Degeneracy explains diversity in interneuronal regulation of pattern separation in heterogeneous dentate gyrus networks
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background and motivations
The Marr-Albus theory postulates that pattern separation is realized by divergent feedforward excitatory connectivity. Yet, there are several lines of evidence for strong but differential regulation of pattern separation by local circuit connectivity, even when feedforward connectivity is divergent. What are the relative contributions of divergent feedforward connectivity and local circuit interactions to pattern separation? How do we reconcile the contrasting lines of evidence on local circuit regulation of pattern separation in circuits endowed with divergent feedforward connectivity? In this study, we quantitatively address these questions in a population of heterogeneous dentate gyrus (DG) networks, where we enforced feedforward connectivity to be identically divergent.
Methodology
We generated tens of thousands of random spiking neuronal models to arrive at thousands of non-repeating heterogeneous single-neuron models of four different DG neuronal subtypes, each satisfying their respective functional characteristics. We connected these heterogeneous populations of neurons with subtype proportions and local connectivity that reflected the DG microcircuit. In a second level of unbiased search, we generated 20,000 identical networks that differed from each other only in their synaptic weight values. Thus, within the Marr-Albus framework, these networks that were identical in terms of neuronal composition and divergent feedforward connectivity should all be capable of performing effective pattern separation. To test this, we fed these networks with morphed sets of input patterns, recorded granule cell outputs, and computed similarity metrics based on correlation measures across input or output patterns. We developed novel quantitative metrics for pattern separation from plots of output similarity vs. input similarity and validated each network with bounds on these metrics.
Results
Despite being identical in terms of divergent feedforward connectivity, we found only a very small proportion (47 of 20,000 or 0.23%) of the randomly generated networks to manifest effective pattern separation. We tested the specific contributions of interneurons by assessing pattern separation in all pattern-separating networks after individually deleting each of the three interneuron subtypes. Strikingly, we found pronounced network-to-network variability in how each interneuron subtype contributed to granule cell sparsity and pattern separation. We traced this variability to differences in local synaptic connectivity, which also resulted in network-to-network variability in firing rates and sparsity of different interneurons. Finally, we found heterogeneous DG networks to be more resilient to synaptic jitter compared to their homogeneous counterparts, with specific reference to pattern separation computed through average firing rate correlations.
Implications
Our population-of-networks approach clearly shows that divergent connectivity of afferent inputs does not guarantee pattern separation in DG networks. Instead, we demonstrate strong yet variable roles for interneurons and local connectivity in implementing pattern separation. Importantly, our analyses unveil degeneracy in DG circuits, whereby similar pattern separation efficacy was achieved through disparate local-circuit interactions. These observations, alongside network-to-network variability in dependencies on different interneurons, strongly advocate the complex adaptive systems approach as a unifying framework to study DG pattern separation.
