Limits...
Internally organized mechanisms of the head direction sense.

Peyrache A, Lacroix MM, Petersen PC, Buzsáki G - Nat. Neurosci. (2015)

Bottom Line: The temporal correlation structure of HD neurons was preserved during sleep, characterized by a 60°-wide correlated neuronal firing (activity packet), both within and across these two brain structures.During rapid eye movement sleep, the spontaneous drift of the activity packet was similar to that observed during waking and accelerated tenfold during slow-wave sleep.These findings demonstrate that peripheral inputs impinge on an internally organized network, which provides amplification and enhanced precision of the HD signal.

View Article: PubMed Central - PubMed

Affiliation: The Neuroscience Institute, School of Medicine and Center for Neural Science, New York University, New York, New York, USA.

ABSTRACT
The head-direction (HD) system functions as a compass, with member neurons robustly increasing their firing rates when the animal's head points in a specific direction. HD neurons may be driven by peripheral sensors or, as computational models postulate, internally generated (attractor) mechanisms. We addressed the contributions of stimulus-driven and internally generated activity by recording ensembles of HD neurons in the antero-dorsal thalamic nucleus and the post-subiculum of mice by comparing their activity in various brain states. The temporal correlation structure of HD neurons was preserved during sleep, characterized by a 60°-wide correlated neuronal firing (activity packet), both within and across these two brain structures. During rapid eye movement sleep, the spontaneous drift of the activity packet was similar to that observed during waking and accelerated tenfold during slow-wave sleep. These findings demonstrate that peripheral inputs impinge on an internally organized network, which provides amplification and enhanced precision of the HD signal.

Show MeSH
Noise correlation of HD cells improves linear decodinga: Further analysis of the waking cross-correlograms shown in Fig. 2c. Superimposed purple curves display expected cross-correlation for independent, rate coding cells. b: Difference between actual and expected pairwise cross-correlations (noise correlation35) at 0-timelag as a function of difference between preferred head directions. Note excess at 0° (purple area) and dip at 60° (gray areas). c: Signal correlation (± s.e.m.) for ADn-ADn pairs, ADn-PoS pairs and PoS-PoS pairs for angular difference at 0°±15° (purple) and at 60°±15° (gray). d: Cross-validated median absolute error (± s.e.m.) of HD reconstruction from signal correlations in the ADn (left) and the PoS (right) using three different decoders. Optimal Linear Estimator (OLE) is based on tuning curve signal correlations (signal, green) or actual pairwise correlations (covariance, cov, yellow) and compared with a non-linear Bayesian Decoder (black) as a function of smoothing time window. e: Ratio of OLE absolute median error based on covariance (εOLEcov) or on tuning curves correlations (εOLEsig) taken at their minimal values (across temporal windows) in the ADn and the PoS (p = 0.009).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4376557&req=5

Figure 4: Noise correlation of HD cells improves linear decodinga: Further analysis of the waking cross-correlograms shown in Fig. 2c. Superimposed purple curves display expected cross-correlation for independent, rate coding cells. b: Difference between actual and expected pairwise cross-correlations (noise correlation35) at 0-timelag as a function of difference between preferred head directions. Note excess at 0° (purple area) and dip at 60° (gray areas). c: Signal correlation (± s.e.m.) for ADn-ADn pairs, ADn-PoS pairs and PoS-PoS pairs for angular difference at 0°±15° (purple) and at 60°±15° (gray). d: Cross-validated median absolute error (± s.e.m.) of HD reconstruction from signal correlations in the ADn (left) and the PoS (right) using three different decoders. Optimal Linear Estimator (OLE) is based on tuning curve signal correlations (signal, green) or actual pairwise correlations (covariance, cov, yellow) and compared with a non-linear Bayesian Decoder (black) as a function of smoothing time window. e: Ratio of OLE absolute median error based on covariance (εOLEcov) or on tuning curves correlations (εOLEsig) taken at their minimal values (across temporal windows) in the ADn and the PoS (p = 0.009).

Mentions: In the waking animal, HD cell dynamic can be explained by inputs arriving from the peripheral sensors7,8. Under this hypothesis, HD neurons in ADn and PoS ‘inherit’ their directional information from upstream neurons and respond independently of each other. However, correlations expected from independent rate-coded neurons could not fully account for the observed pairwise correlations of HD cells (Fig. 4a). ‘Noise’ correlations35 were observed for neuron pairs with overlapping fields (0±15°), with a marked negative signal correlation at 60±15° offset (Fig. 4b,c) and the expected correlations became similar to the observed ones only for pairs with larger angular differences. These signal correlations were stronger for ADn than PoS neurons (p < 10–9, Kruskal-Wallis one-way analysis of variance; ADn-ADn pairs: n = 96 and 200, for 0° and 60° signal correlation values respectively; ADn-PoS pairs: n = 77 and 157; PoS-PoS pairs: n = 29 and 57). Importantly, noise correlations could further improve the decoding of the HD signal. Optimal linear estimates (OLE, see Methods) of the head-direction were enhanced when using spike train covariances (i.e., including noise correlation) instead of the tuning curve covariances (i.e., signal correlation only). Estimate of HD based solely on the actual covariance of the spike trains almost reached the performance of a non-linear Bayesian decoder (Fig. 4d). In the ADn, where noise correlations were highest for overlapping HD cells, the covariance-based OLE was approximately 50% better than the signal-based OLE (Fig. 4d,e), and significantly stronger than in the PoS (Fig. 4e; p < 0.05; Mann-Whitney’s U-test; n = 20 ADn and 8 PoS cell ensembles, respectively).


