Top-down and bottom-up interactions for cortical bursting by Vincent D Tang Submitted to the Department of Brain and Cognitive Sciences in partial fulfillment of the requirements for the degree of DOCTOR OF BRAIN AND COGNITIVE SCIENCE at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY May 2025 © 2025 Vincent D Tang. All rights reserved. The author hereby grants to MIT a nonexclusive, worldwide, irrevocable, royalty-free license to exercise any and all rights under copyright, including to reproduce, preserve, distribute and publicly display copies of the thesis, or release the thesis under an open-access license. Authored by: Vincent D Tang Department of Brain and Cognitive Sciences May 2, 2025 Certified by: Mark T. Harnett Professor of Brain and Cognitive Sciences, Thesis Supervisor Accepted by: Mark T. Harnett Professor of Brain and Cognitive Sciences Graduate Officer, Department of Brain and Cognitive Sciences 2 THESIS COMMITTEE Thesis Supervisor Mark T. Harnett Professor of Brain and Cognitive Sciences Department of Brain and Cognitive Sciences Thesis Readers Mark Bear Picower Professor of Neuroscience Department of Brain and Cognitive Sciences James J. DiCarlo Peter de Florez Professor Department of Brain and Cognitive Sciences Richard Naud Professor of Cellular and Molecular Medicine Department of Cellular and Molecular Medicine 3 4 Top-down and bottom-up interactions for cortical bursting by Vincent D Tang Submitted to the Department of Brain and Cognitive Sciences on May 2, 2025 in partial fulfillment of the requirements for the degree of DOCTOR OF BRAIN AND COGNITIVE SCIENCE ABSTRACT High-frequency burst firing occurs throughout the mammalian cortex in vivo, yet both the underlying mechanisms and functional roles of bursts are unclear. Burst firing in brain slices is strongly modulated by the activity of apical dendrites, which branch extensively in layer 1 (L1) and receive long- range inputs from higher-order cortical and thalamic areas. These properties suggest a powerful subcellular substrate by which single pyramidal neurons could multiplex bottom-up and top-down information via L1-independent tonic spikes and L1-dependent bursts, respectively, and have provided a basis for emerging theoretical models of cortical computation and learning. However, our understanding of burst firing and subcellular processing remains critically limited by a lack of evidence in awake animals. It is unclear whether burst firing a) is preferentially recruited by bottom-up versus top-down inputs, and b) requires apical dendritic engagement. To answer these questions, we performed high-density extracellular recordings in primary visual cortex of awake mice while presenting a battery of Gabor (bottom-up) and inverse (top-down) visual stimuli. We report widespread high-frequency bursts in L2/3 and L5 pyramidal neurons. Contrary to expectation, bursts exhibited extremely short response latencies, and were most strongly recruited by Gabor stimuli. We further tested the causal contribution(s) of apical dendrites to burst firing and top-down visual tuning via two optogenetic manipulations: direct L5 apical tuft inhibition and NDNF interneuron activation. Strikingly, L1 inhibition only modestly reduced the burst fraction, and did not differentially affect Gabor vs inverse responses. Taken together, these results challenge prevailing theories of apical dendritic involvement in burst spike generation and feedback visual tuning, and provide new biological constraints for future theoretical and experimental work. Thesis supervisor: Mark T. Harnett Title: Professor of Brain and Cognitive Sciences 5 6 Acknowledgments It has been an honor and a privilege to pursue this thesis work over the past six years. I am deeply thankful to my advisor, Professor Mark Harnett, for giving me the opportunity to train and embark on this journey in his lab. Your mentorship, rigor, and openness to chasing the unknown have profoundly shaped my development as a scientist. I am grateful to the members of my thesis committee, Professors Mark Bear, Jim DiCarlo, and Richard Naud, for their time, thoughtful feedback, and insightful discussions. It has been a true privilege to have your expertise and perspective over the course of this work. This work would not have been possible without my colleagues in the Harnett lab, who have created the most supportive environment I could have asked for. I am deeply grateful for the scientific collaborations and friendships we have forged together. I am thankful for all the friends and loved ones who have made the past years so wonderful. You all have been an invaluable source of support, and I am deeply grateful for our time together. I owe special thanks to my family — my parents, Xianzhu and Tian, and my brother, Andrew — for their unwavering encouragement and for nurturing my love for science. And finally, I am deeply grateful to my partner Kelly, who has been for being a constant companion and source of inspiration. 7 8 Contents List of Figures 11 1 Introduction 13 1.1 Anatomical organization of cortical inputs . . . . . . . . . . . 15 1.2 The role of dendritic biophysics . . . . . . . . . . . . . . . . . 20 1.3 Cortical burst spiking . . . . . . . . . . . . . . . . . . . . . . . 23 1.4 Bottom-up and top-down signal processing in mouse primary visual cortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 2 Visual tuning of burst and tonic spikes 35 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 2.2.1 Extracellular bursts are ubiquitous and exhibit spatio- temporal features consistent with dendritic recruitment 37 2.2.2 Feedforward recruitment of cortical bursts . . . . . . . 38 2.2.3 Inverse stimuli do not drive increased bursting . . . . . 41 2.2.4 Stimulus driven bursting is invariant to preceding stimuli 42 2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 3 Apical dendritic contributions to burst spiking and top-down visual processing in vivo 57 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.2.1 Apical dendritic activity is not required for burst firing 58 3.2.2 Apical tuft dendrites are not required for inverse tuning 61 3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4 Final Conclusions 65 4.1 Visual tuning of cortical burst and tonic spikes . . . . . . . . . 66 4.2 Causal investigation of the apical tuft . . . . . . . . . . . . . . 67 4.3 A revised framework for top-down and bottom-up integration . 68 9 A Methods 71 A.1 Animals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 A.2 Surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 A.2.1 Acute procedures . . . . . . . . . . . . . . . . . . . . . 72 A.2.2 Chronic procedures . . . . . . . . . . . . . . . . . . . . 74 A.2.3 Silicon probe preparation . . . . . . . . . . . . . . . . . 75 A.3 Electrophysiology . . . . . . . . . . . . . . . . . . . . . . . . . 76 A.3.1 Acute recordings . . . . . . . . . . . . . . . . . . . . . 76 A.3.2 Chronic recordings . . . . . . . . . . . . . . . . . . . . 77 A.4 Optogenetics . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 A.5 Visual stimulation . . . . . . . . . . . . . . . . . . . . . . . . . 79 A.5.1 Gabor vs inverse tuning . . . . . . . . . . . . . . . . . 80 A.5.2 Size tuning . . . . . . . . . . . . . . . . . . . . . . . . 81 A.5.3 Orientation tuning . . . . . . . . . . . . . . . . . . . . 81 A.5.4 Sequential stimuli . . . . . . . . . . . . . . . . . . . . . 81 A.6 Behavioral monitoring . . . . . . . . . . . . . . . . . . . . . . 82 A.7 Histology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 A.8 Data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 A.8.1 Spike-sorting . . . . . . . . . . . . . . . . . . . . . . . 84 A.8.2 Layer identification . . . . . . . . . . . . . . . . . . . . 85 A.8.3 Burst spike analysis . . . . . . . . . . . . . . . . . . . . 85 A.8.4 Spike backpropagation . . . . . . . . . . . . . . . . . . 86 A.8.5 White noise analysis . . . . . . . . . . . . . . . . . . . 86 A.8.6 Receptive field mapping . . . . . . . . . . . . . . . . . 87 A.8.7 Inverse tuning . . . . . . . . . . . . . . . . . . . . . . . 87 A.8.8 Response latency . . . . . . . . . . . . . . . . . . . . . 88 A.8.9 Poisson model . . . . . . . . . . . . . . . . . . . . . . . 88 A.8.10 Optogenetic inhibition . . . . . . . . . . . . . . . . . . 89 A.8.11 Behavioral state . . . . . . . . . . . . . . . . . . . . . . 89 A.9 Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 References 91 10 List of Figures 1.1 Laminar organization of top-down and bottom-up inputs . . . 16 1.2 Axon distribution of inputs to V1 . . . . . . . . . . . . . . . . 19 1.3 V2 to V1 feedback projections . . . . . . . . . . . . . . . . . . 30 1.4 Example layer 5 pyramidal neuron morphology and biophysics 31 1.5 Functional roles of burst spiking . . . . . . . . . . . . . . . . . 32 1.6 Previous evidence for inverse tuning . . . . . . . . . . . . . . . 33 1.7 Model for the integration of top-down and bottom-up inputs via a two-compartment neuron framework . . . . . . . . . . . . . 34 2.1 Detection and characterization of extracellular burst spikes . . 47 2.2 Spike sorting and manual curation . . . . . . . . . . . . . . . . 48 2.3 Extracellular spike bursts decrement in amplitude across time and space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 2.4 Gabor stimuli drive rapid recruitment of burst spiking . . . . . 50 2.5 Unsupervised detection of early bursting neurons . . . . . . . 51 2.6 Tuning of bursts and tonic spikes to white noise stimuli . . . . 52 2.7 Poisson firing rate model . . . . . . . . . . . . . . . . . . . . . 53 2.8 Inverse stimuli do not evoke higher burst firing than Gabors. . 54 2.9 Differences in whisking do not explain lack of inverse stimuli driven bursting . . . . . . . . . . . . . . . . . . . . . . . . . . 55 2.10 Sequential presentation of Gabor, inverse, and full-field stimuli 56 3.1 Optogenetic inhibition of the apical tuft . . . . . . . . . . . . 60 3.2 Effect of apical tuft inhibition on Gabor and inverse responses 62 11 12 Chapter 1 Introduction The mammalian neocortex is a marvel of biological evolution, orchestrating a vast array of cognitive functions that underlie perception, behavior, and phenomenological experience. Deciphering how the brain’s billions of neurons and trillions of connections interact to produce these capabilities remains a critical frontier, with far-reaching implications for both medicine and artificial intelligence. How do we draw core functional principles from a system of such staggering complexity? A pivotal early insight was the discovery that the neocortex is broadly organized into distinct functional areas, each with specialized roles ranging from sensory processing and motor control to goal- directed decision making (Felleman & Van Essen, 1991; Miller & Cohen, 2001; Passingham et al., 2002). Furthermore, many sensory areas are broadly organized into hierarchical networks, whereby sensory information enters the neocortex in lower-order areas and progressively transforms into more abstract representations as it ascends the cortical hierarchy (J. A. Harris et al., 2019; 13 Miller & Cohen, 2001; Siegle et al., 2021; Van Essen & Maunsell, 1983). While this bottom-up framework provides a powerful explanation for basic features of sensory encoding, cortical processing is far more intricate than a simple feedforward flow of information. Instead, the brain is highly recurrent, with extensive connections from higher-order cortical and thalamic areas into their lower-order counterparts. These top-down inputs perform a critical role in cortical function, producing changes in feedforward activity that underlie perception, learning, and adaptive behavior (Banerjee et al., 2020; Gilbert & Li, 2013; Makino & Komiyama, 2015; Manita et al., 2015; Noudoost et al., n.d.; Zhang et al., 2014). How these inputs are integrated within the cortical microcircuit remains a significant gap in knowledge. A key organizational principle of bottom-up and top-down inputs is their spatial segregation across the dendritic compartments of single pyramidal neurons (Schuman et al., 2021). These dendritic compartments, in turn, possess distinct biophysical properties which enable powerful non-linear signal processing (M. E. Larkum, Kaiser, & Sakmann, 1999; Major et al., 2013; Stuart & Spruston, 1998). Specifically, the differential processing of bottom-up and top-down inputs in the peri-somatic and distal apical dendrites (respectively) has provided a prevailing theoretical framework, inspiring powerful algorithmic-level hypotheses for myriad cortical computations. Despite considerable theoretical interest, however, there remains little in vivo evidence for either the subcellular or circuit-level predictions of these models. This is due, in large part, to the difficulty of isolating bottom- up and top-down processes in a model system conducive to experimental interrogation. 14 This thesis investigates a prevailing framework for the integration of bottom- up and top-down inputs via non-linear dendritic mechanisms. In Chapter 1, I review key ex vivo evidence for this model, and identify critical assumptions and predictions which have yet to be tested in vivo. In Chapter 2, I leverage the recent finding of a feedback-dependent visual tuning in the mouse primary visual cortex to test the prediction that top-down inputs preferentially drive high-frequency burst spiking, a non-linear output mode thought to be mediated by distal apical dendrites. In Chapter 3, I perform targeted optogenetic perturbations to causally test the role of the apical tuft in generating burst spikes and top-down visual tuning. Together, the results of these experiments provide direct, in vivo evidence complicating core predictions of existing models. In Chapter 4, I conclude by discussing the functional implications of these findings and directions for future investigation. 1.1 Anatomical organization of cortical inputs The neocortex exhibits a strikingly conserved laminar organization across different cortical areas and mammalian species (Douglas & Martin, 2004; J. A. Harris et al., 2019; Mountcastle, 1997; Rakic, 2009)(Fig. 1.1). This layered structure has long been hypothesized to support a functional organization of inputs, with specific layers and dendritic compartments within each cortical area preferentially receiving inputs from different cortical and thalamic areas. Early retrograde tracing studies found that feedforward axons originating from primary sensory thalamic areas (e.g. lateral and medial geniculate nuclei) 15 Figure 1.1: Laminar organization of top-down and bottom-up inputs. Bottom up (feedforward) inputs from lower order thalamic and cortical areas are preferentially targeted towards L4, and are distributed locally to L2/3 and L5. In contrast, top-down (feedback) inputs from higher order areas largely avoid L4. Instead, they are highly enriched in L1, where they impinge on the apical dendrites of L2/3 and L5 pyramidal neurons, and in deeper layers. are predominantly received in layer 4 of primary sensory cortex (Douglas & Martin, 2004; Gilbert & Wiesel, 1979; E. G. Jones & Powell, 1970; Rockland & Pandya, 1979). Subsequent axon bouton reconstruction and functional studies demonstrated that bottom-up projections from lateral geniculate nucleus synapse heavily onto the peri-somatic dendrites of L4 pyramidal neurons (Ferster & Lindström, 1983; Freund et al., 1985; Lien & Scanziani, 2018b), although they also target the apical oblique dendrites of L5 pyramidal neurons (Constantinople & Bruno, 2013). These inputs are then processed recurrently within the local microcircuits, where they are further distributed to the peri-somatic dendrites of pyramidal neurons in layers 2/3 (L2/3) and 5 (L5) within the same area (Rockland & Pandya, 1979). 16 In contrast, feedback inputs from many higher-order thalamic and cortical regions are noticeably attenuated in L4, and are instead preferentially targeted towards layer 1 (L1) and deeper layers (Fig. 1.1). In the primary visual cortex (V1), for instance, inputs from the lateral geniculate nucleus (LGN) - the primary source of visual information from the retina - terminate predominantly in L4 (Douglas & Martin, 2004; Gilbert & Wiesel, 1979; E. G. Jones & Powell, 1970; Rockland & Pandya, 1979), while inputs from higher order visual areas such as the lateromedial visual cortex (LM) and posteriomedial visual cortex (PM) are targeted towards L1 and L5/6 (Fig. 1.2). The observation of similar connectivity motifs in higher-order cortical areas (Felleman & Van Essen, 1991; Rockland & Pandya, 1979; Van Essen & Maunsell, 1983) has led to the proposal that this laminar segregation of feedforward and feedback inputs represents a canonical organizational principle of the neocortex (Douglas & Martin, 2004; Felleman & Van Essen, 1991). This structured organization is thought to be essential for the integration of externally driven signals (bottom-up) with internally generated predictions, prior knowledge, and contextual information (top-down). Previous studies have shown that descending cortical signals play a crucial role in modulating perception and behavior, contributing to processes such as contextual mod- ulation, predictive coding, attention, expectation, multisensory integration, and learning (Doron et al., 2020; Gilbert & Li, 2013; Jordan & Keller, 2020; Nurminen et al., 2018; Pak et al., 2020; N. L. Xu et al., 2012; Zanto et al., 2011; Zipser et al., 1996). A critical implication of these findings is that long range inputs carrying top-down information which must be adaptively integrated with 17 bottom-up sensory signals, instead of simply overwriting them. However, it remains unclear to what extent the vast array of long range higher order signals share the same canonical laminar organization that was first derived from the hierarchical areas within individual sensory modalities (Felleman & Van Essen, 1991). In V1, for instance, feedback axons from higher order cortical and thalamic areas such as the anterior cingulate cortex (ACC), orbitofrontal cortex (ORB), and the lateral posterior nucleus of the pulvinar (LP) differ noticeably in their laminar organization, with differences in their innervation of deeper layers (ACC preferentially targets L6, ORB targets L5, and LP innervates both L5 and L6), and the relative density of their L1 innervation (Ährlund-Richter et al., 2024; Liu et al., 2024). Even for feedback axons from V2 to V1, previous studies have reported differing laminar distributions, complicating efforts to validate a unified functional motif (Shen et al., 2022; Young et al., 2021). An important caveat to these studies is that axon density does not directly equate to functional connectivity (Oh et al., 2014). Here, direct functional assays such as subcellular channelrhodopsin assisted circuit mapping (sCRACM) have begun to provide critical evidence by distinguishing the cell types and subcompartments which are functionally targeted by specific inputs (Lafourcade et al., 2022; Petreanu et al., 2009; Young et al., 2021). sCRACM mapping of V2 inputs to V1 have found that V2 inputs strongly target the peri-somatic compartments of looped V1 pyramidal neurons which send inputs back to V2 (Young et al., 2021). This evidence is consistent with the substantial deeper layer innervation of V2 axons (J. A. Harris et al., 2019; Marques, Nguyen, et al., 2018; Young et al., 2021), suggesting that they could represent a substantial 18 Figure 1.2: Axon distribution of inputs to V1. Inputs from LGN are highly localized to L4, while inputs from higher order thalamus (LP) and cortex (ACC) are strongly biased towards L1. Inputs from lateromedial (LM) visual cortex are present in L1, but also exhibit substantial innervation in deeper layers (L5 and L6). LGN, LP, and ACC panels adapted from Schuman et al., 2021, images from the Allen Connectivity Map. parallel pathway for V2 to V1 feedback (Fig. 1.3). However, these deep feedback projections remain markedly understudied compared to feedback projections to L1 (Doron et al., 2020; Fişek et al., 2023; Huang et al., 2024; Keller et al., 2020; Marques, Summers, et al., 2018), and their functional role remains unclear. Instead, top-down processing remains strongly associated with L1 inputs (M. E. Larkum et al., 2009; Manita et al., 2015; Rao et al., 2021; Schuman et al., 2021; Siegel et al., 2000; S. Xu et al., 2012). The predominant hypothesis that L1 serves as a privileged integration hub for top-down inputs is due, in part, to its conserved targeting by many long 19 range inputs and its unique anatomical features (Schuman et al., 2021). In addition to being heavily innervated by feedback from higher order cortical and thalamic areas, L1 receives little excitatory input from within the local microcircuit (Chu et al., 2003; Jiang et al., 2015; Petreanu et al., 2009). Furthermore, L1 is unique among cortical layers in that it does not contain any pyramidal neurons. Instead, the dense excitatory inputs from higher-order areas impinge on the apical dendrites of L2/3 and L5 pyramidal neurons, and a unique population of L1-residing interneurons (Cohen-Kashi Malina et al., 2021). Notably, the apical dendrites of the primary thalamo-recipient neurons, L4, are significantly attenuated compared to those of neurons in other layers 1.1. Given this connectivity, L2/3 and L5 pyramidal neurons are uniquely positioned to integrate both bottom-up sensory inputs and top-down modulatory signals. These neurons, with their extensive apical dendrites in L1 and peri-somatic dendrites in deeper layers, may function as powerful computational units that leverage this dendritic compartmentalization to differentially process bottom-up and top-down inputs. 