Report

Reversal of Interlaminar Signal Between Sensory and Memory Processing in Monkey Temporal Cortex

See allHide authors and affiliations

Science  18 Mar 2011:
Vol. 331, Issue 6023, pp. 1443-1447
DOI: 10.1126/science.1199967

Abstract

The primate temporal cortex implements visual long-term memory. However, how its interlaminar circuitry executes cognitive computations is poorly understood. Using linear-array multicontact electrodes, we simultaneously recorded unit activities across cortical layers in the perirhinal cortex of macaques performing a pair-association memory task. Cortical layers were estimated on the basis of current source density profiles with histological verifications, and the interlaminar signal flow was determined with cross-correlation analysis between spike trains. During the cue period, canonical “feed-forward” signals flowed from granular to supragranular layers and from supragranular to infragranular layers. During the delay period, however, the signal flow reversed to the “feed-back” direction: from infragranular to supragranular layers. This reversal of signal flow highlights how the temporal cortex differentially recruits its laminar circuits for sensory and mnemonic processing.

The primate inferotemporal cortex locates at the final stage of the ventral visual pathway and serves as a storehouse for visual long-term memory (14). Previous studies have demonstrated neuronal activity related to presented visual objects and retrieved images at the single-neuron level (46), but the underlying network dynamics (712) remain to be understood. Evidence from the primary sensory cortices suggests that local circuits extending across cortical layers are crucially involved in sensory processing (1315). This raises questions about how the interlaminar circuitry in the inferotemporal cortex is differentially recruited to process presented objects and to retrieve visual long-term memory.

We used two strategies to investigate interlaminar signal flow in awake behaving monkeys. First, we used current source density (CSD) analysis as a tool for layer estimation in each electrode penetration; CSD reflects the gross transmembrane currents in the local neuronal ensemble and is used to estimate the cortical layers that receive afferent inputs (16, 17). Second, we used cross-correlation analysis of spike trains (1821) to infer the functional interactions across cortical layers; asymmetry or peak lag of the cross-correlogram (CCG) reflects the direction of functional connectivity between neurons (22, 23).

Two monkeys were trained to perform a pair-association task, in which they had to retrieve the learned paired associate in response to the presented cue stimulus (Fig. 1A) (35). We recorded single- and multi-unit activities and local field potentials (LFPs) by inserting linear-array multicontact electrodes (16 or 24 contacts with spacing of 150 or 100 μm, respectively) vertically (table S1) (24) into area 36 (A36) (Fig. 1B). CSD was then calculated from depth profiles of stimulus-evoked LFPs in order to physiologically estimate the position of the granular layer (24). A representative CSD profile exhibited the earliest current sink (Fig. 1, D and E, asterisks) at the contact corresponding to the histologically verified granular layer (Fig. 1C, red). This earliest current sink was followed by sinks at more superficial contacts and by sources at deeper contacts (Fig. 1D). Similar CSD profiles were consistently observed for all penetrations (fig. S1). Postmortem histological analyses (24) confirmed that the earliest current sink evoked by cue stimuli consistently corresponded to the granular layer [table S1, the distance between the contact with the earliest current sink (“earliest-sink contact”), and the center of the granular layer was 79 μm (median), n = 6 penetrations]. The histological verifications, together with consistent CSD profiles across penetrations, demonstrated that the CSD profiles can be reliably used to estimate the granular layer (G), the supragranular layer (SG), and the infragranular layer (IG) (24). In this representative penetration, single unit activities were simultaneously recorded in SG and IG (Fig. 1, F and G), both showing stimulus-selective responses during the cue period (Fig. 1H). The CCG (1823) for this unit pair exhibited a significant displaced peak (4 ms lag) on the right side (Fig. 1I), suggesting a functional connectivity from the SG unit to the IG unit, which is consistent with the “feed-forward” signal flow in the primary sensory cortices (1315).

