sensitive network shown in Fig. 2e. The original psychometric curve is shown in orange. b, We fitted the 50 resampled sensitive and insensitive psychometric curves shown in Fig. 2e with a different fitting function P(fn)= b’tan(a’fn) +(1-b’). Here, fn=0 and 1 represent FL and FH, respectively (Methods). The center values and error bars indicate the average and s.d. c, The new sensitivity values S’=a’b’ are plotted for 31 successful learners, which are sorted in the abscissa in an increasing order of the original sensitivity values S. d, The original and new sensitivity values S and S’ are well correlated across all successful learners. In the x-axis and y-axis, error bars show 25 and 75 percentiles of 50 resampled sensitivity values S and S’ of each successful learner, respectively.
Supplementary Fig. 3 Familiar neural population activities projected onto PCs in a rat.
| To reduce the dimensionality, we applied PCA to neural population activities evoked by FH (green) and FL (red) stimuli in left- (solid) and right- (dashed) choice trials (Methods). From the top, neural activities were projected onto PC1, PC2 and PC3. Fig. 3C shows the same neural trajectories in the two-dimensional space spanned by PC2 and PC3. These results were obtained from 379 FH→left, 309 FL→right successful trials, 101 FH→right and 67 FL→left unsuccessful trials.
Supplementary Fig. 4 Reduced neural dynamics of rats and models.
| We approximated neural dynamics with a two-factor (choice and stimuli) regression model for typical three network models (top) and six rats (middle and bottom). We plotted neural trajectories evoked by FH (green) and FL (red). In addition, we plotted trajectories for left (solid) and right (dotted) choices. Gray and colored (green, red) circles indicate stimulus onset and mean reaction time (RT). Out of the 31 successful networks, we selected three such networks that showed sufficiently many (> 10) error trials for both choices to generate robust trajectories. We excluded two rats from the eight rats because these rats did not give sufficiently many neurons due to the shortage of spikes. Shaded areas indicate the period of stimulus application, and black bars above the curves indicate the epochs of significant separation between FH-left and FL-right trajectories according to Cohen’s d (see the subsection "Linear regression analysis" in the Methods). The number of experiments in rat#879, #880, #897, #940, #941 and #949 are: 263, 347, 379, 433, 422 and 347 FH→left and 286, 338, 309, 403, 433 and 333 FL→right successful trials; 50, 61, 67, 0, 6 and 30 FH→right and 24, 50, 101, 24, 43 and 44 FL→left unsuccessful trials. The number of simulations in net #11, #36 and #39 are: 44, 44 and 43 FH → left and 45, 44 and 48 FL → right success trials; 6, 6 and 7 FH → right and 6, 5 and 2 FL → left unsuccess trials.
Supplementary Fig. 5 Neural dynamics without feedback connections.
| Simulations of the model were performed with cutting off feedback connections from the readout units to the reservoirs just after stimulus onset. Neural trajectories mapped onto the choice axis (top) and stimulus axis (middle) are shown in different trial types (FH: green, FL: red, left: solid, right: dotted) with (thin) and without (thick) feedback connections. As in the rats, continuous increases in the separation of neural trajectories are terminated after the removal of familiar cues. The axes were identical as those used in Fig. 3f from neural dynamics in the presence of the feedback. Dynamics of L- (green) and R- (red) readout neurons without feedback indicates that decision-making is possible even without the feedback (bottom). Shaded areas represent the standard deviations of the trajectories over trials. These results were obtained from 44 FH→left, 44 FL→right successful trials and 6 FH→right and 6 FL→left unsuccessful simulation trials in a network model.
Supplementary Fig. 6 Effects of sensitivity on learning process.
| a, Neural responsiveness χ and sensitivity S of various networks that failed to learn association between stimuli and choices (black circles) are shown. Those of successful networks are shown in gray (the same successful networks were previously shown in Fig. 5d). b, c, Learning performance is plotted for the rats (b) and models (c) against sensitivity S. Here, performance is defined as the fraction of correct trials during the entire recording session (that is, a few hundred trials) for each rat or the fraction of correct trials in 100 (50 FH + 50 FL) test trials for each model. d, e, Learning step at which the success rate (the fraction of correct trials during 20 successive learning steps) of each model firstly exceeded 80% is plotted against sensitivity (d) and responsiveness (e). f, The first day of training session on which each rat achieved the criterion percent correct is plotted against sensitivity.
Supplementary Fig. 7 Relationship between reaction times and the sensitivity in psychometric curves.
| a, e, Medians of RT for two familiar tones are plotted against sensitivity in rats (a) and models (e). Error bars show the first and third quartiles of RT distributions. Typical sample sizes are about 100 trials per familiar tone per rat and 50 trials per familiar tone per model network. b, f, Medians of RT for unfamiliar tones are plotted similarly to a and e. Typical sample sizes are about 10 trials per unfamiliar tone per rat and 50 trials per unfamiliar tone per model network. c, g, Differences in the median RTs between familiar (upper) and unfamiliar (lower) tones are plotted similarly to a and e. Pearson’s correlations and p values of two-sided t-test are shown through panels a to g. d, h, RT distributions in a rat and a model, respectively. The results were obtained from about 100 × 2 familiar and 10 × 5 unfamiliar trials for rat (d) and 50 × 2 familiar and 50 × 5 unfamiliar trials for model (h).
Supplementary Fig. 8 Changes in reaction time during training.
| a, b, The histograms of RT are compared between the last day of training session (a) and the first day of recording session (b) for the eight rats. The trial numbers of the eight rats #807, #879, #880, #897, #902, #940, #941 and #949 are (706, 670), (803, 640), (735, 822), (966, 886), (364, 838), (1102, 937), (955, 998) and (798, 795), respectively, where the first and second numbers in each parenthesis refer to trial numbers in the last training session (6429 trials in total) and recording session (6586 trials), respectively. (c), The medians of RT were significantly changed in six rats (filled squares: Mann-Whitney U-test, rat#807, p = 7.8x10−13; #880, p = 2.3x10−34; #897, p = 2.4x10−23; #902, p = 1.9x10−12; #940, p = 5.0x10−5; #941, p = 1.8x10−18), but not in two rats (empty squares: Mann-Whitney U-test, rat#879,p = 0.72; #949,p = 0.34). Error bars show the first and third quartiles.
Supplementary information
Supplementary Information
Supplementary Figures 1–8
Rights and permissions
About this article
Cite this article
Kurikawa, T., Haga, T., Handa, T. et al. Neuronal stability in medial frontal cortex sets individual variability in decision-making. Nat Neurosci 21, 1764–1773 (2018). https://doi.org/10.1038/s41593-018-0263-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41593-018-0263-5