1.2 The role of dendritic biophysics Dendrites receive the vast majority of synaptic inputs in a pyramidal cell. Instead of simply conveying these inputs to the soma unchanged, they leverage a complex array of biophysical properties to transform inputs prior to action potential (AP) generation. Dendrites exhibit passive cable properties that produce significant distance- 20 dependent signal attenuation (Stuart & Spruston, 1998). This electrical com- partmentalization is particularly apparent in large layer 5 pyramidal neurons, whose dendrites span all cortical layers (Mason & Larkman, 1990). Second, the apical dendrites of both layer 2/3 and 5 pyramidal neurons arborize ex- tensively in L1 and possess an array of voltage-dependent ion channels that allow them to actively produce Na+, N-Methyl-D-Aspartate (NMDA), and Ca2+ spikes that mediate supra-linear signal transformation (Palmer et al., 2014; Stuart et al., 1997). The combination of these properties allows dendrites to non-linearly integrate information, and “gate” subthreshold signals. Third, somatic action potentials can backpropagate into dendritic branches (Fig. 1.4) (M. E. Larkum, Kaiser, & Sakmann, 1999; Stuart et al., 1997). The extensive branching and electrotonic properties of apical dendrites have motivated models of single neurons as multi-layer neural networks (David et al., 2019). This differs significantly from most current research in machine learning and artificial intelligence, which overwhelmingly models individual neurons as simple rate or integrate-and-fire units (Higgins et al., 2020; Li & DiCarlo, 2008; Yamins et al., 2014). In layer 5 pyramidal neurons, the expression of high-voltage-activated Ca2+ channels in the apical trunk provides an additional mechanism for supra- linear signal processing. When subthreshold events in the apical dendrite are temporally coincident with somatic spiking, the additional activation of a dendritic Ca2+ spike can generate powerful regenerative currents, transforming single action potentials into high-frequency bursting events (M. E. Larkum, Zhu, & Sakmann, 1999). These regenerative currents can also be engaged by 21 activation of the apical tuft alone, allowing strong distal inputs to overcome their electrotonic isolation and elicit high-frequency burst firing at the soma (Fig. 1.4) (M. Larkum, 2013; M. E. Larkum, Zhu, & Sakmann, 1999; Major et al., 2013). The distinct electrotonic and active properties of the apical tuft versus the peri-somatic dendrites, combined with their differential sampling of feed- back inputs, suggests a powerful two-compartment framework for single cell computation 1.7 (M. Larkum, 2013; Shai et al., 2015; Siegel et al., 2000; Yi et al., 2017). A growing body of theoretical work has leveraged these prop- erties to propose computational models at both the single cell and circuit levels, providing algorithmic-level hypotheses for how cortical circuits could perform both complex signal processing (Hawkins & Ahmad, 2016; Naud et al., 2023) and implement powerful learning rules (Greedy et al., 2022; Guerguiev et al., 2017; Körding & König, 2001; Richards & Lillicrap, 2019; Sacramento et al., 2018). Despite its functional implications, a significant limitation of the two-compartment framework is that its supporting evidence remains almost entirely ex vivo. Ex vivo measurements of neural activity differ substantially from intact circuits in awake animals, with critical differences including the lack of ongoing synaptic activity, spatio-temporally patterned synaptic input, and neuromodulator release, all of which can dramatically affect integrative properties at both the single cell and circuit levels (Destexhe & Paré, 1999; Destexhe et al., 2003; Hasselmo, 1995; London & Häusser, 2005; Marder & Thirumalai, 2002; McCormick, 1992; Okun & Lampl, 2008; Opitz et al., 2017). In fact, the constant barrage of synaptic inputs in vivo has been theorized to 22 drive cells into a high conductance state, under which the non-linear properties of different dendritic compartments are largely normalized (Destexhe et al., 2003). Recent studies of in vivo dendritic activity have only highlighted this gap in our knowledge. Given the high electrotonic separation between the apical tuft and soma, it was widely theorized that apical tuft activity would be decoupled from the soma, and exhibit a substantial amount of independent activity (driven by active spikes). However, recent in vivo studies leveraging two- photon calcium imaging to record simultaneous activity in apical dendrites and their parent somata have found that GCaMP signals are highly synchronized across the two compartments, with little to no independent encoding in the apical tuft (Beaulieu-Laroche et al., 2019; Francioni et al., 2019; Kerlin et al., 2019; Schoenfeld et al., 2021, but see Voigts and Harnett, 2020). Here, electrophysiological recordings of burst firing offers promising complement to these studies, since it can be assayed across large numbers of neurons in vivo, and has a strong biophysical link to apical dendritic engagement. Moreover, the top-down recruitment of burst firing is a direct assumption of a growing body of powerful theoretical models that leverage a two compartment framework. 1.3 Cortical burst spiking Burst spiking powerfully alters neural output, increasing the efficacy and precision of synaptic transmission (Lisman, 1997) and providing a mechanism for multiplexing top-down and bottom-up information in the output of a 23 single pyramidal neuron 1.5 (Kepecs & Lisman, 2003; Naud & Sprekeler, 2018; Naud et al., 2023). In addition to their putative roles in moment-to-moment information processing, bursts are also strongly implicated in learning, where their high-frequency firing and the associated intracellular influx of Ca2+ offer mechanisms to for long-term synaptic modifications (Cichon & Gan, 2015; Froemke et al., 2006; Kampa et al., 2006). In the hippocampus, bursts have been shown to mediate a powerful form of one-shot plasticity known as behavioral timescale synaptic plasticity (Bittner et al., 2017). Emerging theoretical work has consolidated these ideas to suggest that apical dendrite mediated bursting mediates credit assignment within biological neural networks (Greedy et al., 2022; Payeur et al., 2020; Richards & Lillicrap, 2019). Together, these models introduce multiple algorithmic-level descriptions of how cortical circuits could utilize high-frequency bursts to perform both computation and learning. However, key assumptions and predictions of these models have yet to be validated in intact cortical neural circuits. Much of the existing literature on burst spiking, and its potential differences from tonic spikes, derives instead from studies in weakly electric fish (Clarke & Maler, 2017; Gabbiani et al., 1996; Metzen et al., 2016; Metzner et al., 1998; Oswald et al., 2004), thalamus (Alitto et al., 2005; Contreras et al., 1992; Guido et al., 1992; Llinás & Steriade, 2006; Reinagel et al., 1999; Sherman, 2001), or hippocampus (Lowet et al., 2023). Although these studies provide powerful proof-of-principle that burst spikes could serve distinct functional roles within neural circuits, they were conducted in circuits and neurons that differ significantly from those within mammalian neocortex. As such, it remains an open question whether the 24 theorized computational roles of burst spikes represent generalized functional principles. Within cortical circuits, burst spikes have been reported in early studies in the cat and primate visual cortex, where they were found to be more discriminative of stimulus orientation than tonic spikes during microsaccadic eye movements (Cattaneo et al., 1981; Livingstone et al., 1996; Martinez-Conde et al., 2002). Although these studies suggested that cortical burst spikes may play a role in sensory processing, these bursts shared tuning features with tonic spikes, and thus were hypothesized to reflect stimuli salience and optimality rather than encode unique visual features (Martinez-Conde et al., 2002). In the rat primary somatosensory cortex, previous studies have shown that high-frequency burst spikes encode active, but not passive touch (de Kock et al., 2021; De Kock & Sakmann, 2008; De Kock et al., 2007). Interestingly, several studies have shown linked cortical burst firing with the detectability of stimuli during either active whisking or electrical stimulation in the barrel cortex (Doron et al., 2014; Takahashi et al., 2016). As such, few studies have attempted to explicitly disambiguate whether cortical bursting events and tonic spikes differentially encode top-down and bottom-up information (Anderson et al., 2011, 2013). It remains unclear whether bursts are primarily recruited to process bottom-up inputs (i.e. serving a thresholding role in sensory detection or enhancing information transmission), or serve to preferentially encode a privileged class of higher-order signals. In order to distinguish between these two hypotheses, it is critical to directly compare the degree to which burst spikes are recruited by bottom-up and top-down inputs. Furthermore, it remains an open question to what extent 25 burst spikes remain biophysically distinct from tonic spikes in vivo. Although the recruitment of burst spikes by the apical tuft is central to many two- compartment neuron frameworks (M. Larkum, 2013; Shai et al., 2015), a strong causal link has yet to be established in awake animals. Interestingly, although burst spikes are attributed to a subpopulation of L5 neurons with large apical dendritic processes in slice and in anesthetized animals, they have been observed in a broader population of neurons (e.g. L2/3 and putative L5 IT) in awake, in vivo recordings, suggesting that additional biophysical mechanisms may be at play in awake animals (De Kock & Sakmann, 2008; De Kock et al., 2007; Sakmann, 2017). Indeed, other models have generated high-frequency bursts without direct inputs to the distal apical tuft (Bast et al., 2023; Izhikevich, 2003; Rhodes & Gray, 1994), and suggested that bursting may instead serve as an input slope detector (Kepecs et al., 2002). Interestingly, it has also been suggested that bottom-up inputs from sensory thalamus can directly recruit active dendritic processing, bypassing the need for top-down signals altogether (Bast et al., 2023). Directly testing these hypotheses in intact neural circuits is therefore critical for understanding the role of burst spikes cortical computation, and will provide new biological constraints to guide theoretical work. 1.4 Bottom-up and top-down signal processing in mouse primary visual cortex The lack of in vivo evidence for either the subcellular or circuit-level predictions of these models is due, in part, to the difficulty of isolating bottom-up and 26 top-down processes in a model system amenable to experimental interrogation. Here, recent advances in our understanding of the mouse primary visual cortex (V1) present a compelling opportunity to overcome these hurdles. The response properties of neurons in primary visual cortex (V1) have been extensively studied over the past decades (Angelucci et al., 2017; DeAngelis et al., 1993; DeAngelis et al., 1995; Hubel & Wiesel, 1962; Marques, Nguyen, et al., 2018; Niell & Stryker, 2008), and V1’s connectivity to other regions is the most highly explored of all cortical areas (Angelucci et al., 2002; Zhang et al., 2014). As the first cortical visual processing region, V1 receives strong bottom-up visual input from the lateral geniculate nucleus. These inputs are processed on single neurons and within the cortical column, providing a mechanistic basis for classical receptive field localization, and properties such as ON/OFF tuning and orientation selectivity (Angelucci & Bressloff, 2006; Angelucci et al., 2002, 2017; Federer et al., 2021; Hubel & Wiesel, 1959; Hubel & Wiesel, 1962; Iacaruso et al., 2017; Lien & Scanziani, 2018a; Nurminen et al., 2018; Siu et al., 2021). This feedforward tuning is thought to be well approximated as a Gabor filter (J. P. Jones & Palmer, 1987). However, mounting evidence also highlights the role of top-down inputs even in early stages of visual processing, demonstrating that V1 neurons are far more than bottom-up feature detectors. Previous studies have shown that V1 encodes visual flow, reward, and spatial information (Diamanti et al., 2019; Jordan & Keller, 2020; Saleem et al., 2018; Zhang et al., 2014; Zong et al., 2021), and possesses higher-order visual tuning properties such as figure-ground 27 modulation and extra-classical receptive fields (Keller et al., 2020; Lamme et al., 1998; Schnabel et al., 2018; Walker et al., 2019; Zipser et al., 1996). These tuning properties are thought to be mediated by feedback inputs from higher-order visual areas, as well as anterior cingulate cortex and retro-splenial cortex (de Vries et al., 2020; Makino & Komiyama, 2015; Marques, Nguyen, et al., 2018; Marques, Summers, et al., 2018; Nurminen et al., 2018). Of particular interest is the recent discovery that strong extra-classical receptive field tuning in mouse V1 is driven by feedback from V2 (Keller et al., 2020). The tuning to these inverse stimuli exhibits a long temporal latency consistent with a top-down origin, and is abolished by inactivation of V2 areas (Fig. 1.6). Notably, inverse tuning is present in L2/3 and L5 pyramidal neurons but not L4, suggesting that L1 inputs may play a critical role. Previous work investigated this possibility by imaging feedback axons from lateral medial visual (LM) cortex into L1 of V1, finding that L1 feedback inputs are not inverse tuned themselves, but contain the requisite information to generate inverse tuning (Fig. 1.6). Therefore, a supra-linear integration of L1 inputs in the apical dendrites of L2/3 and L5 pyramidal neurons would provide a possible explanation for feedback driven inverse tuning. The feasibility of such a mechanism was directly investigated in a recent study suggesting that inverse stimuli preferentially active dendritic events in the apical dendrites of L5 pyramidal neurons in V1 (Fişek et al., 2023). Together, these studies provide substantial evidence that V1 could differentially process top-down and bottom-up visual inputs via the two compartment neuron framework 1.7. This hypothesis leads to two experimentally testable predictions: 1. Burst and tonic 28 spikes should preferentially encode inverse and Gabor stimuli, respectively, and 2. apical tuft inhibition should selectively abolish burst spike generation and the inverse response. 29 Figure 1.3: V2 to V1 feedback projections. A-B. Axon tracing of inputs from lateral visual cortices (LM) into V1 shows substantial innervation of L5 and L6 in addition to L1. Despite targeting the same area, the laminar specificity varies across studies. C. Subcellular assisted circuit mapping (sCRACM) shows the subcellular targeting of V2 to V1 inputs for a subset of L2/3 and L5 neurons. D. Same data as in C, aligned to the cell body of each neuron instead of laminar depth. L2/3 and L5 neurons receive strong peri-somatic drive from V2 to V1 inputs, with only L5 IT neurons exhibiting significant apical dendritic input from V2. A. adapted from Shen et al., 2022, B,C,D. adapted from Young et al., 2021. 30 Figure 1.4: Example layer 5 pyramidal neuron morphology and biophysics. A. Top-down inputs to single neurons are received in the apical tuft dendrites (L1), while bottom-up inputs are preferentially targeted to the soma. B-D describe simultaneous triple patch experiments performed in vitro. B. Subthreshold stimulation near the apical trunk attenuates heavily along the dendritic trunk. C. Single action potentials evoked in the soma back-propagate to dendritic compartments. D. Combination of subthreshold apical trunk stimulation and a single action potential evokes a supra-linear bursting event. E. Strong apical dendritic stimulation can independently generate somatic bursting. Adapted from M. E. Larkum, Zhu, and Sakmann, 1999 31 Figure 1.5: Functional roles of burst spiking. A. The selective recruitment of burst spikes by distinct input patterns allows single neurons to multiplex dif- ferent information streams. Pathway 1 (putative peri-somatic input) generates tonic spikes when activated alone, but burst spikes when combined with inputs from pathway 2 (putative apical dendritic input) B. Short-term potentiation and depression at the post-synaptic neuron could allow burst spikes to be differentially decoded. C. Differential pairing of bursts and single spikes at the pre and postsynaptic level may lead to either long term potentiation or depression. Adapted from Friedenberger et al., 2023. 32 Figure 1.6: Previous evidence for inverse tuning. A-B. Classical and inverse stimuli share a spatial receptive field. C-D. V2 inactivation selectively abolishes inverse (bottom panel), but not classical (top panel) tuning. E. Inverse responses exhibit a longer latency than classical, consistent with a feedback recruitment. F. Imaging of LM axons in layer 1 of V1 (left panel) showed that LM inputs are not inverse tuned themselves (middle), but the sum of LM activity contains the requisite information to generate inverse tuning (right). G-H. Imaging of L5 apical dendrites suggests that inverse stimuli recruit more active dendritic events than other stimuli. Panels A-F adapted from Keller et al., 2020, panels G-H adapted from Fişek et al., 2023. 