Subjects, surgery, and electrode implantation
Pigeons weighing 400–500 g were obtained from a local supplier (Gongchuang Pigeon Co., Henan, China) and housed in an animal facility with the size of 3 m × 3 m × 2 m for at least 2 weeks before the experiments, which is with plenty of sunlight, good ventilation, and free access to water and food. After completion of all the behavioral experiments and recordings, each animal was humanely euthanized for further histologically examination of the recording locations. Each animal was given a lethal overdose of anesthetic (1.5% pelltobarbitalum natricum) and the tips of recording electrodes in Hp were marked by electrolytic lesions (5 mA for 20 s). Then, each was perfused transcardially with saline followed by 4% paraformaldehyde (prepared in 0.1 M phosphate buffered saline, pH 7.4) to get the brain according to the animal welfare regulations. Finally, frozen coronal sections (50 μm) were collected from the recording sites to be histologically examined. Every effort was made to minimize animal pain, suffering and distress and to reduce the number of animals used. No unexpected mortality or adverse events were observed. All of the experiments were conducted by individuals the Animals Act, 2006 (China), for the care and use of laboratory animals, and approved by the Life Science Ethical Review Committee of Zhengzhou University.
All surgeries were performed after the pigeons were anesthetized with 1.5% pelltobarbitalum natricum (0.25 ml/100 g) injected intraperitoneal. The pigeon was placed in a stereotaxic apparatus and the recording microelectrode array (16 channels: 4 × 4 array, Hong Kong Plexon Inc., Hong Kong, China) was chronically implanted at a location directly in the left Hp (AP 4.5 mm; ML 1.0 mm; DV 0.5–1.5 mm) according to coordinates obtained from the Karten and Hodos stereotaxic atlas of the pigeon brain [52]. The implanting location of Hp, the microelectrode array, and a pigeon with the implanted electrode are shown in Fig. 1.
Goal-directed spatial task and apparatus
After the recovery period of about one week, the pigeons implanted with electrode arrays were taken to carry out the experiment and signal recording in the daytime every day, in which the order of experiments, measurements and caging of animal was random to minimize confounders. In our experiments, the pigeons were trained to carry out a goal-directed spatial cognitive task in a maze (Fig. 2a and b). The maze included a starting position as the waiting area, two alternative goal positions with food. At these above three positions, there were three food hampers providing food rewards. There are multiple optional paths between the starting position and the goal positions. The infrared detectors distributed on all of the pathlets along the paths were used to define the beginning and end time for signal segmentation. The gate was set at the exit of the starting position to control the beginning and end of a trial. All of the designs are used for the simulations of the processes including spatial learning and path adjustment.
In the goal-directed spatial cognitive task, the pigeons were trained to walk from the starting position to the goal location and get food rewards. It is noted that for each pigeon, one of the two alternative goals is chosen to designate as the goal location randomly, in which the randomisation lists are computer generated. Only the investigator is aware of the participant’s goal allocation. At the beginning of the trial, the hamper was opened to provide food in the goal location, and the pigeons were trained to learn a preferred path to the goal. If the pigeon arrived at the goal, this trial was recorded as a correct one. After enjoying the food reward, they were trained to go back to the starting position to start the next trial. The experiment procedures in one session could be divided into three phases (Fig. 1 c and d), while each phase was composed of multiple trials. When a pigeon could reliably perform the trial through a preferred path, in which more than 90% of the total trial numbers are corresponding to the same path on two consecutive days, it was considered that the experiments of Phase 1 (acquisition) were completed. Then the experiments entered Phase 2 (adjustment), in which the learned path was blocked at the pathlet close to the goal. In this phase, the pigeon needs to finish the path adjustment by exploring the maze to find a new path to the goal. Normally, the pigeons could learn a new path after multiple trials of training. Finally, at the end of the experiment is Phase 3 (recovery). The pigeons adapted to the new environment after the barricading, and they learned a new path to get the rewards at the goal position.
Behavioural data and LFPs recording
The behavioural data of the pigeons was recorded by the observation camera placed on the ceiling and stored in the computer during the experiment. The behavioural trajectories and timings of the pigeons were analyzed to obtain the time the pigeons spent and the path length the pigeons walked from the starting position to the goal.
