„Infragranular“ layers V and VI have a strategic location in the auditory cortex: they receive thalamic input, feedback from layers II, III and IV and top-down feedback from higher-order areas. As they diffusely project into layers II, III and IV, they can modulate the processing of cortical input in these layers.

Infragranular layers V and VI thus receive information on the afferent cortical input, information on the way how the cortical column processed this input in layers II, III and IV, and finally the information on the resulting activity in higher-order areas. Consequently, these layers „know“ what the input to the cortex is, what cortical processing the input underwent and what the end-result of this processing was. Provided a mis-match between the processing and the resulting activity in higher-order areas, or provided that the higher-order areas are not able to identify what kind of object could have activated the ear, activity in the primary area can be adaptively modulated to result in a „better“ identification process (comp. Hochsten & Ahissar, 2002). This requires a processing in reversed order within cortical areas: first higher-order areas have to identify the object and then earlier areas can process the details of the input. First we see the forest, second we see the trees.

According to the classical theory of bottom-up and top-down processing, layers V and VI can modulate and direct cortical plasticity. This concept is consistent with the Wagner-Rescorla learning theory and most recent data on imaging that demonstrate that cortical areas in fact compute a „prediction error“, i.e. a difference between expected and actual input (review in Kral et al., 2017). This prediction error signal controls learning. In consequence, the „goal-directed“ auditory learning is made possible by these functions of cortical columns and the infragranular layers.

In adult congenitally deaf, however, deep layers V and VI are only weakly activated during sensory stimulation (Kral et al., 2006) and show significant dystrophic changes in both primary and secondary areas (Berger et al., 2017). Our studies have demonstrated an extensive reduction of activity in the deep cortical layers (V and VI) in the primary auditory cortex of congenitally deaf animals (review in Kral et al., 2006). We could quantify the connectivity within the microcircuit and indded document a reduced connectivity between infragranular and supragranular layers in congenital deafness (Yusuf et al., 2022). In consequence, the neuronal processing and plasticity found in the naive („deaf“) auditory cortex cannot be influenced by higher-order areas, it is decoupled from top-down influences. Plasticity cannot be controlled with respect to the goal of this processing. This explains why prelingually deaf subjects, implanted as adults, hear the sounds, but cannot learn to identify the auditory object that caused the sound. As they cannot identify individual phonemes in the acoustic stream, they cannot understand language.

Additionally, top-down effects allow to fill-in the gaps in acoustic streams: if individual sounds of the speech signal are replaced by noise, normal hearing subjects cannot identify this modification (Warren, 1970, phonemic restauration effect). The non-fitting sound is not „heard“ in the acoustic stream - this helps speech recognition in noisy signals. This process is significantly compromised in congenitally deaf subjects. For more details on filling in, see also Kral (2013). 

In conclusion, one essential function of cortical areas is to perform the comparison of the sensory input with the expectation on this input. The difference (error) signal is also driving learning processes. One function of the cortical minicolumn is to perform a computation of this error signal. The present hypothesis is therfore in line with such theories and suggests deficits in predictive coding in naive auditory cortex (Kral et al., 2017).

Furthermore, connectivity analysis between primary and secondary auditory areas supported predicive coding: in acoustically-stimulated hearing cats strong top-down signaling and weak bottom-up signaling was observed, in electrically-stimulated hearing cats (with a previously unheard, unknown electric stimulus) the top-down signaling was weak and the bottom-up signaling much stronger. This is consistent with prediction being represented by top-down interactions, and prediction error by bottom-up interactions. In congenitally deaf animals, the top-down interaction disappeared, supporting our working hypothesis that experience is required for top-down interactions (Yusuf et al., 2021, Yusuf et al., 2022). This effect not only compromises predictive coding, it also affects the appropriate function of the cortical column in primary auditory cortex and disorganizes the cross-frequency coupling (Yusuf et al., 2024).