33 Figure 1.7: Model for the integration of top-down and bottom-up inputs via a two-compartment neuron framework. Bottom-up inputs are localized to the perisomatic compartments, while top-down inputs are targeted to the apical dendrites in L1. The differential recruitment of voltage-gated supra-linearities by these two streams is hypothesized to produce a differential recruitment of burst and tonic spikes. In the primary visual cortex, Gabor and inverse stimuli are thought to drive bottom-up and top-down inputs, respectively, thus resulting in a differential recruitment of tonic and burst spikes. 34 Chapter 2 Visual tuning of burst and tonic spikes 2.1 Introduction High-frequency burst spiking offers a powerful mechanism by which single pyramidal neurons can multiplex multiple information streams, and is thought to be preferentially recruited by top-down signal processing in apical dendrites. Although a growing number of theoretical models have posited that cortical circuits could leverage burst spikes to perform powerful computation functions, there remains little characterization of their response properties in vivo. It is unknown whether cortical burst and tonic spikes encode different information, and specifically, whether they are preferentially tuned to top-down and bottom- up inputs. In the mouse primary visual cortex, masked drifting gratings (inverse stimuli) 35 can powerfully drive individual V1 neurons, even in the absence of stimuli in their feedforward receptive fields (Keller et al., 2020). Unlike classic bottom-up stimuli (Gabor patches), the tuning to inverse stimuli requires top-down inputs from V2, providing an experimental model for testing the coding of burst and tonic spikes. 2.2 Results We performed extracellular electrophysiology using a battery of inverse and Gabor visual stimuli to test whether burst and single spikes code for distinct visual information in the primary visual cortex of awake, head-fixed mice (Fig. 2.1A). For the Gabor presentation experiments, mice were chronically implanted with Neuropixels 2.0 silicon electrodes, allowing us to record large populations of neurons across all cortical layers with minimal acute damage to layer 1 (Fig. 2.1A). While this extracellular recording approach enabled us to stably record hundreds of neurons for the time periods necessary for our visual stimuli battery, the non-stationarity of waveform shape and amplitude across the course of a burst (Allen et al., 2018; Connors et al., 1982; Fee et al., 1996; Henze et al., 2000; McCormick et al., 1985) poses a considerable challenge for the spike sorting algorithms used to process these data (K. D. Harris et al., 2000; Lewicki, 1998). In particular, the decrementing waveform amplitudes of consecutive spikes within a burst commonly lead to under-detection of burst spikes, and the over-splitting of single bursty neurons into multiple units (Fig. 2.2A). In order to mitigate the under-detection of burst spikes during spike 36 sorting, we utilized state-of-the-art automated algorithms (Kilosort 2.5 and 4) utilizing template-based spike detection and template subtraction, which improve the detection of lower amplitude waveforms and reduce the impact of overlapping spikes, respectively (Pachitariu et al., 2016, 2024). After the initial automated spike detection, we corrected for the over-splitting of bursty units by performing extensive manual curation (Rossant & Harris, 2013). In order to maximize the recovery of over-split bursts, we examined all units on nearby channels, and used cross-correlogram and raw trace inspection in addition to waveform similarity metrics (Fig. 2.2B). After merging, waveform quality control, and removing putative fast spiking interneurons, we identified 1158 neurons across 7 mice. 2.2.1 Extracellular bursts are ubiquitous and exhibit spatio- temporal features consistent with dendritic recruit- ment We defined bursts as events where at least three consecutive spikes occurred with a <10 ms inter-spike interval (Fig. 2.1B). Although the exact criteria of a cortical burst has been debated (Krahe & Gabbiani, 2004; Naud et al., 2023), we chose a stringent burst criteria, consistent with previous literature, in order to maximize distinction between burst and tonic events (Bast et al., 2023). Although our extracellular recordings likely under-represent the true number of bursts, we were able to recover high-frequency bursts in both L2/3 and L5, including those with strongly decrementing waveform amplitudes (Fig. 37 2.1C). Burst spikes were particularly prevalent in L2/3 and deep L5 (Fig. 2.1D), consistent with previous in vivo measurements and the hypothesized differentiation of L5a vs L5b (Senzai et al., 2019). In a subset of recordings made perpendicular to the cortical surface using acute Neuropixel 1.0 probes, we found that consecutive spikes within burst events decreased in their spatial backpropagation along the laminar probe (Fig. 2.3), consistent with previous extracellular reports (Bereshpolova et al., 2007), and with the engagement of unique biophysical mechanisms during high frequency bursts (Connors et al., 1982; Fee et al., 1996; McCormick et al., 1985). 2.2.2 Feedforward recruitment of cortical bursts To investigate the coding properties of these cortical bursts, we first charac- terized their tuning and onset latencies in response to classical feedforward stimuli. We mapped each neurons classical receptive field (RF) (Fig. 2.4A). We identified 669 neurons (n=7 mice) that had a classical spatial receptive field, and performed the subsequent analyses on these neurons. We found that the receptive field tuning for tonic and burst spikes was largely identical for neurons with a significant bursting response to visual stimuli (n=461), and then analyzed the response of burst vs tonic spikes at each neuron’s preferred RF location. Contrary to expectation, bursts exhibited short response latencies, and were recruited in the initial feedforward response to visual stimuli (Fig. 2.4B,C). To further characterize the relative contribution of burst spikes to the visual response, we calculated the burst fraction as the number of burst spikes in a given time bin divided by the total number of spikes (Fig. 2.4D. We found 38 that the majority of Gabor tuned neurons (65.7%, 303 of 461 neurons) had a short latency peak in the burst fraction, defined as a significant burst response within 150 ms after stimuli presentation. When we analyzed the relative timing of the burst fraction and firing rate for these early bursting neurons, we found that the half-peak latency of burst fraction was significantly different, and preceded that of the firing rate at the population level (p=0.0004, Fig. 2.4F,G). This indicated that across the population level, burst spikes were recruited in the initial onset of the visual response, instead of exhibiting a longer latency characteristic of feedback recruitment. To verify our latency threshold for early bursting (150 ms), we performed a k-means clustering on the firing rate and burst fraction latencies, which similarly revealed that the majority of visually responsive neurons resided within an early bursting cluster (Fig. 2.5A, 300 of 461 neurons are assigned to cluster 1). For these neurons, their burst fraction again peaked earlier than their firing rate (Fig. 2.5B,C). In order to test whether the Gabor evoked bursting could explained by the increase in firing rate, we used a Poisson firing rate model to simulate artificial spikes which matched the firing response of each neuron (Fig. 2.7A). We then calculated a model estimated burst rate for each neuron, which correlated closely with the firing rate and provided an estimate of the burst rate predicted solely by firing rate statistics (Fig. 2.7B). The Poisson estimated burst rate differed substantially from the actual burst rate, under-predicting the initial burst rate ramp and over-predicting the subsequent response (Fig. 2.7B,D). The same was true when we analyzed the burst fraction instead of the burst rate (Fig. 2.7C,E), indicating that the Gabor evoked bursting is poorly explained by 39 firing rate alone. Although both the burst rate and burst fraction were well approximated at the population level, the large mean absolute error of the Poisson model indicates that the Poisson model fits extremely poorly on the individual neuron level (Fig. 2.7D,E). Although the burst fraction predominantly preceded the spiking response to the Gabor stimuli, the absolute response latency was likely over-estimated due to the fixed temporal phase at which each Gabor stimulus was presented. In order to perform a more precise estimate of the response latency of burst spikes, we additionally assessed burst vs tonic spike tuning latency by presenting white noise stimuli in a subset of experiments performed using acute recordings with Neuropixel 1.0 probes (Fig. 2.6A). To do so, we first extracted the receptive field of each neuron by calculating tonic and burst spike triggered averages (tSTAs and bSTAs, respectively). Of the 561 neurons we analyzed in these experiments (n=8 mice), the majority (391 neurons) exhibited a significant tuning to white noise stimuli. In individual neurons with both a significant tSTA and bSTA (n=125 neurons), we assessed the white noise response latency by finding the first temporal lag at which a significant tuning was present in the tSTA and bSTA. In contrast to the Gabor driven response latencies, the white noise response latencies were extremely rapid (mean response latency, 68 ms), consistent with previous estimates of the initial LGN driven response latency in V1 (Niell & Stryker, 2008). Interestingly, the tuning latency of burst and tonic spikes (bSTA and tSTA) were not significantly different for bSTA tuned neurons (Fig. 2.6B), again indicating that the initial visual response of V1 neurons is comprised of a high proportion of burst spikes. We additionally 40 compared the white noise response latencies of neurons with a significant bSTA to those without (n=266 only tSTA tuned neurons), and found that bSTA tuned neurons had a significantly faster response latency (p<0.0001, Fig. 2.6C). These results were further inconsistent with the idea that the early bursting responses we observed were due to intra-cortical recurrence or feedback. 2.2.3 Inverse stimuli do not drive increased bursting Having characterized the burst response to classical stimuli, we next tested whether feedback driven visual responses would preferentially recruit burst spiking. We mapped the receptive fields using both Gabor and inverse stimuli, and presented full-field drifting gratings to calculate inverse tuning as in previous literature (Fig. 2.8A) (Keller et al., 2020). Similar to previous reports, we find that neurons in L2/3 and L5 exhibit an inverse receptive field (n=178 neurons, 7 mice), where a masked full field stimuli centered on each neurons classical receptive field produces a spiking response larger than that to a full field grating. These extra-classical responses showed a delayed onset relative to feedforward stimuli, consistent with their recruitment by a feedback process (Fig. 2.8B,C,E). We next compared the burst fraction driven by Gabor and inverse stimuli. Contrary to our expectations, responses to inverse stimuli did not exhibit a higher burst fraction than classical stimuli, even when restricting analysis to the late visual evoked component of the response(Fig. 2.8D,F,G). Instead, the initial feedforward response to classical stimuli recruited a higher fraction of burst spikes than all other conditions (Fig. 2.8). These results suggest that, despite the mechanistic capability of inverse stimuli to drive 41 active dendritic events in the apical tuft, dendritically evoked bursts do not contribute to the generation of inverse tuning. We next investigated whether the differences in stimuli evoked burst fractions could be explained by differences in behavioral state. Previous studies have shown that orofacial movements can explain a large fraction of neural activity of mouse V1 (Stringer et al., 2019), and correlate with metrics of general arousal (Nestvogel & McCormick, 2022; Salkoff et al., 2020). To control for these effects, we used FaceMap to extract orofacial movements from concurrent video recordings of the mouse during visual stimuli experiments, and restricted our analyses to only trials in which the mouse was still (see Methods). However, the differences between Gabor and inverse evoked burst fraction remained the same, indicating that differences in behavioral state do not drive the observed effects (Fig. 2.9A,B). 2.2.4 Stimulus driven bursting is invariant to preceding stimuli The relative absence of top-down driven burst spiking, and the prevalence of burst spikes in the onset response to bottom-up stimuli instead were both inconsistent with our initial two-compartment model. We reasoned that the lack of matched bottom-up inputs during the inverse stimuli may have left the top-down inputs unable to independently drive burst spiking. In this case, the model would predict that the coincidence of bottom-up and top-down inputs would produce increased bursting, consistent with theories that the extra-classical response represents a form of visual prediction. In order to test 42 this, we presented a new stimuli battery in which retinotopically matched Gabor and inverse stimuli were presented sequentially (Fig. 2.10A,B). As a control, we also presented a Gabor preceded by a full-field grating. Here, we reasoned that the full-field gratings would recruit increased surround suppression compared to the inverse response, and as such would produce less apical dendritic drive. However, if the inverse stimuli were recruiting top-down inputs to the apical dendrites to serve as a predictive signal, then the presentation of a Gabor stimuli immediately following its retinotopically matched inverse stimuli (or visa versa) should generate a coincidence of top-down and bottom-up signals which evoke a larger burst fraction response. For these analyses, we compared the firing rate and burst fraction elicited by Gabor and inverse stimuli with different preceding stimuli without subtracting the baseline due to the previous stimuli, since we reasoned that the absolute amplitude of the response should differ. We found that the firing rate elicited by the Gabor stimuli did not change depending on whether they were preceded by inverse or full-field stimuli (n=130 neurons, 4 mice) Fig. 2.10C,D). Contrary to our initial hypothesis, the Gabor-evoked burst fraction did not increase when the Gabor stimuli was preceded by inverse, and instead exhibited a modest decrease compared to when it was preceded by a full-field stimulus (p=0.035, Fig. 2.10E,F). Furthermore, we did not find an amplification of the burst fraction when the inverse stimuli was preceded by its RF matched Gabor compared to when it was preceded by gray screen (Fig. 2.10G,H). These results were inconsistent with the idea that inverse stimuli preferentially recruit top-down signals to the apical dendrites to serve as a predictive coincidence detector. 43 2.3 Discussion Our results provide evidence that burst spiking in mouse V1 is not uniquely, nor preferentially, driven by top-down stimuli, but is instead robustly recruited by classical bottom-up stimuli (Gabors). Notably, Gabor-evoked increases in the burst fraction were largely recruited in the initial feedforward component of the visual response. These rapid onset times are more consistent with feedforward drive from the LGN or local recurrence than with delayed feedback from higher cortical regions. These findings are inconsistent with previous theories in which the burst fraction selectively encode top-down information, and the prediction that cortical burst spiking is dependent on feedback inputs to the apical tuft. Importantly, our findings are not easily explained by models that predict burst generation based solely on firing rate. Burst fractions were highest at stimulus onset, preceding peak firing rates, and neither the latency nor magnitude of these visually evoked bursts could be predicted by a Poisson model. Furthermore, the burst fraction and firing rate profiles across laminar depth differ substantially, suggesting that burst spike generation is decoupled from firing rate in a cell-type dependent manner. Taken together, this suggests that the burst spikes we measured reflect an intrinsic biophysical or circuit-level computation beyond simple rate coding. Interestingly, our results are consistent with a previous study suggesting that sensory thalamic inputs can directly recruit active calcium channels along the apical trunk to generate early sensory evoked bursting (Bast et al., 2023). Another possibility is that the spatio-temporal pattern of bottom-up inputs 44 within in vivo neural circuits generates enough somatic depolarization to recruit active apical dendritic voltage channels through the backpropagation of somatic action potentials, again allowing the rapid recruitment of burst spiking without longer-latency feedback signals. Such a regime would allow bottom-up driven bursts to act as a nonlinear thresholding mechanism, encoding the rapid temporal derivative, or slope, of incoming excitation. Despite previous evidence that inverse stimuli drive feedback inputs to L1 of V1, and that these L1 inputs can engage active dendritic events, we did not find that inverse stimuli drove a higher burst fraction than our bottom-up control stimuli. These results provide direct in vivo evidence contradicting previous predictions of top-down encoding via burst spikes. The lack of inverse driven increase in burst fraction (relative to Gabor) could have several explanations. One, despite the strong ex vivo results tying apical dendritic inputs to burst spiking, the concurrent recruitment of balanced circuit level inhibition could prevent apical tuft inputs from generating somatic burst spikes. Second, it is possible that inverse stimuli are not causally dependent on feedback to the apical tuft. Previous studies have shown that a substantial amount of V2 feedback arrives peri-somatically, both for L2/3 and L5 neurons. Whether these deeper-layer inputs carry the requisite information to generate inverse tuning and are causally involved remains an open question. We additionally tested whether the sequential presentation of bottom-up and top-down stimuli would cause an amplification of the burst fraction, as coincidence detection or visual predictive coding models might predict. Again contrary to these models, our data showed that the sequential presentation 45 of top-down and bottom-up stimuli (or visa versa) did not produce further increases to the burst fraction. Instead, we found that the response to Gabor stimuli was consistently recruited regardless of its preceding stimuli. These data suggest that inverse stimuli do not form a type of visual prediction which amplifies future presentations of classical stimuli (or visa versa). Taken together, these findings suggest a need to revise current theories of burst coding in the cortex. Rather than acting as a dedicated channel for top-down information, burst spiking may play a broader role in encoding salient or strongly driving features of the sensory environment - functions typically associated with bottom-up pathways. This reframing has implications for both computational models of cortical processing and our understanding of how dendritic integration contributes to perception and behavior. 46 Figure 2.1: Detection and characterization of extracellular burst spikes. A. Extracellular electrophysiological recordings were performed in the primary visual cortex of awake, head-fixed mice using both acute Neuropixels 1.0 probes and chronically implanted Neuropixels 2.0 probes. B. Burst criteria and example burst and single spikes (denoted by blue and red arrows, respectively) from an example L5 pyramidal neuron. C. Example bursts recorded from neurons in L2/3 and L5. D. Firing rate and burst fraction as a function of distance from the L5 MUA peak. Putative L5a neurons (centered on the MUA peak) exhibit the highest firing rate, whereas burst fraction peaks are centered on L2/3 and just below the MUA peak (putative L5b). 