Fig. 1

(A) Sequence of pair-association task. Monkeys had to retrieve the learned paired associate in response to the presented cue stimulus. (B) Lateral (left) and ventral (right) views of monkey brain. Unit activities and LFPs were recorded across cortical layers in area 36 (blue) by using a linear-array multicontact electrode. Scale bar, 10 mm; amts, anterior middle temporal sulcus. (C to I) A representative data set. (C) Electrolytic lesion marks made at two contacts of the electrode (yellow contacts) were identified in a Nissl-stained histological section. Scale bar, 200 μm. [(D) and (E)] Stimulus-evoked CSDs. The earliest significant current sink appeared at 91 ms after cue onset (asterisks), (red contact). The red contact corresponded to the granular layer in histological section (C). Red and green bars in (E) indicate significant and nonsignificant current sink, respectively. G, SG, and IG represent granular, supragranular, and infragranular layers, respectively. (F) Waveforms, (G) auto-correlograms, and (H) poststimulus time histograms of two single units simultaneously recorded in SG and IG [(F), blue contacts]. (I) Raw CCG between spike trains of the SG and IG units in the cue period (black line). Gray line, IFR-predictor (24). Bin width, 1 ms. (Inset) IFR-predictor–subtracted CCG. The CCG exhibited a significant peak on the right side (lag time, 4 ms). Horizontal gray line indicates the confidence limit.

We made 20 penetrations in three hemispheres of two monkeys and conducted cross-correlation analyses for three populations of unit pairs: G-SG pairs (cue period, n = 52 pairs; delay-period, n = 49 pairs ), G-IG pairs (n = 128 pairs; n = 121 pairs), and SG-IG pairs (n = 252 pairs; n = 211 pairs) [both single units and multi-units were included; for details, see supporting online material (SOM) text and table S2]. A CCG was calculated only when both constituent units responded to at least one common stimulus during either the cue or delay period. CCG peak was detected within 10 ms lag (1922) so as to evaluate its significance (Z > 2.82, P < 0.05) (24). We then compared the proportions of unit pairs with significant CCG peaks among the G-SG, G-IG, and SG-IG pairs (fig. S2). The proportion of unit pairs with a significant CCG peak was greater for G-SG pairs than for G-IG pairs during both the cue period (33% versus 11%; χ2 test with post-hoc pair-wise comparisons followed by Bonferroni’s correction, P < 0.005) and delay period (27% versus 12%; P < 0.05). The proportion of unit pairs with a significant CCG peak was greater for G-SG pairs than for SG-IG pairs during the cue period (33% versus 16%; P < 0.05), and the same tendency was observed during the delay period (27% versus 19%; P = 0.24).

We next examined the direction of functional connectivity between units in different layers during each task period. For G-SG pairs, the distribution of asymmetry index (AI) (2224) of individual CCGs during the cue period was shifted to the feed-forward direction: from G to SG (Fig. 2B, blue) [Wilcoxon signed-rank test; cue period (Fig. 2B, blue), P < 0.01, n = 17 pairs]. This directional bias was not significant during the delay period (Fig. 2B, red) (P > 0.4, n = 14 pairs). Similar results were obtained using the center of mass (CoM) of the CCG peak (Fig. 2C) [cue period (Fig. 2C, blue), P < 0.04; delay period (Fig. 2C, red), P > 0.5]. During the fix period, only four pairs exhibited a significant CCG peak, and thus the directional bias was not statistically evaluated. These results were further substantiated by the population-averaged CCGs (Fig. 2A and fig. S4A): The CCG showed a prominent peak [P < 0.001, (24)] on the right side during the cue period (G to SG) (Fig. 2A, blue). In G-IG pairs, no bias was observed in their signal flow directions during any of the task periods (fig. S3).

Fig. 2

Population results of the functional connectivity. (A to C) Cross-correlation between spike trains of G-SG pairs. (A) Population-averaged CCGs in the (gray, left) fix, (blue, middle) cue, and (red, right) delay periods, respectively. (B) Asymmetry index and (C) center of mass of individual CCGs. Asterisk indicates significant bias to either side of the histogram. Filled histogram indicates a task period for which significant bias in the directionality was observed. (D) Schematic diagram of interlaminar signal flow between G and SG. (E to H) Same as in (A) to (D), but for SG-IG pairs.