Internally organized mechanisms of the head direction sense.

Peyrache A, Lacroix MM, Petersen PC, Buzsáki G - Nat. Neurosci. (2015)

Noise correlation of HD cells improves linear decodinga: Further analysis of the waking cross-correlograms shown in Fig. 2c. Superimposed purple curves display expected cross-correlation for independent, rate coding cells. b: Difference between actual and expected pairwise cross-correlations (noise correlation35) at 0-timelag as a function of difference between preferred head directions. Note excess at 0° (purple area) and dip at 60° (gray areas). c: Signal correlation (± s.e.m.) for ADn-ADn pairs, ADn-PoS pairs and PoS-PoS pairs for angular difference at 0°±15° (purple) and at 60°±15° (gray). d: Cross-validated median absolute error (± s.e.m.) of HD reconstruction from signal correlations in the ADn (left) and the PoS (right) using three different decoders. Optimal Linear Estimator (OLE) is based on tuning curve signal correlations (signal, green) or actual pairwise correlations (covariance, cov, yellow) and compared with a non-linear Bayesian Decoder (black) as a function of smoothing time window. e: Ratio of OLE absolute median error based on covariance (εOLEcov) or on tuning curves correlations (εOLEsig) taken at their minimal values (across temporal windows) in the ADn and the PoS (p = 0.009).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4376557&req=5

Figure 4: Noise correlation of HD cells improves linear decodinga: Further analysis of the waking cross-correlograms shown in Fig. 2c. Superimposed purple curves display expected cross-correlation for independent, rate coding cells. b: Difference between actual and expected pairwise cross-correlations (noise correlation35) at 0-timelag as a function of difference between preferred head directions. Note excess at 0° (purple area) and dip at 60° (gray areas). c: Signal correlation (± s.e.m.) for ADn-ADn pairs, ADn-PoS pairs and PoS-PoS pairs for angular difference at 0°±15° (purple) and at 60°±15° (gray). d: Cross-validated median absolute error (± s.e.m.) of HD reconstruction from signal correlations in the ADn (left) and the PoS (right) using three different decoders. Optimal Linear Estimator (OLE) is based on tuning curve signal correlations (signal, green) or actual pairwise correlations (covariance, cov, yellow) and compared with a non-linear Bayesian Decoder (black) as a function of smoothing time window. e: Ratio of OLE absolute median error based on covariance (εOLEcov) or on tuning curves correlations (εOLEsig) taken at their minimal values (across temporal windows) in the ADn and the PoS (p = 0.009).
Mentions: In the waking animal, HD cell dynamic can be explained by inputs arriving from the peripheral sensors7,8. Under this hypothesis, HD neurons in ADn and PoS ‘inherit’ their directional information from upstream neurons and respond independently of each other. However, correlations expected from independent rate-coded neurons could not fully account for the observed pairwise correlations of HD cells (Fig. 4a). ‘Noise’ correlations35 were observed for neuron pairs with overlapping fields (0±15°), with a marked negative signal correlation at 60±15° offset (Fig. 4b,c) and the expected correlations became similar to the observed ones only for pairs with larger angular differences. These signal correlations were stronger for ADn than PoS neurons (p < 10–9, Kruskal-Wallis one-way analysis of variance; ADn-ADn pairs: n = 96 and 200, for 0° and 60° signal correlation values respectively; ADn-PoS pairs: n = 77 and 157; PoS-PoS pairs: n = 29 and 57). Importantly, noise correlations could further improve the decoding of the HD signal. Optimal linear estimates (OLE, see Methods) of the head-direction were enhanced when using spike train covariances (i.e., including noise correlation) instead of the tuning curve covariances (i.e., signal correlation only). Estimate of HD based solely on the actual covariance of the spike trains almost reached the performance of a non-linear Bayesian decoder (Fig. 4d). In the ADn, where noise correlations were highest for overlapping HD cells, the covariance-based OLE was approximately 50% better than the signal-based OLE (Fig. 4d,e), and significantly stronger than in the PoS (Fig. 4e; p < 0.05; Mann-Whitney’s U-test; n = 20 ADn and 8 PoS cell ensembles, respectively).

Bottom Line: The temporal correlation structure of HD neurons was preserved during sleep, characterized by a 60°-wide correlated neuronal firing (activity packet), both within and across these two brain structures.During rapid eye movement sleep, the spontaneous drift of the activity packet was similar to that observed during waking and accelerated tenfold during slow-wave sleep.These findings demonstrate that peripheral inputs impinge on an internally organized network, which provides amplification and enhanced precision of the HD signal.

View Article: PubMed Central - PubMed

Affiliation: The Neuroscience Institute, School of Medicine and Center for Neural Science, New York University, New York, New York, USA.

ABSTRACT
The head-direction (HD) system functions as a compass, with member neurons robustly increasing their firing rates when the animal's head points in a specific direction. HD neurons may be driven by peripheral sensors or, as computational models postulate, internally generated (attractor) mechanisms. We addressed the contributions of stimulus-driven and internally generated activity by recording ensembles of HD neurons in the antero-dorsal thalamic nucleus and the post-subiculum of mice by comparing their activity in various brain states. The temporal correlation structure of HD neurons was preserved during sleep, characterized by a 60°-wide correlated neuronal firing (activity packet), both within and across these two brain structures. During rapid eye movement sleep, the spontaneous drift of the activity packet was similar to that observed during waking and accelerated tenfold during slow-wave sleep. These findings demonstrate that peripheral inputs impinge on an internally organized network, which provides amplification and enhanced precision of the HD signal.

Show MeSH