47 Figure 2.2: Spike sorting and manual curation. A. Raw Neuropixel data was time-shifted (t-shift) to correct for asynchronous temporal sampling across channels and artifact-corrected using a common average reference (CAR). Extracellular spike sorting was performed to identify putative single units using Kilosort 2.5 and 4. After automated spike sorting, bursty units can be erroneously split into separate units due to characteristic differences in burst amplitude and shape. B. Following spike sorting, extensive manual curation was performed to re-merge over-split bursty units. Left panel: For each unit, cross-correlograms (CCGs) were examined for all units on adjacent channels, with a particular focus on identifying heavily asymmetric CCGs with minimal refractory period violations. Right panel: Putative mis-splits identified via CCG profiles were further inspected for waveform similarities, and raw trace inspection was used to confirm that mis-splits units spiked consecutively during high-frequency bouts of action potentials. 48 Figure 2.3: Extracellular spike bursts decrement in amplitude across time and space. A. Averaged waveforms at the peak channel (putative somatic location) for an example L5 neuron. Waveforms are grouped by tonic vs burst spikes and the position of each spike within a burst. Somatic waveform amplitude decreases across the duration of a burst. B. Spike-triggered LFP (sLFP) reveals action potential backpropagation across the cortical depth. Scale denotes voltage clipped between ±8 µV. C. Spike-triggered CSD (sCSD) profiles of the field potentials shown in B. Markers indicate significant sources, showing decreasing action potential backpropagation over the course of a burst. Scale denotes boot-strapped z-score clipped between ±4. 49 Figure 2.4: Gabor stimuli drive rapid recruitment of burst spiking. A. Ex- perimental design. Extracellular Neuropixel recording in V1 while presenting Gabor stimuli. B. Raster plot showing burst and tonic spike firing in response to a Gabor stimuli presented in the receptive field of an example neuron. Inset shows long trains of burst spikes occurring at the initial onset of the visual response. C. Peri-stimulus time histogram of the same neuron as in B. D. PSTH showing the response of all spikes, and the separation of burst and tonic spikes across the population (n=669 neurons, 7 mice). Dashed lines indicate the mean peak latency of each spike type across the population. E. Firing rate and burst fraction of an example neuron. F. Firing rate and burst fraction across the population (n=461 neurons, 7 mice). G. Half-peak latencies of burst spikes compared to all spikes for all neurons which exhibit an early bursting response (significant burst fraction peak within 150 ms of the Gabor onset, n=303/461 neurons). The burst spike response occurs significantly earlier across the population (p<0.001). 50 Figure 2.5: Unsupervised detection of early bursting neurons. A. K-means clustering of burst vs all-spike response latency for all neurons with a significant bursting response. Inset: selection of clusters based on within-cluster sum of squares. B. Same analysis of burst and all spike response latency as in Fig. 2.4G, but for early bursting neurons identified using k-means clustering instead of a latency threshold (cluster 1, n=300/461 neurons). Burst spikes respond at a significantly shorter latency than all spikes (p<0.0001). C. Burst - all spike latency for all neurons. The majority of neurons are in cluster 1, which exhibits a faster burst latency (significantly left shifted relative to 0). 51 Figure 2.6: Tuning of bursts and tonic spikes to white noise stimuli. A. Full- field gaussian white noise stimuli are presented for 32 ms with no inter-stim interval. B. Example tonic and burst spike triggered averages (tSTA and bSTA, respectively) for a single neuron. C. In single neurons with a significant bSTA (n=125, 7 mice), the tuning latency of burst and tonic spikes was not significantly different. D. Neurons with a significantly tuned bSTA exhibit a faster tuning onset than those without a significant burst spike tuning (n=266 neurons, 7 mice). The rapid tuning latency of burst spikes, even to spatio- temporally random noise stimuli, is inconsistent with their dependence on cortical feedback. 52 Figure 2.7: Poisson firing rate model. A. Example Poisson model fitting to a single neuron. Poisson spike trains are generated to match the firing rate of the Gabor response, and then the same burst spike criteria is applied to the Poisson generated spike trains. Top: Raster plot of tonic spikes (orange) and burst spikes (purple) for an example L5 neuron. Middle: Raster plot of Poisson simulated spikes and bursts for the sample example neuron. Bottom: Same data as in the top and middle panels, but zoomed in on the initial onset window (50-200ms). Poisson generated bursts lack the trial-to-trial structure and early onset latency of the actual neuron. B. Actual vs Poisson predicted burst rate for the example neuron in A. Top: Actual firing rate and burst rate plotted against the burst rate predicted by the Poisson model. Bottom: Model error (actual - predicted burst rate) across time. The Poisson model fails to capture the early onset of the actual burst rate, and then over-predicts over the course of the response. C. Same as B, but plotted using the burst fraction instead of the burst rate. The burst fraction is left shifted compared to the burst rate due to periods during which few spikes are fired before the response onset, but shows the same general trend as the burst rate. D. Actual and Poisson predicted burst rates across the population (n=461 neurons). Although the Poisson model approximates burst rate at population level, the mean absolute error (model error, red) indicates an extremely poor fit for individual neurons. Model error peaks early, and remains high throughout the response window. E. Same as in D, but for the population burst fraction. 53 Figure 2.8: Inverse stimuli do not evoke higher burst firing than Gabors. A. Visual stimuli battery of Gabor, inverse, and full field drifting gratings. Matching Gabor and inverse receptive fields for an example neuron. B. Example tuning of a single pyramidal neuron with receptive fields shown in A. Top and middle: Raster plot and PSTH of Gabor, inverse, and full-field responses. The response to inverse stimuli is delayed in onset compared to Gabor, and significantly greater than the full-field response. Bottom: Burst fraction driven by different grating stimuli. C. Population level firing rate (n=178 neurons, 7 mice). D. Population level burst fraction. E. Firing rate for different visual stimuli separate by short and long latency bins (short and long latency responses are defined as 0-150 ms and 150-500 ms, respectively). F. Burst fraction depicted in the same format as E. The short latency response to Gabor stimuli evokes a larger burst fraction than all other conditions (p<0.0001). Inverse stimuli do not drive a higher burst fraction than Gabor in any condition. F. Same data as in F, but with the pairwise difference between Gabor and inverse stimuli driven burst fractions for each neuron. Gabor stimuli evoke a higher burst fraction in the short (p<0.0001) but not long latency conditions (p=0.083). 54 Figure 2.9: Differences in whisking do not explain lack of inverse stimuli driven bursting. A. Population firing rate for inverse and Gabor stimuli (n=87, n=89, respectively) during trials in which animals were not whisking. B. Burst fraction for inverse and Gabor stimuli during the same stationary trials as in A. Gabor stimuli drive equal or more bursting than inverse stimuli at all latencies. 55 Figure 2.10: Sequential presentation of Gabor, inverse, and full-field stimuli. A. Visual stimuli battery. Inverse stimuli are presented directly following retinotopically matched Gabor stimuli, and visa versa. As a control condition, Gabor stimuli are presented directly following full-field stimuli. B. Population level firing rate and burst fraction across the three stimuli conditions. Dotted line indicates when the second visual stimuli in the sequence was presented. C. Population firing rate across the three conditions, synchronized to the onset of the Gabor stimuli (n=130 neurons, 4 mice). D. Mean normalized firing rate between 75-275 ms after Gabor presentation. Mean firing rate does not change depending on whether the Gabor was preceded by an inverse of full-field stimuli. E. Population burst fraction locked to the onset of Gabor stimuli. F. Burst fraction quantified for the same time period as in D. The burst fraction is not higher when Gabor stimuli are preceded by inverse compared to any condition. Instead, Gabors preceded by full-field stimuli recruit a slightly larger burst fraction than those preceded by inverse stimuli (p=0.035) G. Population burst fraction aligned to the onset of inverse stimuli. H. Burst fraction quantification of the data shown in G. Inverse stimuli do not recruit different burst fractions when preceded by Gabor vs full-field stimuli. 56 Chapter 3 Apical dendritic contributions to burst spiking and top-down visual processing in vivo 3.1 Introduction Although decades of ex vivo physiology have shown that apical dendrites can powerfully mediate burst spiking, the causal contribution of apical dendrites to burst spike generation in awake animals remains unknown. In vivo neural circuits differ substantially in the nature of their synaptic activity, which could dramatically alter the input-output functions of single neurons. For instance, top-down inputs to the apical tuft could be accompanied by circuit-mediated changes in excitatory-inhibitory balance, resulting in a lower burst rate despite changes in tuning (Anderson et al., 2013). 57 Independent from their role in driving burst spiking, it is also unclear the extent to what excitatory inputs to the apical tuft are responsible for feedback driven visual tuning. Compared to top-down inputs from LP or ACC, feedback projections from V2 are more heavily innervated in the deeper layers of V1, and functional mapping has shown that LM inputs strongly target the peri-somatic dendrites of both L2/3 and L5 neurons (Fig. 1.3) (Shen et al., 2022; Young et al., 2021). Although previous studies have largely focused on the functional role of L1 inputs (Fişek et al., 2023; Keller et al., 2020; Marques, Nguyen, et al., 2018), it remains possible that these deep layer and/or peri-somatic projections are heavily involved in visual feedback tuning. 3.2 Results We therefore combined extracellular electrophysiology with targeted optogenetic perturbations to directly investigate the causal role of apical dendrites in both burst spike generation and top-down visual tuning. 3.2.1 Apical dendritic activity is not required for burst firing We first performed graded inhibition of the apical tuft vs soma via a direct optogenetic inhibition of L5 pyramidal neurons. To do so, we used a Cre- dependent viral approach to express a halorhodopsin variant (eNphR3.0) in a semi-sparse population of L5 (Fig. 3.1A) and performed acute Neuropixels recordings (n=40 neurons, 4 mice). Similar to previous work, we delivered 58 titrated optogenetic inhibition using a fiber optic positioned at the surface of the brain (Ranganathan et al., 2018). Due to the exponential drop-off of light intensity as a function of distance and diffusion through cortical tissue, we rea- soned that this optogenetic inhibition would preferentially inactivate the apical tuft, particularly at lower LED powers. We validated that electrophysiological recordings were co-localized with the opsin expression and selectively targeted to L5 neurons (n=4 mice, Fig. 3.1A,B). Strikingly, the burst fraction during optogenetic inhibition remained the same as control even at the maximal laser power, which corresponded to a 40% reduction in firing rate (Fig. 3.1C,D). As a complementary approach, we next performed a selective inactivation of apical tuft activity via activation of layer 1 neuron-derived neurotrophic factor positive (NDNF) interneurons (Fig. 3.1E). NDNF interneurons represent a distinct subclass of interneurons which are localized to L1 and form direct inhibitory synapses with the apical tuft dendrites of L2/3 and L5 pyramidal neurons. NDNF driven inhibition is thought to be mediated by a combination of GABAA and GABAB release, allowing for long-timescale dendritic inhibi- tion(Cohen-Kashi Malina et al., 2021). NDNF interneurons additionally target parvalbumin-positive (PV) interneurons in deeper cortical layers. Thus, NDNF interneurons provide long lasting apical dendritic inhibition paired with somatic disinhibition, and are thought to enable a circuit level switch from top-down to bottom-up signal processing. In order to test the effects of NDNF mediated inhibition on cortical bursting, we expressed the excitatory opsin ChrMine in a genetically targeted population of NDNF interneurons, and performed chronic electrophysiology using Neuropixels 2.0 probes. We found that transient 59 Figure 3.1: Optogenetic inhibition of the apical tuft. A. Experimental design for L5 inhibition experiments. ENpHR3 is expressed in a subset of layer 5 pyramidal neurons targeted using rbp4-Cre mice (n=40 neurons, 4 mice). Example histology shows co-localization of the recording electrode and viral expression. B. Optogenetic inhibition is restricted to layer 5 pyramidal neurons. C. Firing rate inhibition caused by optogenetic perturbation. D. Burst fraction does not differ between opto and control conditions at the maximal LED intensity. E. Experimental design for NDNF-interneuron mediated dendritic inhibition. ChrMine is expressed in L1 NDNF neurons via NDNF-Cre mice. F. Opto/control for both firing rate and burst fraction. Burst fraction and firing rate modulation are not significantly different . G. Burst fraction during control and opto conditions. H. Distribution of burst fraction differences between opto and control conditions. activation of NDNF inteneurons significantly decreased the burst fraction (134 neurons, 3 mice, Fig. 3.1G,H). However, the magnitude of this inhibition was fairly modest (20% decrease in burst fraction), and not significantly different from the decrease in firing rate (Fig. 3.1F). Contrary to the prevailing model of burst spike generation, these results showed that apical tuft engagement 60 was not necessary for the generation of the majority of the burst spikes we observed in vivo. 3.2.2 Apical tuft dendrites are not required for inverse tuning Despite the relatively small contribution of apical tuft activity to burst spike generation, we reasoned that it was still possible that tuft inputs were selectively involved in inverse stimuli processing. As such, we performed simultaneous visual stimuli and apical tuft inhibition via NDNF interneuron activation, and analyzed the effects of L1 inhibition on both the overall firing rate response and burst fraction driven by classical and inverse stimuli (n=130 Gabor tuned neurons, 48 inverse tuned neurons, Fig. 3.2A-D). We found no differences between the degree of firing rate or burst fraction inhibition between Gabor and inverse stimuli, strongly suggesting that inverse stimuli are not preferentially processed by apical tuft inputs (Fig. 3.2B,D). Interestingly, inverse-tuned neurons were significantly inhibited in their firing rate, but showed no burst fraction inhibition (Fig. 3.2D). Finally, we tested whether the size tuning curves to Gabor and inverse stimuli were differentially affected by L1 inhibition. Previous work has shown that direct inhibition of higher order visual areas preferentially abolishes the size tuning to inverse, but not Gabor stimuli (Keller et al., 2020). However, we found that selective apical tuft inhibition did not affect the size tuning for either Gabor or inverse stimuli, and instead produced a uniform decrease in response amplitude (Fig. 3.2E). 61 Figure 3.2: Effect of apical tuft inhibition on Gabor and inverse responses. A. Population firing rate during control and optogenetic inhibition during Gabor (left panel) and inverse (right panel) stimuli presentation. Responses are normalized to the maximum firing rate during Gabor presentation for each neuron. B. Opto-control firing rate for both Gabor and inverse conditions. L1 inactivation does not differentially Gabor and inverse firing rates across time. C. Normalized burst fraction across the population in response to Gabor and inverse stimuli (left and right panels, respectively). D. Burst fraction difference (opto-control) for Gabor and inverse. E. Top: Tuning to increasing sizes of Gabor stimuli centered at the RF location for opto and control conditions. Bottom: Size tuning for RF centered inverse stimuli. Similar to Gabors, the size tuning to inverse stimuli is not abolished by L1 inactivation. 3.3 Discussion In these experiments, we performed the first causal test of whether the apical tuft is required for in vivo burst spiking or feedback-driven visual tuning. Contrary to prevailing models derived from ex vivo and computational studies, our results reveal that apical tuft engagement is not a necessary condition for the generation of burst spikes in awake animals. Both direct optogenetic and 62 interneuron-mediated inhibition of the apical dendrites failed to abolish the majority of burst firing, despite robust suppression of overall neuronal activity. This dissociation between firing rate and burst fraction suggests that burst generation can occur through mechanisms that do not rely on apical tuft input, and perhaps more surprisingly, may be predominantly governed by peri-somatic processes in the intact brain. These findings challenge a widely held assumption that burst firing is a hallmark of apical dendrite-mediated top-down integration. While some component of burst activity may still be supported by dendritic spikes in the tuft, our data indicate that this is not the dominant mode in awake, non- anesthetized animals. One interpretation is that the tightly regulated excitatory- inhibitory (EI) balance in vivo, particularly in L1, actively suppresses the conditions required for dendritic spike initiation. This aligns with prior reports suggesting that apical dendrites may serve more modulatory or subthreshold integrative functions, rather than acting as discrete burst generators under passive behavioral states. Furthermore, our experiments show that top-down visual tuning, such as size tuning to inverse stimuli, is not selectively abolished by inhibition of the apical tuft. Despite anatomical projections from higher-order areas like LP or V2 into L1, and prior work implicating these projections in feedback modulation, we found no evidence that tuft input is necessary for stimulus-specific tuning effects. Even in superficial layers, top-down inputs from V2 show substantial peri-somatic targeting to L2/3 pyramidal neurons in V1, suggesting that L1 inputs need not be integrated by the apical tuft. Another possibility is that 63 feedback-driven tuning may be mediated predominantly through alternative pathways—specifically, projections to deeper layers of V1, which are often overlooked in favor of L1 inputs. This hypothesis is supported by anatomical studies showing that V2 and other feedback sources robustly innervate layers beyond L1, including L5 and L6. Importantly, our findings do not entirely rule out the functional significance of apical tuft activity. It is possible that under specific behavioral states, such as multi-sensory processing or active learning, the apical tuft could play a more prominent role in modulating output or biasing sensory processing. Future experiments could explore these possibilities using more complex behavioral paradigms paired with temporally precise dendritic inhibition. 