, 2010a, 2010b; Lisman et al., 1998;
Wong and Wang, 2006). Thus, in our results, the emergence of the highly cue-tuned neurons with a low basal activity may also result from plastic changes of local recurrent circuits after learning. Ablation of the activated area before and after the training demonstrated that it is required specifically for the long-term storage of the memory of reinforcement learning. Concomitantly, we observed calcium activity only during recall of the consolidated long-term memory. These results suggest that a region in fish telencephalon homologous to mammalian cortex can consolidate and retrieve a long-term memory. Although the transfer of a memory has been reported in rodents (Frankland et al., 2004; Maviel et al., 2004), the current study represents the first report of the visualization and physiological analysis of long-term MK-1775 molecular weight consolidated memory in vivo. The telencephalic activity that we observed may represent a neural program for cue association, cue contingency, and avoidance behavior established by learning, rather than a simple find more motor command for swimming. Several lines of evidences support this idea. First, we did not observe the activity 30 min after training
when the fish had already learned the avoidance program. Second, and more importantly, we observed the calcium signals even in the stay memory retrieval acquired by two-color conditioned learning or by a change in the behavioral rule from the avoidance to stay task. The telencephalic activity should disappear in this context if it simply encoded a motor output command. The activated areas for memory retrieval in the stay task were broader than that in the
avoidance task, suggesting the engagement of a subset of neurons that were required specifically for the learning contingency for stay. One possible explanation aminophylline for this broader activity pattern in the stay task would be the requirement of the activity of an additional telencephalic region that suppresses the activity of the avoidance ensemble to accomplish the stay behavior in fish. In the rodent cued fear conditioning paradigm, the infralimbic cortex is required to suppress the expression of the learned fear, i.e., freezing (Sotres-Bayon and Quirk, 2010). We showed that a change in the learning contingency from the avoidance to the stay task induced a rapid change in activation patterns. However, the telencephalic signals for the stay task after the avoidance task faded by 24 hr. This is in distinctive contrast with the case when the fish first learned the avoidance task, in which the signal was detected only 24 hr later.
The average training duration of participants here was 73 hr, with up to 10 hr devoted to learning to read using the SSD. As part of the training program, the participants were taught (using verbal explanations and palpable images; see Figure 1D
and Supplemental Experimental Procedures) how to process 2D still (static) images, including hundreds of images of seven structured categories: geometric shapes, Hebrew letters and digital numbers, body postures, everyday objects, textures (sometimes with geometric shapes placed over visual texture, used to teach object-background segregation), Selleck Erastin faces, and houses (see Figure 1E; see Movie S1 for a demo of the visual stimuli and their soundscape representations). For full details on the training technique and protocol, see the Supplemental Experimental Procedures.
After the structured training, participants could tell upon hearing a soundscape which category it represented. This required Gestalt object perception and generalization of the category principles and shape perception to novel stimuli. They could also determine multiple features of the stimulus, enabling them to differentiate between objects within categories. For an example, see Movie S2, depicting one congenitally blind Microtubule Associated inhibitor participant reading a three-letter word and another participant recognizing emotional facial expressions. In order to assess the efficiency of training in terms of visual recognition, six of the participants in the training protocol underwent a psychophysical evaluation of their through ability to identify different object categories. They were required to categorize 35 visual images (in pseudorandomized order) as belonging to the seven object categories.
Each stimulus was displayed using headphones for four repetitions (totaling 8 s), followed by a verbal response. The average rate of object classification success in the blind was 78.1% (±8.8% SD), significantly better than chance (14%; see Figure 1F, t test p < 0.00005). Letter category recognition did not differ from that of the other object categories (all p > 0.05, corrected for multiple comparisons). In order to minimize sensory-motor artifacts, no recording of performance was conducted during the fMRI scan. Prior to each scan, we verified that the subjects were able to easily recognize learned stimuli from the tested categories (see more detail in Supplemental Experimental Procedures). The main study included six experimental conditions presented in a block design paradigm. Each condition included ten novel soundscapes representing unfamiliar images from the trained object categories: letters, faces, houses, body shapes, everyday objects, and textures. Each condition was repeated five times, in a pseudorandom order. In each epoch, three different stimuli of the same category were displayed, each for 4 s (two repetitions of a 2 s stimulus). For instance, in each letter epoch, the subject was presented with a novel meaningless three-consonant letter string.
Fukushima et al. (2012) are inclined toward the position that the signal arises primarily
from neuronal spiking in the superficial layers of auditory cortex, based on a proximity argument and on a prior study in rodent auditory cortex. This seems to us to be unlikely, given that in the auditory cortices Sorafenib clinical trial of the awake monkey, the massive weight of both stimulus-evoked and spontaneous firing is in the granular layers compared to the relatively sparse firing seen in the more superficial layers (see e.g., Kajikawa and Schroeder, 2011). Assuming, as the authors do, that high-gamma power is related to multiunit firing, high gamma generated by high-volume firing in the middle layers is likely to overwhelm any generated by the much more sparse firing in supragranular sites. Fukushima et al. (2012) raise a number of logical possibilities regarding underlying causes of structure in ongoing auditory cortical activity, based on a detailed consideration of the relevant anatomical connectivity
patterns between core and higher-order IPI 145 cortices and between auditory core and thalamic regions. They also discuss a provocative idea that ongoing activity in auditory cortex represents a playback of recently experienced stimulation. Continuing down this path to longer time scales, it is noteworthy that the dynamical structure of spontaneous activity across the spectrum in auditory cortex bears a remarkable, and likely noncoincidental, resemblance to the 1/f statistics of the natural auditory environment (Garcia-Lazaro et al., 2006). This fits with the idea that the
blueprint for macaque auditory cortex evolved under the pressures of this natural environment and that in ontogeny, individuals’ auditory Resminostat cortices further tune to the statistics of that same environment (Berkes et al., 2011). It will be interesting to investigate these relationships further and to see how nature and nurture collaborate in this arena. Needless to say, the causes of “spontaneous order” in auditory as well as other cortices are a prime area for future research, as currently there are many more questions than answers. For example, the authors note work by Raichle and colleagues on so-called “resting state” fMRI as evidence that the brain is constantly active, a line of work that has virtually exploded as a means of mapping large-scale brain functional connectivity networks using graph theoretic analyses (Bullmore and Sporns, 2009). To connect the dots, it is interesting to note that this approach is in principle applicable at smaller scales such as those dealt with here, which would in effect represent subsets or nodes in a larger network. This in turn underscores the point (see also below) that it will be important to relate high-gamma to lower-frequency dynamics, extending down to the infraslow ranges that approximate the time frame of hemodynamic oscillations.