We repeated the same analyses for SG-IG pairs (Fig. 2, E to G). The distribution of AIs during the cue period was significantly shifted toward the direction from SG to IG (Fig. 2F, blue) (P < 0.01, n = 41 pairs). However, the distribution during the delay period exhibited a bias in the opposite direction, IG toward SG (Fig. 2F, red) (P < 0.01, n = 41 pairs). No significant directional bias was observed during the fix period (P > 0.2, n =12 pairs). Similar results were obtained using the CoM [cue period (Fig. 2G, blue), P < 0.03; delay-period (Fig. 2G, red), P < 0.02; fix-period, P > 0.8] (24). Population-averaged CCGs again confirmed these results (Fig. 2E and fig. S4B): During the cue period, a significant peak was observed on the right side (P < 0.001; SG to IG, Fig. 2E, blue), whereas a significant peak appeared on the left side during the delay period (P < 0.001; IG to SG, Fig. 2E, red). Consistent results were obtained using only single unit data (SOM text and fig. S5).

These results demonstrated the signal flow from G to SG and from SG to IG during the cue period, as in the canonical feed-forward processing (Fig. 2, D and H, left) (1315). During the delay period, however, the direction of signal flow reversed, suggesting recruitment of a “feed-back” pathway (Fig. 2H, right).

We then examined the temporal dynamics of the functional connectivity for individual SG-IG pairs. Figure 3A shows the time course of the correlation strength (CS) and asymmetry index (AI) for each pair (n = 70 pairs) that exhibited a significant peak during either the cue or delay period (24). Although the connectivity of individual pairs exhibited a variety of dynamics, as a whole the direction of connectivity gradually changed: SG to IG (Fig. 3A, blue) during the cue period and IG to SG (Fig. 3A, orange) during the delay period (Fig. 3B) (for the temporal dynamics of firing rates, see SOM text and figs. S6 and S7). To investigate these observations quantitatively, we divided SG-IG pairs according to the sign of the AI in the cue and delay periods. Nearly half of the pairs (47%, 33 of 70) exhibited directional changes in the connectivity between the cue and delay periods (“flipped pairs”) (Fig. 3C), suggesting that the signal flow direction can be modulated in individual pairs. Of these, a significantly greater proportion (73%, 24 of 33 pairs; χ2 test, P < 0.01) exhibited connectivity from SG to IG during the cue period and reversed their direction during the delay period. Furthermore, unit pairs that did not change the sign of AI between the cue and delay periods (“non-flipped pairs”; 53%, 37 of 70 pairs) also contributed to the overall changes in the signal flow (Fig. 3D): For unit pairs with connectivity direction from SG to IG (Fig. 3D, blue), AIs in the delay period were closer to zero than those in the cue period (Wilcoxon signed-rank test, P < 0.05, n = 17 pairs), and for unit pairs with connectivity direction from IG to SG (Fig. 3D, orange), AIs in the delay period were more negative than those in the cue period (P < 0.01, n = 20 pairs). Together, the reversal of connectivity direction between the cue and delay periods (Fig. 2) was the result of both the directional changes of the flipped pairs and consistent small directional shifts of the non-flipped pairs.

Fig. 3

Connectivity dynamics of individual SG-IG pairs. (A) Time course of spike correlation for individual unit pairs. AI and CS of CCGs were color-coded as shown in the inset. Unit pairs were sorted according to AI value during the latter half of the delay period. (B) Population average of all the unit pairs. (C and D) Polar plot of CS and AI dynamics for the (C) flipped- and (D) non-flipped pairs. Radius, CS. Angle from the vertical axis, AI. Positions of base and tip of an arrow correspond to AI/CS values during cue and delay periods, respectively. [(C), right] Proportion of each type of flipped pairs. S and I represent SG and IG, respectively. [(D), right] AI of non-flipped pairs during the cue and delay periods. Blue, SG→IG pairs; orange, IG→SG pairs.