64 Chapter 4 Final Conclusions The ability to integrate sensory information from the outside world with inter- nal predictions, contextual knowledge, and goals is a fundamental computation performed by the neocortex. Decades of anatomical and physiological research have suggested a framework for how this computation could be implemented at the single cell and circuit level, through the combination of spatially segre- gated inputs and dendritic biophysics. This framework, in turn, has informed theoretical models proposing algorithmic-level explanations for how the brain could integrate different streams of information to perform both computation and learning. At present, these models have exposed key predictions for how bottom-up and top-down information are processed at the single cell level which have not yet been validated in vivo. Here, we bridge this gap by specifically investigating the role of burst and tonic spikes in encoding top-down and bottom-up information, and the causal contributions of apical tuft dendrites to burst spiking and top-down visual processing. 65 4.1 Visual tuning of cortical burst and tonic spikes In Chapter 2, we found that burst spiking is not preferentially recruited by top-down visual stimuli. Instead, the highest burst fraction that we observed was in the short latency response to a classical receptive-field stimuli thought to be mediated primarily by bottom-up inputs. When we sequentially presented top-down and bottom-up stimuli, we did not observe supra-linear amplification of the responses that would be predicted by theories of dendritic coincidence detection. In these experiments, we found that the Gabor driven firing and burst fraction were not affected by the preceding stimuli. Taken together, results strongly contradict the idea that the burst fraction preferentially encodes higher- order inputs to the apical tuft. Instead, the strong feedforward recruitment of high-frequency firing suggests that burst generation in vivo is regulated by strong peri-somatic drive. Interestingly, many of the extracellular burst spikes that we recorded showed decrements in both spike amplitude and signal backpropagation, which are both consistent with the presence of the regenerative currents characteristic of apical-dendrite mediated bursting ex vivo. These findings pose a considerable biological constraint for models which rely on burst spiking to encode top-down driven activity for either computational or learning related functions. The lack of inverse stimuli driven burst spiking posed a further question - did we fail to observe these events because (1) the apical tuft is not strongly coupled to burst spiking, or (2) the inverse stimuli did not strongly drive tuft activation? 66 4.2 Causal investigation of the apical tuft To answer these questions, we directly tested the causal involvement of apical dendrites in Chapter 3. Contrary to the ex vivo evidence linking apical dendritic engagement to burst firing, we found that inhibition of apical dendrites—either via direct optogenetic silencing of L5 apical tufts or NDNF interneurons L1 inhibition—had a minimal effect on the burst fraction. This suggests that apical dendrites are not essential for generating the majority of the burst spikes we observed in vivo, consistent with our finding that they are strongly recruited by bottom-up stimuli. This may be explained by more cell-intrinsic mechanisms of burst firing (Izhikevich, 2003; Kepecs et al., 2002). Our perturbation experiments also showed that the burst fraction is highly robust to decreases in the overall firing rate. In particular, it was striking that the burst fraction did not change at any laser titration during our halorhodpsin-mediated inhibition of L5 neurons - even when the firing rate was inhibited by 40%. Although inverse did not drive a higher burst fraction than Gabor stimuli, we reasoned that it was still possible that their tuning was preferentially mediated by the apical tuft. This could be the case if the inverse driven inputs simultaneously recruited feedback inhibition. However, we found that optogenetic activation of NDNF-interneurons did not preferentially affect either the firing rate or burst fraction for inverse versus Gabor stimuli. Unlike the direct inhibition of V2 reported in previous work (Keller et al., 2020), the targeted inhibition of layer 1 in V1 did not cause a selective abolishment of inverse tuning to stimuli size. Instead, the uniform decreases across both Gabor 67 and inverse stimuli in response to apical dendritic inhibition suggests that inverse tuning is not driven by feedback projections into L1. Thus, our findings challenge the conventional view that top-down feedback relies predominantly on L1 and apical tuft signaling, and instead suggest a much more prominent role for deeper layer feedback projections. 4.3 A revised framework for top-down and bottom- up integration Taken together, these results provide new insights into the mechanisms of burst spike generation and top-down visual processing in vivo. They suggest that, contrary to prevailing models, apical dendrites are not critical for high- frequency burst firing, and that feedback modulation of visual processing is not exclusive to the canonical L1 pathway. At the cellular level, the strong bottom-up recruitment of burst firing supports previous theories that bursts serve to enhance the transmission of sensory information (lisman_bursts_1997; Kepecs et al., 2002). Instead of relying on apical dendritic inputs, these bursts may be strongly driven by mechanisms such as spatio-temporally precise inputs or more proximal-driven activation of the apical trunk (Bast et al., 2023). These results suggest that the burst fraction does not uniquely encode a multiplexed representation of top-down and bottom-up inputs. However, it remains possible that bursts could be strongly recruited by top-down inputs or learning signals under different experimental conditions. One caveat to these experiments is that it is unclear 68 whether the extracellular recording technique utilized in these experiments can capture action potentials when neurons enter the depolarization block associated with large intracellular plateau potentials. As a result, these experiments may be unable to sample the intracellular plateau potential driven bursts most strongly associated with plasticity. In the hippocampus, for instance, high-frequency bursts are ubiquitous and contain different information than tonic spikes (Lowet et al., 2023), and yet are clearly distinct from the plateau potentials which cause one-shot remapping of place cell tuning (bittner_conjunctive_2015; Bittner et al., 2017). Do tonic spikes, bursts, and plateaus represent three distinct output modes that single neurons can differentially leverage? Future work, including ground-truth validation of extracellular techniques, will be critical for a more nuanced understanding of these phenomena. At the circuit level, these findings point towards a prominent role for deep layer feedback inputs from V2 to V1 in generating extra-classical visual tuning. Despite the fact that these projections represent the majority of V2 to V1 feedback, they remain critically under-characterized. It remains unknown whether their functional tuning or progenitor neurons overlap with those of the superficial V2 to V1 projections. Compared to higher order visual cortices, feedback projections from many motor and prefrontal cortices and high-order thalamus to V1 are more precisely restricted to L1. This raises the possibility that different long range inputs may exploit functionally different laminar targeting despite having a qualitatively similar organization (i.e. targeting to L1 and deeper layers). The significance of this differentiation between hierarchical feedback within the same sensory modality and cross-modal projections offers 69 a promising direction for further investigation. That the apical dendrites in L1 do not process passive visual information may actually represent a critical feature for more complex functions of top-down processing, although it remains an open question what their role is in vivo. Here, laminar specific inactivation approaches such as the direct apical and NDNF-activation strategies utilized in this work present powerful tools for future causal studies of apical dendritic computation. 70 Appendix A Methods A.1 Animals All experiments were compliant with guidance and regulation from the NIH and the Massachusetts Institute of Technology Committee on Animal Care. We used C57Bl6 mice for acute Neuropixel 1.0 and chronic Neuropixel 2.0 experiments, Rbp4-Cre heterozygous mice for all acute Neuropixel 1.0 experiments with L5 optogenetic inactivations, and NDNF-Cre homo/heterozygous for all chronic Neuropixel 2.0 experiments with NDNF interneuron mediated suppression of L1. Male and female mice were used for all experiments, and were 8-12 weeks of age at the beginning of experiments. All mice were maintained on a 12-hour light/dark cycle in a temperature and humidity-controlled room with ad libitum food and water access. 71 A.2 Surgery Mice were anesthetized using 4% isoflurane in 1 l/min oxygen and subsequently maintained using 1-2% isoflurane over the course of the surgery. Following anesthetization, mice were placed on a closed-loop heating pad to maintain physiological body temperature, and eye ointment (Bepanthen, Bayer) was applied to prevent the eyes from drying. We delivered Dexamethasone (4 mg/kg) and Buprenorphine (extended release, 0.5 mg/kg) subcutaneously to provide post-surgical analgesia. The scalp was first shaved using hair clippers, and then remaining hairs were removed using a depilatory cream (Nair). The position of the head was leveled along the anterior-posterior (AP) and mediolateral (ML) axes. The exposed skin and surrounding area were cleaned using alternating passes of iodine solution and ethanol, and the skin over the skull was excised. We then marked the target location over left monocular V1 (3.6 mm AP, 2.7 mm ML from bregma), which was the same for all experiments. Additional surgical procedures were performed for acute and chronic electrophysiology experiments as follows. A.2.1 Acute procedures For acute recordings without optogenetic perturbations, a custom metal head- plate was fixed to the skull using clear dental cement (C&B Metabond, Parkell), and mice were transferred to post-operative care. For L5 inhibition experiments in Rbp4-Cre mice, a small burr hole (2-300 µm) was made at the target site, and a glass pipette containing pAAV-hSyn-eNpHR 3.0-EYFP (Addgene, catalog # 72 26972-AAV5) at a dilution of 1x1012 vg/ml was lowered to 575 µm below the cortical surface. We waited at least 5 minutes for the pipette to settle , and then pressure injected 100 nL of virus at a rate of 20 nL/min. The virus was allowed to diffuse for at least 5 minutes following completion of the injection before the pipette was removed. The burr hole was covered using a clear dental cement (Ivoclar Vivadent Tetric EvoFlow), and a custom metal headplate was then fixed to the skull using opaque cement (C&B Metabond, Parkell). Mice were subcutaneously injected with up to 25 mL/kg of Ringer’s solution at the end of the procedure. Recordings were performed 4-6 weeks post-surgery. Prior to the recording experiment, we habituated mice to human handling and head fixation on a treadmill for at least 3 consecutive days, and progressively increased head fixation duration up to 45 minutes. On the day of recording, we performed a short second surgery to expose the cortex. For these surgeries, mice were induced at 2% isoflurane, and maintained at 1-2% isoflurane over the course of the surgery. Body temperature was maintained using a heating pad as described above, and a protective eye ointment was applied immediately following anesthetization (Bepanthen, Bayer). We removed the clear cement over the target recording site, and performed either a burr hole or small craniotomy to expose the cortical surface. The dura was left intact. In order to maintain a saline well during recording, and to limit off-target light leakage, a small circular well was made with dental cement surrounding the exposed cortex. This well was filled with a mixture of low melting point agarose (Durand et al., 2023), followed by a layer of silicon elastiomer (Kwik-Cast, WPI). We applied thin stripes of dental 73 cement over the well to provide additional mechanical protection. Mice were subcutaneously injected with Ringer’s solution to maintain hydration, then removed from anesthesia and allowed to recover in a heated cage for 3-4 hours before recordings. A.2.2 Chronic procedures For chronic Neuropixel implants, we drilled a 2 mm craniotomy centered over the same V1 target location described above (3.6 mm AP, 2.7 mm ML from bregma). The dura was left intact, and surgical foam (Surgifoam) was used to maintain cortical hydration. We drilled a small burr hole over the cerebellum of the opposite hemisphere, and implanted a sharpened silver wire for grounding. A custom-printed plastic base plate was affixed to the skull using dental cement (Ivoclar Vivadent Tetric EvoFlow). We then centered a custom-fabricated microdrive (see A.x.x Silicon probe preparation) with a Neuropixel 2.0 probe over the V1 target site. The four shanks of the probe were angled at a 45°angle relative to the midline. Small adjustments in the angle and placement of the four shanks (+-100 µm AP, ML) were made to avoid blood vessels. To accommodate for these adjustments to the four-shank placement, we positioned the probe before performing viral injections for optogenetic experiments to ensure maximal co-localization between the injection and recording sites. For these experiments, we took reference images of each shank at the cortical surface. We then removed the probe, and used a glass pipette to inject 100 nL of pAAV-nEF-Con/Foff 2.0-ChRmine-oScarlet (Addgene, catalog # 137161- AAV8) at each of the four electrode shank positions. Injections were performed 74 at a depth of 175 µm below the cortical surface, and injected at a rate of 20 nL/s. To prevent backflow, we waited at least 5 minutes before and after each injection. We then repositioned the probe microdrive to the same location, and performed the same implant procedure for NDNF-Cre and BL6 mice. Probes were inserted to a depth of 3.5-4 mm at a speed of 10 µm/s using a programmable micromanipulator (Scientifica). Following probe insertion, we attached the microdrive to the skull using dental cement, and filled the craniotomy using Dura-gel (Cambridge NeuroTech). The implanted ground wire was soldered to the previously prepared grounding/reference wire on the Neuropixel probe. We installed an electrically shielded cage to the base plate, and used a self-adhesive bandage to enclose the implant. The design of the cage and base plate system was adapted from Vöröslakos et al., 2021. We delivered multiple injections of Ringers solution (2-3 deliveries) subcutaneously over the course of the surgery to maintain hydration, up to a final dose of 25 mL/kg. Mice were allowed to recover for at least three days post-operation, and were habituated to human handling and head fixation for at least 3 days prior to the beginning of experiments. A.2.3 Silicon probe preparation To minimize damage to cortical tissue during probe insertion, we sharpened all Neuropixel probes (1.0 and 2.0) using a custom-built sharpening device, follow- ing a publicly available protocol (https://github.com/billkarsh/SpikeGLX/blob/gh- pages/Support/NPix_sharpening.docx). We soldered a silver ground wire to the ground and reference pins for all probes. Probes were coated in either DiI or 75 DiO lipophilic dyes (Invitrogen, ThermoFisher) immediately prior to insertion or implantation to allow for histological reconstruction of the recording site. Acute recordings were performed using Neuropixel 1.0 probes. Following each acute recording, probes were soaked in overnight in 1% Terg-a-zyme (Sigma- Aldrich) and then rinsed with DiI water and sterilized in ethanol. We used Neuropixels 2.0 probes for all chronic recording experiments, and attached them to recoverable, custom-fabricated metal microdrives (design adapted from Vöröslakos et al., 2021). To clean probes recovered from chronic implants, we soaked the electrode shanks in alternating baths of 1% Terg-a-zyme, Dowsil DS-2025 (Dow-Corning), and ethanol. A.3 Electrophysiology A.3.1 Acute recordings Acute electrophysiological recordings were performed at least 3-4 hours after the burr hole surgery (section A.2.1), and after mice exhibited full motor recovery and self-groomed to remove the protective eye cream. We first head fixed the mice and removed the protective layers of dental cement and Kwik- cast. We then filled the well over the recording site was with saline, and placed a chlorinated silver wire into the bath and connected to the ground and reference sites of the Neuropixel probe. A micromanipulator (Luigs & Neumann) was used to position a Neuropixel 1.0 probe at the burr hole site. We adjusted the angle of the probe in order to approximate a normal insertion relative to the cortical surface. For optogenetic experiments, we used a second 76 micromanipulator to position an optical fiber directly above the cortical surface before probe insertion. The probe was then advanced at 1.5 µm/s until the cortical depth was spanned ( 1500 µm), and then allowed to settle for 15 minutes before the beginning of recording. Data acquisition was performed using a National Instruments system (PXIe-1071) and bandpass filtered into AP and LFP bands (0.3-10 kHz and 0.5-500 Hz, respectively) through SpikeGLX. AP and LFP band data were collected at 30 kHz and 2.5 kHz, respectively. At the end of each session, we slowly withdrew the electrode at the same speed as insertion (1.5 µm/s) and covered the recording sites with Kwik-cast before preparing the animal for histology. A.3.2 Chronic recordings On each day of chronic recording, mice were head fixed and the implant covering was removed. For optogenetic experiments in NDNF-Cre mice, we used a micromanipulator to position an optical fiber over the craniotomy site, and wrapped the entire implant with opaque tape in order to minimize light leakage. The recording headstage was then attached, and a custom channel map was loaded for each animal. We sampled uniformly across all four shanks, resulting in 96 channels spanning 720 µm of contiguous cortical depth per shank. Due to constraints in the flexibility of channel selection, we biased channel selection to maximize coverage of cortical layers 1-5. Spiking and LFP data were acquired in the same data stream at 30 kHz using a National Instruments system (PXIe-1083) and bandpass filtered (0.5 Hz - 10 kHz) through SpikeGLX. In a subset of mice, we performed freely behaving recordings in the home cage 77 to capture periods of putative sleep. For these recordings, we temporarily fixed the headstage to the implant body and then replaced the implant covering. We then removed the mice from head fixation and transferred them into their home cages. At the end of each recording session, we removed the headstage and replaced the implant covering. We recorded from individual mice for up to two months after probe implantation, and recovered the probes before histology. A.4 Optogenetics We controlled light delivery for optogenetic perturbations using a Cyclops LED driver (OpenEphys, Catalog # OEPS-6602) triggered by an external Arduino signal. The Cyclops driver was connected to a fiber-coupled LED, which terminated in a 400 µm optical fiber positioned above the cortical surface. We used a 595 nm LED (8.7 mW, 1000 mA, ThorLabs catalog # M595F2) to provide excitation for ChRmine experiments, and a 455 LED (17 mW, 1000 mA, Thorlabs catalog # M455F3) for eNpHR3.0 experiments. Titrations in light intensity were programmed through the Cyclops library (Newman et al., 2015). Light delivery was initiated at the onset of visual stimuli, and was performed for a fixed duration (400 or 500 ms) with a square wave profile. For direct L5 pyramidal neuron inhibition experiments (eNpHR3.0), we activated the LED for the entire duration of the visual stimuli (500 ms), and for NDNF interneuron activation experiments (ChRmine) we reduced the optogenetic perturbation to 400 ms to account for the long timescale of NDNF mediated inhibition (Cohen-Kashi Malina et al., 2021). 