A 128 channel Cerebus™ Multichannel Acquisition Processor (Blackrock Microsystems, Salt Lake City, UT, USA) was used to record local field potential (LFP) signals from the Hp region of the pigeon. The LFPs with a sampling rate of 2 kHz were filtered by a 0–250 Hz Butterworth low-pass filter.
Signals of interest segmentation
In our experiments, we recorded the LFPs in the Hp of the pigeons when they performed the goal-directed spatial cognitive task. To eliminate the negative influence of bad channels caused by detached electrode contacts, intermittent electrical connection, or line noise, we selected the channels with high reliability from all 16 ones of different pigeons. Then the signals of interest (SOIs) corresponding to different states from the three phases of the experiments were segmented. Finally, for each SOI segment, we filtered the LFPs and obtained the signals corresponding to the following six bands, including delta (1–4 Hz), theta (5–12 Hz), beta (13–30 Hz), slow-gamma (31–45 Hz), and fast-gamma (55–80 Hz).
Functional network analysis
The brain can be considered as an extremely complex network and functional connectivity is a powerful tool used to characterize the brain networks in the local brain region. To obtain the mathematical representation of the brain, the nodes and edges are defined based on complex networks theory in brain functional network analysis. Then, the correlation intensity between the network nodes can be measured by the incident matrix. As a commonly used criterion to construct an incident matrix, coherence [53] can be used not only to analyze the degree of the synchronization but also to represent the linear or interdependent relationship of the variables in the frequency domain, which is defined as the normalized result of two independent signals. In this paper, we used coherence to measure the relationship between the LFPs corresponding to the channels in Hp, and the coherence coefficient is calculated as follows:
$${Coh}_{x,y}(f)=\frac{{\left|{p}_{x,y}(f)\right|}^2}{\left|{p}_x(f)\right|\times \left|{p}_y(f)\right|}$$
(1)
where
$${p}_{x,y}(f)=\frac{1}{n}\sum \limits_{i=1}^n{x}_i(f){y}_i^{\ast }(f)$$
(2)
For a given frequency f, px(f) and py(f) represent the auto-power spectrums of two LFP time series x and y respectively, and px, y(f) is the cross-power spectrum. In this paper, we calculated the LFPs coherence of each pair of channels from all channels in Hp. The coherence matrices of different frequency bands were obtained to construct the functional network. In our work, the channels in the Hp can be defined as the nodes of the network and the connection between any two nodes can be defined as the edge of the network. We can binarize the above coherence matrices based on the appropriate thresholds to realize the visualization of the network connections [54].
The topological characteristics of the brain functional network can be used to reflect the functional connectivity of the brain and help to reveal the cognitive state in the brain. In this paper, we selected two representative characteristics, clustering coefficient (Coef) and global efficiency (Eff) for statistical analysis of the networks [55]. The clustering coefficient reflects the intensity of the connection between different nodes (channels in Hp) in the functional network. The calculation formula is as follows:
$$\mathrm{Coef}=\frac{1}{M}\sum \limits_{i=1}^M\frac{2{E}_i}{k_i\left({k}_i-1\right)}$$
(3)
where ki indicates that the i-th node has k edges connected with the other nodes. Ei indicates the number of edges in the network that connects with the i-th node. M is the total number of nodes in the network. The value of Coef ranges from 0 to 1. The larger the value is, the closer the information between the two nodes in the network will be.
Global efficiency can be used not only to measure the efficiency and ability of the information transmission and processing in the brain functional network but also to reflect the degree of network integration. It is defined as:
$$\mathrm{Eff}=\frac{1}{M}\sum \limits_{i\in M}\frac{\sum_{j\in M}{\left({d}_{ij}\right)}^{-1}}{M-1}$$
(4)
where dij indicates the shortest path length between two nodes i and j.
Statistical analysis
All statistical analyses were performed by Matlab R2014a software (The MathWorks, Inc., Natick, MA, USA), using the rank-sum test. Statistical results were presented as mean ± standard deviation (std), and the statistically significant difference level was set to 5%. Statistically significant were indicated by p value as follows: * p < 0.05, ** p < 0.01, *** p < 0.001.
