In this dissertation, we described several distinct experiments that explore how
neuronal activity the primary auditory cortex adapts to variations in sound stimulus
contexts. In Chapter 2, we tested how stimulus temporal correlation modulates the
input-response profile of neurons in the primary auditory cortex, and our linear non-
linear modeling revealed that they adaptively compensate for this particular stimulus
feature through gain-control. In Chapter 3, we used optogenetic methods to test how
interneuron populations in the primary auditory cortex modulate responses to standard
versus deviant tones, showing that inhibition from SOMs suppresses responses
specifically to repeated tones while inhibition from PVs simply amplifies the differences
in response magnitude. Further we found that SOMs and PVs themselves exhibit
adaptation at similar levels as local pyramidal neurons. These results were incorporated
into a rate model of cortical dynamics, which found that SOM interneurons play a
special role in SSA. In Chapter 4, we again used optogenetic methods to more closely
examine how PVs and SOMs modulate responses before and after adaptation across
frequency tuning curves. We found that inhibition from SOMs increased, modulating
frequency tuning in a gain-like manner, as adaptation developed and was also
correlated with the magnitude of adaptation. By contrast, inhibition from PVs was
results deliver insights to the extent, mechanisms and functions of cortical sensory
adaptation.
Though the approaches are quite different, these studies share a common goal
of exploring how the auditory cortex adapts to persistent temporal regularities in sound
stimuli. Natural sounds contain a mix of spectrally and temporally correlated features
with parametrically constrained statistical variations. In order to inform how the
auditory cortex encodes natural sounds, we chose to focus separately on temporal
aspects of stimulus dynamics rather than spectro-temporal correlations that could
confound interpretation of the results. In Chapter 2, our custom designed dynamic
random cord stimuli exhibited a range in temporal correlation independent from
variation in spectral modulation. In Chapters 3 and 4, oddball sequences and single
frequency tone trains explore regularity across narrow frequency bands. The adaptive
phenomena tested in each experiment may operate at overlapping or different
timescales. Accordingly, we tested temporal amplitude modulations ranging from 200
Hz to 12.5 Hz among the dynamic random cord stimuli and only 2.5 Hz for tone trains.
We found that the time constant of adaptation ranged widely across neurons from 100
ms up to 7 s for changes in temporal correlation, and similarly from 100 ms to greater
than 3 seconds for tone train stimuli. We found that across these difference dynamic
ranges, adaptation plays an important role in controlling cortical responses. Future
studies can address the downstream mechanisms for inferring temporal dynamics in the
Our findings indicate that PVs and SOMs may both play a role in adaptation to
broadband stimuli; PVs by non-selectively amplifying stimulus-response differences and
SOMs by inducing gain-like adaptation across the spectrum. Interestingly, the
interneurons themselves may operate on different time scales of adaptation. Through
direct thalamocortical input to PVs, inhibitory feedforward circuit dynamics could
mediate adaptive responses more quickly. For instance, PVs have been shown to
modulate the stimulus integration time window by around 10 ms (Gabernet et al.,
2005), and could produce fast acting adaptation within very short time scales. Feedback
and top-down circuits, which SOMs may participate in, would generate an adaptive
signal more slowly. In Chapter 3, the tone-by-tone increase in SOM mediated inhibition
indicates that SOMs may operate on longer timescales. Perhaps the difference in
adaptation time scales among pyramidal neurons is due to differential weighting of PV
and SOM mediated inhibitory input. These findings may help to constrain future models
and understanding of temporally dynamic aspects of auditory cortical responses to
natural sounds.
Our findings likely generalize to other sensory modalities and cortical areas.
Beyond auditory cortex, studies have described adaptation to stimulus contrast in visual
(Lesica et al., 2007) and olfactory cortex (Kadohisa and Wilson, 2006). Our experiments
showed that gain-control underlying contrast adaptation similarly underlies adaptation
to temporal correlation. By extension, cortical receptive fields in vision,
somatosensation and olfaction may also exhibit sensitivity to temporal correlations in
variations over their receptive fields. Though there are variations, interneurons in
primary sensory cortices largely share common circuit architecture (Douglas and Martin,
2004). Thus, the roles of PV and SOM inhibition in adaptation is likely analogous across
modalities, even if the statistics of temporal modulations in natural auditory scenes
differs from those of natural stimuli in other modalities. In fact, a recent study confirms
that SOMs modulate cortical responses to standard and deviant visual stimuli
differentially (Hamm and Yuste, 2016). Thus, we expect that our experimental findings
uncover general principles and mechanisms of cortical sensory adaptation.
The adaptive properties we describe may also underlie complex adaptive
phenomena beyond temporal context sensitivity in sensory cortex. Divisive
normalization, considered a canonical cortical computation, describes a spatial context-
specific adaptation (Carandini and Heeger, 2011); a sensory neuron’s spiking response
to stimuli falling within a central excitatory receptive field are normalized by the mean
intensity of the stimuli falling within a broader suppressive surround subfield.
Importantly, the center and surround subfields exhibit independent sensitivity to the
temporal regularities of the stimulus (Durand et al., 2007). The sensitivity to temporal
correlations described in Chapter 2 may constrain models of divisive normalization in
this process; a receptive field’s smaller excitatory and larger suppressive subfields likely
each integrate rapid stimulus variation over proportional timescales. Thus it would be
interesting to test how gain-control acts independently between these subfield, and
controls their combined output. Such computation structure adds a spatial component