78 A.5 Visual stimulation We generated visual stimuli using the open-source PsychoPy toolbox based in Python (Peirce et al., 2019). All stimuli were presented on an LED-backlit LCD monitor with a refresh rate of 60 Hz, at a distance of 15 cm to the right eye (approximately 135° x 75° coverage of visual space). We used screen refresh cycles to define presentation time in order to maximize timing accuracy, and sent TTL pulses at the onset of each visual stimuli for post hoc synchronization to the electrophysiological data. All visual stimuli were block randomized. Our visual stimuli battery contained white noise stimuli and three different types of drifting gratings: Gabor stimuli, inverse stimuli, and full field drifting gratings. For the white noise stimuli, we generated a 25 x 14 grid of squares spanning the full screen ( 5.4 °per square), and randomly sampled the color (black to white) of each square from a uniform distribution. White noise stimuli were presented for 2 frames ( 33.3 ms), with no inter-trial interval between stimuli. Gabor stimuli consisted of a circular drifting grating patch presented on a full-field gray screen, and inverse stimuli consisted of a full-field drifting grating masked by a circular gray patch. The location of Gabor and inverse stimuli in visual space were defined as the center of their respective circular patches. To mitigate confounds caused by visual edge effects, we used a raised cosine filter to blur the edges of the patches for both classical and extra-classical stimuli. All drifting gratings were sinusoidal (0.04 cycles per degree, 2 Hz temporal frequency) and were presented at 100% contrast. (Keller et al., 2020). Unless otherwise specified, grating stimuli were presented for 30 frames (500 79 ms) and separated by an inter-trial interval of 450-500 ms, during which the screen was held an equiluminant gray. We presented all drifting gratings at the same initial phase (0°, black grating centered at onset) to ensure that all latency analyses were performed across phase-matched conditions (see A.9.x Response latency). Full field gratings were centered at the center of the screen unless otherwise specified. For optogenetic perturbation experiments, we multiplied the number of trials by the number of laser power titrations, and adjusted the spatial sampling of Gabor and inverse stimuli as necessary. Optogenetic perturbation trials were block randomized (i.e. different light titrations and non-opto control trials were randomly interleaved). In the first recording session for each chronic mouse, we presented 20000 white noise stimuli and/or Gabor stimuli tiling the entire screen. The Gabor stimuli (20°visual size) covered an even grid pattern with 15°of separation between the center of each stimuli. We presented 2 different directions per location, and repeated each unique Gabor stimuli for 10-15 trials. We estimated the receptive field of each neuron offline using either the white noise or Gabor (see A.9.3 Receptive field mapping). In subsequent experiments, we used the averaged population RF to reposition the monitor as necessary, and to fine-tune the positioning of Gabor and inverse stimuli. A.5.1 Gabor vs inverse tuning For baseline comparisons between Gabor and inverse, we presented Gabor, inverse, and full field stimuli. The Gabor and inverse stimuli were tiled along a square grid of visual locations (15°spacing) centered at the population RF 80 estimated for each mouse. The patch size was identical (20°) for Gabor and inverse stimuli. Each stimuli was presented at 2 different directions (0 and 90°) per location, and each unique stimuli was presented for 15-20 trials. A.5.2 Size tuning To evaluate size tuning, we presented patches of 6 different sizes (5, 15, 25, 35, 45, and 65°). We presented each unique stimuli at 2 different directions, and adjusted the extent of spatial tiling for the Gabor and inverse stimuli to maintain at least 15 trials per stimulus. A.5.3 Orientation tuning For orientation tuning experiments, we presented Gabor, inverse, and full field stimuli at 8 different directions (equally spaced from 0 to 360°, 20 trials per stimulus). For these experiments, we presented 20°Gabor stimuli, and increased the size of the gray mask in the inverse stimuli to 35°to minimize classical RF-related contamination of the inverse response. A.5.4 Sequential stimuli We created three types of sequential stimuli (Gabor -> inverse, inverse -> Gabor, full field -> Gabor) by presenting pairs of drifting grating stimuli with no inter-trial interval. Sequentially delivered Gabor and inverse stimuli (and visa-versa) were presented at the same location and had the same patch size (20°), such that the drifting grating and gray portions of the screen were exactly 81 inverted at each transition. For these sequential stimuli, we also matched the "center" location of each full field grating to their paired Gabor stimuli. All pairs of stimuli were presented with at the same direction (0 or 90°), initial phase (0°), and spatial frequency (2 Hz). Each individual stimuli was presented for 500 ms, resulting in an exact phase match (occurring after 1 full cycle of the preceding sinusoidal grating) at the sequential transitions. Each pair of sequential stimuli was followed by an inter-trial gray screen (450-500 ms). We presented 20 trials for each unique configuration. A.6 Behavioral monitoring Mice were awake and free to run on a treadmill during all head-fixed experiments. We recorded locomotion using an optical encoder (E6, US Digital, 2500 cycles per revolution), and facial movements using an infrared camera (Flir Blackfly, Teledyne) trained on the left side of the mouse’s face. Behavioral data was acquired and saved using Bonsai (Lopes et al., 2015), and synchronized to the neural recording post-hoc via shared TTLs. During homecage recordings, the monitor was turned off, and an overhead infrared camera was used to track motion. A.7 Histology For chronic Neuropixel experiments, the probe microdrives were recovered before histology. We anesthetized mice with 5% isoflurane, and maintained anesthesia at 1-2%. The implant covering and protective cage were removed, 82 and the silver ground wire connecting the probe to the skull was carefully disconnected using wire clippers. The probe microdrive was affixed to a stereotaxic arm, and then withdrawn from the brain at 2.5 µm/s. Following probe removal, the Dura-gel covering the craniotomy was reinforced using Kwik-Cast, and mice were prepared for histology. Mice were deeply anesthetized with 5% isoflurane and transcardially per- fused with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in PBS. After overnight fixation in 4% PFA, we washed the brains in PBS and took 100 µm thick coronal sections using a floating section vibratome (Leica VT1000s). Slices were mounted and coverslipped in a mounting medium containing DAPI (Vectashield H-1500-10, Vector Laboratories). Confocal im- ages were acquired using a Leica TCS SP8 microscope with a 10x objective (NA 0.40). Image pre-processing and curation was performed using Fiji (Schindelin et al., 2012), and all slices containing probe tracks (identified by fluorescent DiI or DiO tracks) were registered to the Allen Common Coordinate Frame- work (CCFv3) using SHARP-Track (Shamash et al., 2018). SHARP-Track reconstructions were used to validate the targeting of probes to V1. We used eYFP (co-expressed with eNpHR3.0) and oScarlet (co-expressed with ChRmine) fluorescence to verify that opsin expression was co-localized with the electrode tracks for the apical tuft perturbation experiments. 83 A.8 Data analysis A.8.1 Spike-sorting Raw electrophysiological data was time-shifted and common-average referenced using SpikeInterface (Buccino et al., 2020). We performed automated spike sort- ing using either Kilosort 2.5 or 4 (Pachitariu et al., 2016, 2024). Putative single units were manually curated using Phy 2.0 (Rossant & Harris, 2013) as follows. First, all clusters that were identified by Kilosort as ‘mua’ or ‘noise’, or had a fir- ing rate of < 0.1 Hz, were automatically rejected. Then, we split or merged clus- ters based on the auto and cross correlograms, following the procedures detailed in the Manual Clustering Practical User’s Guide (by S. Lenzi and N. Steinmetz, publicly available https://phy.readthedocs.io/en/latest/sorting_user_guide/). Special care was taken to look for erroneously split clusters stemming from single bursting cells. These over splits are likely due to changes in waveform shape and amplitude across consecutive action potentials within a burst, which has been characterized in both intra and extracellular recordings (Allen et al., 2018; Connors et al., 1982; Fee et al., 1996; Henze et al., 2000; McCormick et al., 1985). We separated regular spiking (RS) and fast spiking (FS) units based on the peak-to-trough time of each unit’s average waveform. Putative excitatory neurons (RS) were defined as those with a peak-to-trough greater than 0.4 ms, and putative inhibitory neurons (FS) as those with a peak-to-trough less than or equal to 0.4 ms (Barthó et al., 2004; Niell & Stryker, 2008; Sirota et al., 2008). RS and FS units were separated for all analyses. 84 A.8.2 Layer identification We used the depth profile of previously reported physiological landmarks to estimate laminar depth (Senzai et al., 2019). We first bandpass filtered the local field potential (LFP) between 0.5-500 Hz and averaged it over visual stimuli presentations. Neuropixel 2.0 data was temporally downsampled to 2.5 kHz to match Neuropixel 1.0. For L4 estimation, we spatially smoothed the LFP (Gaussian kernel, σ = 30 µm) and took its second spatial derivative to estimate the current source density (CSD) (Mitzdorf, 1985). L4 was identified by the prominent short latency current sink (Niell & Stryker, 2008). To identify L5, we band-pass filtered the raw AP band signal between 0.5 - 5000 Hz, and used the Hilbert transform to calculate the analytic amplitude at each channel (MUA power). The channel with the highest MUA power was designated the center of L5 (Oude Lohuis et al., 2024; Senzai et al., 2019). For layer specific analysis, we conservatively assigned units within +-100 µm of the MUA peak to L5, and units >275 µm above MUA peak to L2/3. A.8.3 Burst spike analysis We defined high-frequency bursts as events with three or more consecutive spikes with ISIs 10 ms. We classified burst spikes (b) as all events occurring within a burst, and tonic spikes (t) as all events outside a burst. The burst fraction (BF ) was defined as the number of burst spikes divided by the total number of spikes (BF = b/(b+t)). For estimating the burst fraction across time, we first calculated smoothed peri-stimulus time histograms (PSTHs) separately 85 for burst and tonic spikes (1 ms binning, Gaussian smoothing with σ = 5-10 ms). The resulting smoothed burst and tonic firing rates were then used to estimate an instantaneous burst fraction similar to above (BF = FRb/(FRb + FRt)). A.8.4 Spike backpropagation For a subset of recordings, we calculated the spatial spread of action potentials along the vertical axis of the recording electrode. For individual neurons, we extracted AP band (0.5 - 10 kHz) data across all channels triggered on spikes detected at the soma. We averaged these signals separately for burst and tonic spikes, and took the second spatial derivative to remove volume conducted signals. Spike triggered activity at each channel was z-scored against the window directly preceding the spike. A.8.5 White noise analysis For each neuron, we computed calculated the spike-triggered average (STA) of the white noise stimuli using burst, tonic, and all spikes. STAs were calculated by reverse correlating each spike with the preceding white noise stimuli (Chichilnisky, n.d.; Schwartz et al., 2006). To identify tuned neurons, we used a shuffled STA (generated by randomly circularly shuffling the timing of the stimuli with respect to the spikes) to compute a z-scored significance. We considered neurons visually tuned if they exhibited a significant response (z-score > 4) at any point within 200 ms. For each neuron, we defined the response latency as the time at which z-scored tuning crossed threshold, and the classical receptive field as the maximally tuned position in visual space. 86 A.8.6 Receptive field mapping Classical receptive fields were mapped using white noise (see above) and Gabor stimuli. For each neuron, we computed the evoked response (average firing rate from 0-500 ms) to Gabor stimuli at each spatial location. We next generated a null distribution by computing the evoked response after shuffling the location across all Gabor presentations. We considered neurons receptive field tuned if they exhibited a significant evoked response at any spatial location (p < 0.05 compared to the shuffle distribution) and their evoked response was significantly greater than the pre-stimulus baseline (average firing rate 100 ms before stimuli presentation). We defined the classical receptive field as the position at which each RF tuned neuron responded maximally to Gabor stimuli. Only Gabor stimuli 25°in size were used for receptive field mapping. A.8.7 Inverse tuning For all neurons with a classical receptive field, we calculated the evoked response to inverse stimuli centered at the classical RF location. Neuron were considered inverse tuned if their cRF centered inverse response (of any size) was significantly greater than baseline (100 ms preceding) and the evoked response to full-field stimuli. We did not approximate the response to full-field stimuli using Gabor or inverse stimuli of different sizes. 87 A.8.8 Response latency To estimate response latency, we calculated the smoothed firing rate and burst fraction for grating stimuli presented in each neuron’s cRF as described above (see A.x.x Burst fraction). We defined the response latency as the half-peak time of the first peak for both firing rate and burst fraction. For burst latency comparisons, we excluded units which did not exhibit a significant burst fraction peak. For analyses in Fig. ??, we considered neurons to be early bursting if their burst fraction response latency was < 150 ms. We performed an unsupervised categorization of response latency using k-means clustering of each neuron’s firing rate and burst fraction latencies. The optimal number of clusters was determined by calculating the within-cluster sum of squares (inertia) across different cluster sizes (Fig. 2.6. A.8.9 Poisson model We used a Poisson spiking model to examine whether the Gabor stimuli evoked bursting could be explained by firing rate (Fig. 2.7). To do so, we simulated the spiking response for individual neurons using an inhomogeneous Poisson model matched to each neuron’s smoothed firing rate responses (Marmarelis & Berger, 2005). We then calculated the smoothed burst fraction of the simulated spike train using the same criteria as for the neural data (see Burst spike analysis). The model error for each neuron was defined as the absolute difference between the actual and Poisson burst fractions. 88 A.8.10 Optogenetic inhibition To identify optogenetically inhibited neurons, we compared the baseline cor- rected firing rates (trial averaged) across opto and non-opto trials. For sessions in which multiple LED powers were delivered, we used the maximum power to identify inhibited neurons. Only neurons which were significantly inhibited were included in subsequent analyses. A.8.11 Behavioral state We used FaceMap (Syeda et al., 2024) to extract whisking and eye movement at a sampling frequency of 200 Hz. These behavioral variables, were trial synchro- nized to visual stimuli presentation. We calculated the trial averaged changes in behavior for each variable. For each distribution of trial averaged behavioral data, we noticed that the median value corresponded to a stationary state (i.e. no whisking or eye movement). To separate stationary from movement trials, we fit a half-Gaussian model to data below the median. We then classified stationary trials as those within 2 standard deviations (based on the Gaussian model) of the median value. A.9 Statistics All statistics were performed using the scipy package in Python. For normally distributed data, we used Student’s t-tests for paired and independent sample comparisons. Otherwise, Wilcoxon signed-rank tests were used for independent 89 sample comparisons, and Mann-Whitney U tests were used for paired tests. Barplot error bars indicate the 95% confidence interval. All other error bars represent the standard error of the mean (SEM). 90 References Ährlund-Richter, S., Osako, Y., Jenks, K., Odom, E., Huang, H., Arnold, D., & Sur, M. (2024). Prefrontal Cortex subregions provide distinct visual and behavioral feedback modulation to the Primary Visual Cortex. bioRxiv, 2024.08.06.606894. https://doi.org/10.1101/2024.08.06.606894 Alitto, H. J., Weyand, T. G., & Usrey, W. M. (2005). Distinct properties of stimulus-evoked bursts in the lateral geniculate nucleus. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 25 (2), 514–523. https://doi.org/10.1523/JNEUROSCI.3369-04.2005 Allen, B. D., Moore-Kochlacs, C., Bernstein, J. G., Kinney, J. P., Scholvin, J., Seoane, L. F., Chronopoulos, C., Lamantia, C., Kodandaramaiah, S. B., Tegmark, M., & Boyden, E. S. (2018). Automated in vivo patch-clamp evaluation of extracellular multielectrode array spike recording capability [Publisher: American Physiological Society]. Journal of Neurophysiology, 120 (5), 2182–2200. https://doi.org/10.1152/jn.00650.2017 Anderson, E. B., Mitchell, J. F., & Reynolds, J. H. (2011). Attentional Modu- lation of Firing Rate Varies with Burstiness across Putative Pyramidal Neurons in Macaque Visual Area V4 [Publisher: Society for Neuro- 91 science Section: Articles]. Journal of Neuroscience, 31 (30), 10983–10992. https://doi.org/10.1523/JNEUROSCI.0027-11.2011 Anderson, E. B., Mitchell, J. F., & Reynolds, J. H. (2013). Attention-dependent reductions in burstiness and action-potential height in macaque area V4. Nature Neuroscience, 16 (8), 1125–1131. https://doi.org/10.1038/ nn.3463 Angelucci, A., & Bressloff, P. C. (2006, January). Contribution of feedforward, lateral and feedback connections to the classical receptive field center and extra-classical receptive field surround of primate V1 neurons. In S. Martinez-Conde, S. L. Macknik, L. M. Martinez, J. .-. Alonso, & P. U. Tse (Eds.), Progress in Brain Research (pp. 93–120, Vol. 154). Elsevier. https://doi.org/10.1016/S0079-6123(06)54005-1 Angelucci, A., Levitt, J. B., Walton, E. J. S., Hupé, J.