Lastly, we examined the spatial patterns of functional connectivity by calculating the laminar positions of units by parametrically using the distances from the estimated granular layer (Fig. 4 and fig. S8) (25, 26). Compared with the connectivity during the fix period, two prominent connectivity patterns appeared during the cue period, corresponding to the feed-forward pathways from G to SG and from SG to IG (Fig. 4, A and B, middle). During the delay period, the feed-forward connectivity was attenuated, and the feed-back connectivity from IG to SG emerged (Fig. 4, A and B, right). In addition, outward signal flow (from superficial to deep parts) within IG was found during the delay period (Fig. 4, A and B, right). Putative target units of this outward flow were located at significantly deeper positions than those of the putative source units (fig. S9) [paired t test, P < 0.02, n = 19 pairs; median distances from the granular layer were 0.45 mm (source units) and 1.05 mm (target units)].

Fig. 4

Interlaminar connectivity matrices. (A) AI matrix for each task period. Abscissa and ordinate represent recorded positions of the putative source and target units relative to the earliest-sink contact, respectively. Size of a circle in each matrix indicates the proportion of unit pairs with significant CCG peak to the total number of unit pairs for which CCGs were calculated at the corresponding site. Saturation of color of each circle indicates the average of AI across unit pairs. (B) CS matrix, as in (A). All laminar positions plotted in the AI and CS matrices were recorded in at least three penetrations. (C) Summary diagrams showing all the laminar signal flows identified in the present study.

The present study demonstrated that canonical feed-forward signal flow across cortical layers during sensory coding reverse to the feed-back direction during memory retrieval phase, which suggests flexible recruitment of interlaminar connectivity depending on the cognitive demands in the monkey association cortices (Fig. 4C). We used CSD analysis to estimate cortical layers (Fig. 1, C to E, and fig. S1), and the observed stimulus-evoked CSD profiles were quite similar to those in the primary sensory cortices (17, 27). For some penetrations, we observed that the current sink positioned superficially next to the earliest-sink contact exhibited larger peak amplitudes and much longer durations than that of the earliest current sink. This observation might reflect the cytoarchitectural nature of A36 as a dysgranular cortex (28) as well as the direct inputs to the deepest part of the superficial layer, which is consistent with anatomical observations (29).

A recent study in the rat primary auditory cortex demonstrated that the direction of interlaminar signal flow depends on the cortical “state”: Sensory-evoked responses were initiated in the thalamorecipient layers and then propagated to the superficial and deep layers, whereas in spontaneously active “up-states,” neuronal activity was initiated in the deep layers and then propagated to the superficial layers (27). These state-dependent changes in the interlaminar signal flows in rats are consistent with our results obtained in monkeys performing a memory task. Together, these findings highlight the flexibility of cortical laminar circuits. Further experiments will be needed to determine whether such flexible interlaminar connectivity is also implemented and used in other cortical areas for other cognitive demands.

Supporting Online Material

www.sciencemag.org/cgi/content/full/331/6023/1443/DC1

Materials and Methods

SOM Text

Figs. S1 to S9

Tables S1 and S2

References

References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. This work was supported by a Grant-in-Aid for Specially Promoted Research from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) to Y.M. (19002010) and Grant-in-Aid for Young Scientists from MEXT to T.H. (18700378); a grant from Takeda Science Foundation to Y.M.; and Japan Society for the Promotion of Science (JSPS) Research Fellowships for Young Scientists to D.T. (1811234) and K.T. (211438). This work was also supported in part by Global Center of Excellence Program from MEXT. We thank M. Takeda for discussions and comments and H. Kasahara, R. Fujimichi, T. Matsui, and K. W. Koyano for advice and help with experiments.
View Abstract

Stay Connected to Science

Navigate This Article