-M., Bullier, J., & Lund, J. S. (2002). Circuits for Local and Global Signal Integration in Primary Visual Cortex [Publisher: Society for Neuroscience Section: ARTICLE]. Journal of Neuroscience, 22 (19), 8633–8646. https://doi.org/10.1523/ JNEUROSCI.22-19-08633.2002 Angelucci, A., Bijanzadeh, M., Nurminen, L., Federer, F., Merlin, S., & Bressloff, P. C. (2017). Circuits and Mechanisms for Surround Modulation in Visual Cortex [_eprint: https://doi.org/10.1146/annurev-neuro-072116- 031418]. Annual Review of Neuroscience, 40 (1), 425–451. https://doi. org/10.1146/annurev-neuro-072116-031418 Banerjee, A., Parente, G., Teutsch, J., Lewis, C., Voigt, F. F., & Helmchen, F. (2020). Value-guided remapping of sensory cortex by lateral orbitofrontal 92 cortex [Publisher: Springer US]. Nature, 585 (September). https://doi. org/10.1038/s41586-020-2704-z Barthó, P., Hirase, H., Monconduit, L., Zugaro, M., Harris, K. D., & Buzsáki, G. (2004). Characterization of neocortical principal cells and interneurons by network interactions and extracellular features. Journal of Neuro- physiology, 92 (1), 600–608. https://doi.org/10.1152/jn.01170.2003 Bast, A., Guest, J. M., Fruengel, R., Narayanan, R. T., Kock, C. P. J. d., & Oberlaender, M. (2023, July). Thalamus drives active dendritic com- putations in cortex [Pages: 2021.10.21.465325 Section: New Results]. https://doi.org/10.1101/2021.10.21.465325 Beaulieu-Laroche, L., Toloza, E. H., Brown, N. J., & Harnett, M. T. (2019). Widespread and Highly Correlated Somato-dendritic Activity in Cortical Layer 5 Neurons [Publisher: Elsevier Inc.]. Neuron, 103 (2), 235–241.e4. https://doi.org/10.1016/j.neuron.2019.05.014 Bereshpolova, Y., Amitai, Y., Gusev, A. G., Stoelzel, C. R., & Swadlow, H. A. (2007). Dendritic Backpropagation and the State of the Awake Neocortex. The Journal of Neuroscience, 27 (35), 9392–9399. https : //doi.org/10.1523/JNEUROSCI.2218-07.2007 Bittner, K. C., Milstein, A. D., Grienberger, C., Romani, S., & Magee, J. C. (2017). Behavioral time scale synaptic plasticity underlies CA1 place fields. Science, 357 (6355), 1033–1036. https://doi.org/10.1126/science. aan3846 Buccino, A. P., Hurwitz, C. L., Garcia, S., Magland, J., Siegle, J. H., Hurwitz, R., & Hennig, M. H. (2020). SpikeInterface, a unified framework for 93 spike sorting (L. L. Colgin, S. Grün, & F. Kloosterman, Eds.) [Publisher: eLife Sciences Publications, Ltd]. eLife, 9, e61834. https://doi.org/10. 7554/eLife.61834 Cattaneo, A., Maffei, L., & Morrone, C. (1981). Two firing patterns in the discharge of complex cells encoding different attributes of the visual stimulus. Experimental Brain Research, 43 (1). https://doi.org/10.1007/ BF00238819 Chichilnisky, E. J. (n.d.). A simple white noise analysis of neuronal light responses. Chu, Z., Galarreta, M., & Hestrin, S. (2003). Synaptic interactions of late- spiking neocortical neurons in layer 1. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 23 (1), 96–102. https://doi.org/10.1523/JNEUROSCI.23-01-00096.2003 Cichon, J., & Gan, W. B. (2015). Branch-specific dendritic Ca2+ spikes cause persistent synaptic plasticity. Nature, 520 (7546), 180–185. https://doi. org/10.1038/nature14251 Clarke, S. E., & Maler, L. (2017). Feedback Synthesizes Neural Codes for Motion. Current Biology, 27 (9), 1356–1361. https://doi.org/10.1016/j. cub.2017.03.068 Cohen-Kashi Malina, K., Tsivourakis, E., Kushinsky, D., Apelblat, D., Shtiglitz, S., Zohar, E., Sokoletsky, M., Tasaka, G.-i., Mizrahi, A., Lampl, I., & Spiegel, I. (2021). NDNF interneurons in layer 1 gain-modulate whole cortical columns according to an animal’s behavioral state. Neuron, 109 (13), 2150–2164.e5. https://doi.org/10.1016/j.neuron.2021.05.001 94 Connors, B. W., Gutnick, M. J., & Prince, D. A. (1982). Electrophysiolog- ical properties of neocortical neurons in vitro [Publisher: American Physiological Society]. Journal of Neurophysiology, 48 (6), 1302–1320. https://doi.org/10.1152/jn.1982.48.6.1302 Constantinople, C. M., & Bruno, R. M. (2013). Deep Cortical Layers Are Activated Directly by Thalamus. Science, 340 (6140), 1591–1594. https: //doi.org/10.1126/science.1236425 Contreras, D., Curró Dossi, R., & Steriade, M. (1992). Bursting and tonic discharges in two classes of reticular thalamic neurons. Journal of Neurophysiology, 68 (3), 973–977. https://doi.org/10.1152/jn.1992.68.3. 973 David, B., Idan, S., & Michael, L. (2019). Single Cortical Neurons as Deep Artificial Neural Networks. bioRxiv. https://doi.org/10.1101/613141 de Kock, C. P. J., Pie, J., Pieneman, A. W., Mease, R. A., Bast, A., Guest, J. M., Oberlaender, M., Mansvelder, H. D., & Sakmann, B. (2021). High-frequency burst spiking in layer 5 thick-tufted pyramids of rat pri- mary somatosensory cortex encodes exploratory touch. Communications Biology, 4 (1), 709. https://doi.org/10.1038/s42003-021-02241-8 de Vries, S. E. J., Lecoq, J. A., Buice, M. A., Groblewski, P. A., Ocker, G. K., Oliver, M., Feng, D., Cain, N., Ledochowitsch, P., Millman, D., Roll, K., Garrett, M., Keenan, T., Kuan, L., Mihalas, S., Olsen, S., Thompson, C., Wakeman, W., Waters, J., . . . Koch, C. (2020). A large- scale standardized physiological survey reveals functional organization of the mouse visual cortex [Publisher: Nature Publishing Group]. Nature 95 Neuroscience, 23 (1), 138–151. https://doi.org/10.1038/s41593-019- 0550-9 De Kock, C. P. J., & Sakmann, B. (2008). High frequency action potential bursts ( 100 Hz) in L2/3 and L5B thick tufted neurons in anaesthetized and awake rat primary somatosensory cortex [_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1113/jphysiol.2008.155580]. The Journal of Physiology, 586 (14), 3353–3364. https://doi.org/10. 1113/jphysiol.2008.155580 De Kock, C. P. J., Bruno, R. M., Spors, H., & Sakmann, B. (2007). Layer- and cell-type-specific suprathreshold stimulus representation in rat primary somatosensory cortex [_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1113/jphysiol.2006.124321]. The Journal of Physiology, 581 (1), 139–154. https://doi.org/10.1113/ jphysiol.2006.124321 DeAngelis, G. C., Ohzawa, I., & Freeman, R. D. (1993). Spatiotemporal organi- zation of simple-cell receptive fields in the cat’s striate cortex. I. General characteristics and postnatal development. Journal of Neurophysiology, 69 (4), 1091–1117. https://doi.org/10.1152/jn.1993.69.4.1091 DeAngelis, G. C., Ohzawa, I., & Freeman, R. D. (1995). Receptive-field dy- namics in the central visual pathways. Trends in Neurosciences, 18 (10), 451–458. https://doi.org/10.1016/0166-2236(95)94496-R Destexhe, A., & Paré, D. (1999). Impact of network activity on the integrative properties of neocortical pyramidal neurons in vivo. Journal of Neuro- physiology, 81 (4), 1531–1547. https://doi.org/10.1152/jn.1999.81.4.1531 96 Destexhe, A., Rudolph, M., & Paré, D. (2003). The high-conductance state of neocortical neurons in vivo. Nature Reviews Neuroscience, 4 (9), 739–751. https://doi.org/10.1038/nrn1198 Diamanti, E. M., Reddy, C. B., Schröder, S., Muzzu, T., Harris, K., Saleem, A., & Carandini, M. (2019). Spatial modulation of visual signals arises in cortex with active navigation, 1–15. https://doi.org/10.1101/832915 Doron, G., von Heimendahl, M., Schlattmann, P., Houweling, A. R., & Brecht, M. (2014). Spiking Irregularity and Frequency Modulate the Behavioral Report of Single-Neuron Stimulation [Publisher: Elsevier]. Neuron, 81 (3), 653–663. https://doi.org/10.1016/j.neuron.2013.11.032 Doron, G., Shin, J. N., Takahashi, N., Drüke, M., Bocklisch, C., Skenderi, S., De Mont, L., Toumazou, M., Ledderose, J., Brecht, M., Naud, R., & Larkum, M. E. (2020). Perirhinal input to neocortical layer 1 controls learning. Science, 370 (6523), eaaz3136. https://doi.org/10.1126/science.aaz3136 Douglas, R. J., & Martin, K. A. (2004). NEURONAL CIRCUITS OF THE NEOCORTEX. Annual Review of Neuroscience, 27 (1), 419–451. https: //doi.org/10.1146/annurev.neuro.27.070203.144152 Federer, F., Ta’afua, S., Merlin, S., Hassanpour, M. S., & Angelucci, A. (2021). Stream-specific feedback inputs to the primate primary visual cortex [Number: 1 Publisher: Nature Publishing Group]. Nature Communica- tions, 12 (1), 228. https://doi.org/10.1038/s41467-020-20505-5 Fee, M. S., Mitra, P. P., & Kleinfeld, D. (1996). Variability of extracellular spike waveforms of cortical neurons [Publisher: American Physiological 97 Society]. Journal of Neurophysiology, 76 (6), 3823–3833. https://doi.org/ 10.1152/jn.1996.76.6.3823 Felleman, D. J., & Van Essen, D. C. (1991). Distributed Hierarchical Processing in the Primate Cerebral Cortex. Cerebral Cortex, 1 (1), 1–47. https: //doi.org/10.1093/cercor/1.1.1-a Ferster, D., & Lindström, S. (1983). An intracellular analysis of geniculo-cortical connectivity in area 17 of the cat. The Journal of Physiology, 342, 181– 215. https://doi.org/10.1113/jphysiol.1983.sp014846 Fişek, M., Herrmann, D., Egea-Weiss, A., Cloves, M., Bauer, L., Lee, T.-Y., Russell, L. E., & Häusser, M. (2023). Cortico-cortical feedback engages active dendrites in visual cortex [Publisher: Nature Publishing Group]. Nature, 617 (7962), 769–776. https://doi.org/10.1038/s41586-023-06007- 6 Francioni, V., Padamsey, Z., & Rochefort, N. L. (2019). High and asymmetric somato-dendritic coupling of v1 layer 5 neurons independent of visual stimulation and locomotion. eLife, 8, 1–25. https://doi.org/10.7554/ eLife.49145 Freund, T. F., Martin, K. a. C., & Whitteridge, D. (1985). Innervation of cat vi- sual areas 17 and 18 by physiologically identified X- and Y- type thalamic afferents. I. Arborization patterns and quantitative distribution of postsy- naptic elements [_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/cne.902420208]. Journal of Comparative Neurology, 242 (2), 263–274. https://doi.org/10. 1002/cne.902420208 98 Friedenberger, Z., Harkin, E., Tóth, K., & Naud, R. (2023). Silences, spikes and bursts: Three-part knot of the neural code [_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1113/JP281510]. The Journal of Physiology, 601 (23), 5165–5193. https://doi.org/10. 1113/JP281510 Froemke, R. C., Tsay, I. A., Raad, M., Long, J. D., & Dan, Y. (2006). Contri- bution of Individual Spikes in Burst-Induced Long-Term Synaptic Mod- ification [Publisher: American Physiological Society]. Journal of Neuro- physiology, 95 (3), 1620–1629. https://doi.org/10.1152/jn.00910.2005 Gabbiani, F., Metzner, W., Wessel, R., & Koch, C. (1996). From stimulus encoding to feature extraction in weakly electric fish [Publisher: Nature Publishing Group]. Nature, 384 (6609), 564–567. https://doi.org/10. 1038/384564a0 Gilbert, C. D., & Li, W. (2013). Top-down influences on visual processing [Number: 5 Publisher: Nature Publishing Group]. Nature Reviews Neu- roscience, 14 (5), 350–363. https://doi.org/10.1038/nrn3476 Gilbert, C. D., & Wiesel, T. N. (1979). Morphology and intracortical projections of functionally characterised neurones in the cat visual cortex [Publisher: Nature Publishing Group]. Nature, 280 (5718), 120–125. https://doi. org/10.1038/280120a0 Greedy, W., Zhu, H. W., Pemberton, J., Mellor, J., & Costa, R. P. (2022, June). Single-phase deep learning in cortico-cortical networks [arXiv:2206.11769 [cs, q-bio]]. https://doi.org/10.48550/arXiv.2206.11769 99 Guerguiev, J., Lillicrap, T. P., & Richards, B. A. (2017). Towards deep learning with segregated dendrites (P. Latham, Ed.) [Publisher: eLife Sciences Publications, Ltd]. eLife, 6, e22901. https://doi.org/10.7554/eLife.22901 Guido, W., Lu, S. M., & Sherman, S. M. (1992). Relative contributions of burst and tonic responses to the receptive field properties of lateral geniculate neurons in the cat. Journal of Neurophysiology, 68 (6), 2199– 2211. https://doi.org/10.1152/jn.1992.68.6.2199 Harris, J. A., Mihalas, S., Hirokawa, K. E., Whitesell, J. D., Choi, H., Bernard, A., Bohn, P., Caldejon, S., Casal, L., Cho, A., Feiner, A., Feng, D., Gaudreault, N., Gerfen, C. R., Graddis, N., Groblewski, P. A., Henry, A. M., Ho, A., Howard, R., . . . Zeng, H. (2019). Hierarchical organization of cortical and thalamic connectivity [Publisher: Nature Publishing Group]. Nature, 575 (7781), 195–202. https://doi.org/10.1038/s41586- 019-1716-z Harris, K. D., Henze, D. A., Csicsvari, J., Hirase, H., & Buzsáki, G. (2000). Accuracy of tetrode spike separation as determined by simultaneous in- tracellular and extracellular measurements. Journal of Neurophysiology, 84 (1), 401–414. https://doi.org/10.1152/jn.2000.84.1.401 Hasselmo, M. E. (1995). Neuromodulation and cortical function: Modeling the physiological basis of behavior. Behavioural Brain Research, 67 (1), 1–27. https://doi.org/10.1016/0166-4328(94)00113-T Hawkins, J., & Ahmad, S. (2016). Why Neurons Have Thousands of Synapses, a Theory of Sequence Memory in Neocortex [arXiv: 1511.00083]. Frontiers 100 in neural circuits, 10 (March), 23. https://doi.org/10.3389/fncir.2016. 00023 Henze, D. A., Borhegyi, Z., Csicsvari, J., Mamiya, A., Harris, K. D., & Buzsáki, G. (2000). Intracellular Features Predicted by Extracellular Recordings in the Hippocampus In Vivo [Publisher: American Physiological Society]. Journal of Neurophysiology, 84 (1), 390–400. https://doi.org/10.1152/jn. 2000.84.1.390 Higgins, I., Chang, L., Langston, V., Hassabis, D., Summerfield, C., Tsao, D., & Botvinick, M. (2020). Unsupervised deep learning identifies semantic disentanglement in single inferotemporal neurons [arXiv: 2006.14304], 1–24. http://arxiv.org/abs/2006.14304 Huang, S., Wu, S. J., Sansone, G., Ibrahim, L. A., & Fishell, G. (2024). Layer 1 neocortex: Gating and integrating multidimensional signals [Publisher: Elsevier]. Neuron, 112 (2), 184–200. https://doi.org/10.1016/j.neuron. 2023.09.041 Hubel, D. H., & Wiesel, T. N. (1959). Receptive fields of single neurones in the cat’s striate cortex. The Journal of Physiology, 148 (3), 574–591. Retrieved April 8, 2025, from https://www.ncbi.nlm.nih.gov/pmc/ articles/PMC1363130/ Hubel, D. H., & Wiesel, T. (1962). Receptive Fields, Binocular Interaction and Functional Architecture in the Cat’s Visual Cortex. J. Neurophysiology, 106–154. 101 Iacaruso, M. F., Gasler, I. T., & Hofer, S. B. (2017). Synaptic organization of vi- sual space in primary visual cortex [Publisher: Nature Publishing Group]. Nature, 547 (7664), 449–452. https://doi.org/10.1038/nature23019 Izhikevich, E. (2003). Simple model of spiking neurons. IEEE Transactions on Neural Networks, 14 (6), 1569–1572. https://doi.org/10.1109/TNN.2003. 820440 Jiang, X., Shen, S., Cadwell, C. R., Berens, P., Sinz, F., Ecker, A. S., Patel, S., & Tolias, A. S. (2015). Principles of connectivity among morphologically defined cell types in adult neocortex [Publisher: American Association for the Advancement of Science]. Science, 350 (6264), aac9462. https: //doi.org/10.1126/science.aac9462 Jones, E. G., & Powell, T. P. (1970). An anatomical study of converging sensory pathways within the cerebral cortex of the monkey. Brain: A Journal of Neurology, 93 (4), 793–820. https://doi.org/10.1093/brain/93.4.793 Jones, J. P., & Palmer, L. A. (1987). An evaluation of the two-dimensional Gabor filter model of simple receptive fields in cat striate cortex [Publisher: American Physiological Society]. Journal of Neurophysiology, 58 (6), 1233–1258. https://doi.org/10.1152/jn.1987.58.6.1233 Jordan, R., & Keller, G. (2020). Opposing influence of top-down and bottom-up input on different types of excitatory layer 2/3 neurons in mouse visual cortex [Publisher: Elsevier Inc.]. Neuron, 1–13. https://doi.org/10.1101/ 2020.03.25.008607 Kampa, B. M., Letzkus, J. J., & Stuart, G. J. (2006). Requirement of den- dritic calcium spikes for induction of spike-timing-dependent synap- 102 tic plasticity. The Journal of Physiology, 574 (Pt 1), 283–290. https: //doi.org/10.1113/jphysiol.2006.111062 Keller, A. J., Roth, M. M., & Scanziani, M. (2020). Feedback generates a second receptive field in neurons of the visual cortex [Publisher: Nature Publishing Group]. Nature, 582 (7813), 545–549. https://doi.org/10. 1038/s41586-020-2319-4 Kepecs, A., & Lisman, J. (2003). Information encoding and computation with spikes and bursts. Network: Computation in Neural Systems, 14 (1), 103–118. https://doi.org/10.1088/0954-898X/14/1/306 Kepecs, A., Wang, X.-J., & Lisman, J. (2002). Bursting Neurons Signal Input Slope. The Journal of Neuroscience, 22 (20), 9053–9062. https://doi. org/10.1523/JNEUROSCI.22-20-09053.2002 Kerlin, A., Boaz, M., Flickinger, D., Maclennan, B. J., Dean, M. B., Davis, C., Spruston, N., & Svoboda, K. (2019). Functional clustering of dendritic activity during decision-making. eLife, 8, 1–32. https://doi.org/10.7554/ eLife.46966 Körding, K. P., & König, P. (2001). Supervised and Unsupervised Learning with Two Sites of Synaptic Integration. Journal of Computational Neu- roscience, 11 (3), 207–215. https://doi.org/10.1023/A:1013776130161 Krahe, R., & Gabbiani, F. (2004). Burst firing in sensory systems [Publisher: Nature Publishing Group]. Nature Reviews Neuroscience, 5 (1), 13–23. https://doi.org/10.1038/nrn1296 Lafourcade, M., van der Goes, M.-S. H., Vardalaki, D., Brown, N. J., Voigts, J., Yun, D. H., Kim, M. E., Ku, T., & Harnett, M. T. (2022). Differential 103 dendritic integration of long-range inputs in association cortex via subcellular changes in synaptic AMPA-to-NMDA receptor ratio. Neuron, 110 (9), 1532–1546.e4. https://doi.org/10.1016/j.neuron.2022.01.025 Lamme, V. A. F., Zipser, K., & Spekreijse, H. (1998). Figure-ground activity in primary visual cortex is suppressed by anesthesia [Publisher: Proceedings of the National Academy of Sciences]. Proceedings of the National Academy of Sciences, 95 (6), 3263–3268. https://doi.org/10.1073/pnas. 95.6.3263 Larkum, M. E., Kaiser, K. M. M., & Sakmann, B. (1999). Calcium electroge- nesis in distal apical dendrites of layer 5 pyramidal cells at a critical frequency of back-propagating action potentials. Proceedings of the Na- tional Academy of Sciences of the United States of America, 96 (25), 14600–14604. Retrieved March 31, 2025, from https://www.ncbi.nlm. nih.gov/pmc/articles/PMC24482/ Larkum, M. (2013). A cellular mechanism for cortical associations: An organiz- ing principle for the cerebral cortex [Publisher: Elsevier Ltd]. Trends in Neurosciences, 36 (3), 141–151. https://doi.org/10.1016/j.tins.2012.11. 006 Larkum, M. E., Zhu, J. J., & Sakmann, B. (1999). A new cellular mechanism for coupling inputs arriving at different cortical layers. Nature, 398 (6725), 338–341. https://doi.org/10.1038/18686 Larkum, M. E., Nevian, T., Sandler, M., Polsky, A., & Schiller, J. (2009). Synaptic Integration in Tuft Dendrites: A New Unifying Principle. Science, 756 (2009), 756–761. 104 Lewicki, M. S. (1998). A review of methods for spike sorting: The detection and classification of neural action potentials. Network: Computation in Neural Systems. Li, N., & DiCarlo, J. (2008). Unsupervised Natural Experience Rapidly Alters Invariant Object Representation in Visual Cortex. Science, 321. https: //doi.org/10.1126/science.1162981 Lien, A. D., & Scanziani, M. (2018a). Cortical direction selectivity emerges at convergence of thalamic synapses [Publisher: Springer US]. Nature, 558 (7708), 80–86. https://doi.org/10.1038/s41586-018-0148-5 Lien, A. D., & Scanziani, M. (2018b). Cortical direction selectivity emerges at convergence of thalamic synapses [Number: 7708 Publisher: Nature Publishing Group]. Nature, 558 (7708), 80–86. https://doi.org/10.1038/ s41586-018-0148-5 Lisman, J. E. (1997). Bursts as a unit of neural information: Making unreliable synapses reliable [Publisher: Elsevier]. Trends in Neurosciences, 20 (1), 38–43. https://doi.org/10.1016/S0166-2236(96)10070-9 Liu, Y., Zhang, J., Jiang, Z., Qin, M., Xu, M., Zhang, S., & Ma, G. (2024). Organization of corticocortical and thalamocortical top-down inputs in the primary visual cortex [Publisher: Nature Publishing Group]. Nature Communications, 15 (1), 4495. https://doi.org/10.1038/s41467-024- 48924-8 Livingstone, M. S., Freeman, D. C., & Hubel, D. H. (1996). Visual Responses in V1 of Freely Viewing Monkeys [Publisher: Cold Spring Harbor Labo- 105 ratory Press]. Cold Spring Harbor Symposia on Quantitative Biology, 61, 27–37. https://doi.org/10.1101/SQB.1996.061.01.006 Llinás, R. R., & Steriade, M. (2006). Bursting of thalamic neurons and states of vigilance. Journal of Neurophysiology, 95 (6), 3297–3308. https://doi. org/10.1152/jn.00166.2006 London, M., & Häusser, M. (2005). Dendritic computation. Annual Review of Neuroscience, 28, 503–532. https://doi.org/10.1146/annurev.neuro.28. 061604.135703 Lopes, G., Bonacchi, N., Frazão, J., Neto, J. P., Atallah, B. V., Soares, S., Moreira, L., Matias, S., Itskov, P. M., Correia, P. A., Medina, R. E., Calcaterra, L., Dreosti, E., Paton, J. J., & Kampff, A. R. (2015). Bonsai: An event-based framework for processing and controlling data streams [Publisher: Frontiers]. Frontiers in Neuroinformatics, 9. https://doi.org/ 10.3389/fninf.2015.00007 Lowet, E., Sheehan, D. J., Chialva, U., Pena, R. D. O., Mount, R. A., Xiao, S., Zhou, S. L., Tseng, H.-a., Gritton, H., Shroff, S., Kondabolu, K., Cheung, C., Wang, Y., Piatkevich, K. D., Boyden, E. S., Mertz, J., Hasselmo, M. E., Rotstein, H. G., & Han, X. (2023). Theta and gamma rhythmic coding through two spike output modes in the hippocampus during spatial navigation [Publisher: Elsevier]. Cell Reports, 42 (8). https://doi.org/10.1016/j.celrep.2023.112906 Major, G., Larkum, M. E., & Schiller, J. (2013). Active properties of neocortical pyramidal neuron dendrites [ISBN: 0621111503]. Annual Review of 106 Neuroscience, 36, 1–24. https://doi.org/10.1146/annurev-neuro-062111- 150343 Makino, H., & Komiyama, T. (2015). Learning enhances the relative impact of top-down processing in the visual cortex. Nature Neuroscience, 18 (8), 1116–1122. https://doi.org/10.1038/nn.4061 Manita, S., Suzuki, T., Homma, C., Matsumoto, T., Odagawa, M., Yamada, K., Ota, K., Matsubara, C., Inutsuka, A., Sato, M., Ohkura, M., Yamanaka, A., Yanagawa, Y., Nakai, J., Hayashi, Y., Larkum, M. E., & Murayama, M. (2015). A Top-Down Cortical Circuit for Accurate Sensory Perception. Neuron, 86 (5), 1304–1316. https://doi.org/10.1016/j.neuron.2015.05.006 Marder, E., & Thirumalai, V. (2002). Cellular, synaptic and network effects of neuromodulation. Neural Networks: The Official Journal of the Inter- national Neural Network Society, 15 (4-6), 479–493. https://doi.org/10. 1016/s0893-6080(02)00043-6 Marmarelis, V. Z., & Berger, T. W. (2005). General methodology for nonlinear modeling of neural systems with Poisson point-process inputs. Mathe- matical Biosciences, 196 (1), 1–13. https://doi.org/10.1016/j.mbs.2005. 04.002 Marques, T., Nguyen, J., Fioreze, G., & Petreanu, L. (2018). The functional organization of cortical feedback inputs to primary visual cortex [Pub- lisher: Springer US]. Nature Neuroscience, 21 (5), 757–764. https://doi. org/10.1038/s41593-018-0135-z Marques, T., Summers, M. T., Fioreze, G., Fridman, M., Dias, R. F., Feller, M. B., & Petreanu, L. (2018). A Role for Mouse Primary Visual Cortex 107 in Motion Perception. Current Biology, 28 (11), 1703–1713.e6. https: //doi.org/10.1016/j.cub.2018.04.012 Martinez-Conde, S., Macknik, S. L., & Hubel, D. H. (2002). The function of bursts of spikes during visual fixation in the awake primate lateral geniculate nucleus and primary visual cortex [Publisher: Proceedings of the National Academy of Sciences]. Proceedings of the National Academy of Sciences, 99 (21), 13920–13925. https://doi.org/10.1073/ pnas.212500599 Mason, A., & Larkman, A. (1990). Correlations between morphology and electrophysiology of pyramidal neurons in slices of rat visual cortex. II. Electrophysiology. Journal of Neuroscience, 10 (5), 1415–1428. https: //doi.org/10.1523/jneurosci.10-05-01415.1990 McCormick, D. A. (1992). Neurotransmitter actions in the thalamus and cerebral cortex and their role in neuromodulation of thalamocortical activity. Progress in Neurobiology, 39 (4), 337–388. https://doi.org/10. 1016/0301-0082(92)90012-4 McCormick, D. A., Connors, B. W., Lighthall, J. W., & Prince, D. A. (1985). Comparative electrophysiology of pyramidal and sparsely spiny stellate neurons of the neocortex. Journal of Neurophysiology, 54 (4), 782–806. https://doi.org/10.1152/jn.1985.54.4.782 Metzen, M. G., Krahe, R., & Chacron, M. J. (2016). Burst Firing in the Elec- trosensory System of Gymnotiform Weakly Electric Fish: Mechanisms and Functional Roles [Publisher: Frontiers]. Frontiers in Computational Neuroscience, 10. https://doi.org/10.3389/fncom.2016.00081 108 Metzner, W., Koch, C., Wessel, R., & Gabbiani, F. (1998). Feature Extraction by Burst-Like Spike Patterns in Multiple Sensory Maps [Publisher: Society for Neuroscience Section: ARTICLE]. Journal of Neuroscience, 18 (6), 2283–2300. https://doi.org/10.1523/JNEUROSCI.18-06-02283.1998 Miller, E. K., & Cohen, J. D. (2001). An Integrative Theory of Prefrontal Cortex Function (tech. rep.). Retrieved April 19, 2019, from www.annualreviews. org Mitzdorf, U. (1985). Current source-density method and application in cat cerebral cortex: Investigation of evoked potentials and EEG phenomena. Physiological Reviews, 65 (1), 37–100. https://doi.org/10.1152/physrev. 1985.65.1.37 Mountcastle, V. B. (1997). The columnar organization of the neocortex. Brain: A Journal of Neurology, 120 ( Pt 4), 701–722. https://doi.org/10.1093/ brain/120.4.701 Naud, R., & Sprekeler, H. (2018). Sparse bursts optimize information transmis- sion in a multiplexed neural code. Proceedings of the National Academy of Sciences of the United States of America, 115 (27), E6329–E6338. https://doi.org/10.1073/pnas.1720995115 Naud, R., Friedenberger, Z., & Toth, K. (2023, February). Silences, Spikes and Bursts: Three-Part Knot of the Neural Code [arXiv:2302.07206 [q-bio]]. Retrieved April 19, 2023, from http://arxiv.org/abs/2302.07206 Nestvogel, D. B., & McCormick, D. A. (2022). Visual thalamocortical mecha- nisms of waking state-dependent activity and alpha oscillations. Neuron, 110 (1), 120–138.e4. https://doi.org/10.1016/j.neuron.2021.10.005 109 Newman, J. P., Fong, M.-f., Millard, D. C., Whitmire, C. J., Stanley, G. B., & Potter, S. M. (2015). Optogenetic feedback control of neural activity (M. Bartos, Ed.) [Publisher: eLife Sciences Publications, Ltd]. eLife, 4, e07192. https://doi.org/10.7554/eLife.07192 Niell, C. M., & Stryker, M. P. (2008). Highly Selective Receptive Fields in Mouse Visual Cortex. The Journal of Neuroscience, 28 (30), 7520–7536. https://doi.org/10.1523/JNEUROSCI.0623-08.2008 Noudoost, B., Chang, M. H., Steinmetz, N. A., & Moore, T. (n.d.). TOP- DOWN CONTROL OF VISUAL ATTENTION. https://doi.org/10. 1016/j.conb.2010.02.003 Nurminen, L., Merlin, S., Bijanzadeh, M., Federer, F., & Angelucci, A. (2018). Top-down feedback controls spatial summation and response amplitude in primate visual cortex. Nature Communications, 9, 2281. https://doi. org/10.1038/s41467-018-04500-5 Oh, S. W., Harris, J. A., Ng, L., Winslow, B., Cain, N., Mihalas, S., Wang, Q., Lau, C., Kuan, L., Henry, A. M., Mortrud, M. T., Ouellette, B., Nguyen, T. N., Sorensen, S. A., Slaughterbeck, C. R., Wakeman, W., Li, Y., Feng, D., Ho, A., . . . Zeng, H. (2014). A mesoscale connectome of the mouse brain. Nature, 508 (7495), 207–214. https://doi.org/10.1038/nature13186 Okun, M., & Lampl, I. (2008). Instantaneous correlation of excitation and inhibition during ongoing and sensory-evoked activities [Publisher: Na- ture Publishing Group]. Nature Neuroscience, 11 (5), 535–537. https: //doi.org/10.1038/nn.2105 110 Opitz, A., Falchier, A., Linn, G. S., Milham, M. P., & Schroeder, C. E. (2017). Limitations of ex vivo measurements for in vivo neuroscience. Proceedings of the National Academy of Sciences of the United States of America, 114 (20), 5243–5246. https://doi.org/10.1073/pnas.1617024114 Oswald, A.-M. M., Chacron, M. J., Doiron, B., Bastian, J., & Maler, L. (2004). Parallel Processing of Sensory Input by Bursts and Isolated Spikes [Publisher: Society for Neuroscience Section: Behavioral/Sys- tems/Cognitive]. Journal of Neuroscience, 24 (18), 4351–4362. https: //doi.org/10.1523/JNEUROSCI.0459-04.2004 Oude Lohuis, M. N., Marchesi, P., Olcese, U., & Pennartz, C. M. A. (2024). Triple dissociation of visual, auditory and motor processing in mouse primary visual cortex [Publisher: Nature Publishing Group]. Nature Neuroscience, 27 (4), 758–771. https://doi.org/10.1038/s41593-023- 01564-5 Pachitariu, M., Steinmetz, N., Kadir, S., Carandini, M., & D, H. K. (2016, June). Kilosort: Realtime spike-sorting for extracellular electrophysiology with hundreds of channels [Pages: 061481 Section: New Results]. https: //doi.org/10.1101/061481 Pachitariu, M., Sridhar, S., Pennington, J., & Stringer, C. (2024). Spike sorting with Kilosort4 [Publisher: Nature Publishing Group]. Nature Methods, 21 (5), 914–921. https://doi.org/10.1038/s41592-024-02232-7 Pak, A., Ryu, E., Li, C., & Chubykin, A. A. (2020). Top-down feedback controls the cortical representation of illusory contours in mouse primary visual 111 cortex. Journal of Neuroscience, 40 (3), 648–660. https://doi.org/10. 1523/JNEUROSCI.1998-19.2019 Palmer, L. M., Shai, A. S., Reeve, J. E., Anderson, H. L., Paulsen, O., & Larkum, M. E. (2014). NMDA spikes enhance action potential generation during sensory input [Publisher: Nature Publishing Group]. Nature Neuroscience, 17 (3), 383–390. https://doi.org/10.1038/nn.3646 Passingham, R. E., Stephan, K. E., & Kötter, R. (2002). The anatomical basis of functional localization in the cortex [Publisher: Nature Publishing Group]. Nature Reviews Neuroscience, 3 (8), 606–616. https://doi.org/ 10.1038/nrn893 Payeur, A., Guerguiev, J., Zenke, F., Richards, B. A., & Naud, R. (2020). Burst- dependent synaptic plasticity can coordinate learning in hierarchical circuits. bioRxiv, 1–30. https://doi.org/10.1101/2020.03.30.015511 Peirce, J., Gray, J. R., Simpson, S., MacAskill, M., Höchenberger, R., Sogo, H., Kastman, E., & Lindeløv, J. K. (2019). PsychoPy2: Experiments in behavior made easy. Behavior Research Methods, 51 (1), 195–203. https://doi.org/10.3758/s13428-018-01193-y Petreanu, L., Mao, T., Sternson, S. M., & Svoboda, K. (2009). The subcellular organization of neocortical excitatory connections [Publisher: Nature Publishing Group]. Nature, 457 (7233), 1142–1145. https://doi.org/10. 1038/nature07709 Rakic, P. (2009). Evolution of the neocortex: A perspective from develop- mental biology [Publisher: Nature Publishing Group]. Nature Reviews Neuroscience, 10 (10), 724–735. https://doi.org/10.1038/nrn2719 112 Ranganathan, G. N., Apostolides, P. F., Harnett, M. T., Xu, N.-L., Druckmann, S., & Magee, J. C. (2018). Active dendritic integration and mixed neocortical network representations during an adaptive sensing behavior [Number: 11 Publisher: Nature Publishing Group]. Nature Neuroscience, 21 (11), 1583–1590. https://doi.org/10.1038/s41593-018-0254-6 Rao, A., Legenstein, R., Subramoney, A., & Maass, W. (2021). Synaptic plas- ticity in distal dendrites can optimize probabilistic top-down predictions in pyramidal cells, 1–20. Reinagel, P., Godwin, D., Sherman, S. M., & Koch, C. (1999). Encoding of Visual Information by LGN Bursts [Publisher: American Physiological Society]. Journal of Neurophysiology, 81 (5), 2558–2569. https://doi.org/ 10.1152/jn.1999.81.5.2558 Rhodes, P. A., & Gray, C. M. (1994). Simulations of Intrinsically Bursting Neocortical Pyramidal Neurons. Neural Computation, 6 (6), 1086–1110. https://doi.org/10.1162/neco.1994.6.6.1086 Richards, B. A., & Lillicrap, T. P. (2019). Dendritic solutions to the credit assignment problem [Publisher: Elsevier Ltd]. Current Opinion in Neu- robiology, 54, 28–36. https://doi.org/10.1016/j.conb.2018.08.003 Rockland, K. S., & Pandya, D. N. (1979). Laminar origins and terminations of cortical connections of the occipital lobe in the rhesus monkey. Brain Research, 179 (1), 3–20. https://doi.org/10.1016/0006-8993(79)90485-2 Rossant, C., & Harris, K. D. (2013). Hardware-accelerated interactive data visualization for neuroscience in Python [Publisher: Frontiers]. Frontiers in Neuroinformatics, 7. https://doi.org/10.3389/fninf.2013.00036 113 Sacramento, J., Bengio, Y., Costa, R. P., & Senn, W. (2018). Dendritic cor- tical microcircuits approximate the backpropagation algorithm [arXiv: 1810.11393]. Advances in Neural Information Processing Systems, 2018- Decem(NeurIPS), 8721–8732. Sakmann, B. (2017). From single cells and single columns to cortical networks: Dendritic excitability, coincidence detection and synaptic transmission in brain slices and brains [_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1113/EP085776]. Experimental Physiology, 102 (5), 489–521. https://doi.org/10.1113/ EP085776 Saleem, A. B., Diamanti, E. M., Fournier, J., Harris, K. D., & Carandini, M. (2018). Coherent encoding of subjective spatial position in visual cortex and hippocampus. Nature, 562 (7725), 124–127. http://www.nature.com/ articles/s41586-018-0516-1%0Ahttp://dx.doi.org/10.1038/s41586-018- 0516-1 Salkoff, D. B., Zagha, E., McCarthy, E., & McCormick, D. A. (2020). Movement and Performance Explain Widespread Cortical Activity in a Visual Detection Task. Cerebral Cortex (New York, N.Y.: 1991), 30 (1), 421– 437. https://doi.org/10.1093/cercor/bhz206 Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., Preibisch, S., Rueden, C., Saalfeld, S., Schmid, B., Tinevez, J.-Y., White, D. J., Hartenstein, V., Eliceiri, K., Tomancak, P., & Cardona, A. (2012). Fiji: An open-source platform for biological-image analysis [Publisher: Nature Publishing Group]. Nature Methods, 9 (7), 676–682. https://doi.org/10.1038/nmeth.2019 114 Schnabel, U. H., Bossens, C., Lorteije, J. A., Self, M. W., Op de Beeck, H., & Roelfsema, P. R. (2018). Figure-ground perception in the awake mouse and neuronal activity elicited by figure-ground stimuli in primary visual cortex [Publisher: Springer US]. Scientific Reports, 8 (1), 1–14. https://doi.org/10.1038/s41598-018-36087-8 Schoenfeld, G., Kollmorgen, S., Lewis, C., Bethge, P., Reuss, A. M., Aguzzi, A., Mante, V., & Helmchen, F. (2021). Dendritic integration of sensory and reward information facilitates learning. bioRxiv. https://doi.org/10. 1101/2021.12.28.474360 Schuman, B., Dellal, S., Prönneke, A., Machold, R., & Rudy, B. (2021). Neo- cortical Layer 1: An Elegant Solution to Top-Down and Bottom-Up Integration. Annual Review of Neuroscience, 44 (1), 221–252. https : //doi.org/10.1146/annurev-neuro-100520-012117 Schwartz, O., Pillow, J. W., Rust, N. C., & Simoncelli, E. P. (2006). Spike- triggered neural characterization. Journal of Vision, 6 (4), 13. https: //doi.org/10.1167/6.4.13 Senzai, Y., Fernandez-Ruiz, A., & Buzsáki, G. (2019). Layer-Specific Physio- logical Features and Interlaminar Interactions in the Primary Visual Cortex of the Mouse. Neuron, 101, 500–513. https://doi.org/10.1016/j. neuron.2018.12.009 Shai, A. S., Anastassiou, C. A., Larkum, M. E., & Koch, C. (2015). Physiol- ogy of Layer 5 Pyramidal Neurons in Mouse Primary Visual Cortex: Coincidence Detection through Bursting [Publisher: Public Library of 115 Science]. PLOS Computational Biology, 11 (3), e1004090. https://doi. org/10.1371/journal.pcbi.1004090 Shamash, P., Carandini, M., Harris, K., & Steinmetz, N. (2018, October). A tool for analyzing electrode tracks from slice histology [Pages: 447995 Section: New Results]. https://doi.org/10.1101/447995 Shen, S., Jiang, X., Scala, F., Fu, J., Fahey, P., Kobak, D., Tan, Z., Zhou, N., Reimer, J., Sinz, F., & Tolias, A. S. (2022). Distinct organization of two cortico-cortical feedback pathways [Publisher: Nature Publishing Group]. Nature Communications, 13 (1), 6389. https://doi.org/10.1038/s41467- 022-33883-9 Sherman, S. M. (2001). Tonic and burst firing: Dual modes of thalamocortical relay [Publisher: Elsevier]. Trends in Neurosciences, 24 (2), 122–126. https://doi.org/10.1016/S0166-2236(00)01714-8 Siegel, M., Körding, K. P., & König, P. (2000). Integrating Top-Down and Bottom-Up Sensory Processing by Somato-Dendritic Interactions. Jour- nal of Computational Neuroscience, 8 (2), 161–173. https://doi.org/10. 1023/A:1008973215925 Siegle, J. H., Jia, X., Durand, S., Gale, S., Bennett, C., Graddis, N., Heller, G., Ramirez, T. K., Choi, H., Luviano, J. A., Groblewski, P. A., Ahmed, R., Arkhipov, A., Bernard, A., Billeh, Y. N., Brown, D., Buice, M. A., Cain, N., Caldejon, S., . . . Koch, C. (2021). Survey of spiking in the mouse visual system reveals functional hierarchy. Nature, 592 (7852), 86–92. https://doi.org/10.1038/s41586-020-03171-x 116 Sirota, A., Montgomery, S., Fujisawa, S., Isomura, Y., Zugaro, M., & Buzsáki, G. (2008). Entrainment of neocortical neurons and gamma oscillations by the hippocampal theta rhythm. Neuron, 60 (4), 683–697. https : //doi.org/10.1016/j.neuron.2008.09.014 Siu, C., Balsor, J., Merlin, S., Federer, F., & Angelucci, A. (2021). A direct interareal feedback-to-feedforward circuit in primate visual cortex [Num- ber: 1 Publisher: Nature Publishing Group]. Nature Communications, 12 (1), 4911. https://doi.org/10.1038/s41467-021-24928-6 Stringer, C., Pachitariu, M., Steinmetz, N., Reddy, C. B., Carandini, M., & Harris, K. D. (2019). Spontaneous behaviors drive multidimensional, brainwide activity [Publisher: American Association for the Advance- ment of Science]. Science, 364 (6437), eaav7893. https://doi.org/10. 1126/science.aav7893 Stuart, G., & Spruston, N. (1998). Determinants of voltage attenuation in neocortical pyramidal neuron dendrites. Journal of Neuroscience, 18 (10), 3501–3510. https://doi.org/10.1523/jneurosci.18-10-03501.1998 Stuart, G., Schiller, J., & Sakmann, B. (1997). Action potential initiation and propagation in rat neocortical pyramidal neurons. Journal of Physiology, 505 (3), 617–632. https://doi.org/10.1111/j.1469-7793.1997.617ba.x Syeda, A., Zhong, L., Tung, R., Long, W., Pachitariu, M., & Stringer, C. (2024). Facemap: A framework for modeling neural activity based on orofacial tracking [Publisher: Nature Publishing Group]. Nature Neuroscience, 27 (1), 187–195. https://doi.org/10.1038/s41593-023-01490-6 117 Takahashi, N., Oertner, T. G., Hegemann, P., & Larkum, M. E. (2016). Active cortical dendrites modulate perception. Science, 354 (6319), 1587–1590. https://doi.org/10.1126/science.aah6066 Van Essen, D. C., & Maunsell, J. H. (1983). Hierarchical organization and functional streams in the visual cortex. Trends in Neurosciences, 6 (100), 370–375. https://doi.org/10.1016/0166-2236(83)90167-4 Voigts, J., & Harnett, M. T. (2020). Somatic and Dendritic Encoding of Spatial Variables in Retrosplenial Cortex Differs during 2D Navigation. Neuron, 105 (2), 237–245.e4. https://doi.org/10.1016/j.neuron.2019.10.016 Vöröslakos, M., Petersen, P. C., Vöröslakos, B., & Buzsáki, G. (2021). Metal microdrive and head cap system for silicon probe recovery in freely moving rodent (L. L. Colgin, L. M. de la Prida, & A. L. Juavinett, Eds.) [Publisher: eLife Sciences Publications, Ltd]. eLife, 10, e65859. https://doi.org/10.7554/eLife.65859 Walker, E. Y., Sinz, F. H., Cobos, E., Muhammad, T., Froudarakis, E., Fahey, P. G., Ecker, A. S., Reimer, J., Pitkow, X., & Tolias, A. S. (2019). Inception loops discover what excites neurons most using deep predictive models [Publisher: Springer US]. Nature Neuroscience, 22 (12), 2060– 2065. https://doi.org/10.1038/s41593-019-0517-x Xu, N. L., Harnett, M. T., Williams, S. R., Huber, D., O’Connor, D. H., Svoboda, K., & Magee, J. C. (2012). Nonlinear dendritic integration of sensory and motor input during an active sensing task [Publisher: Nature Publishing Group]. Nature, 492 (7428), 247–251. https://doi. org/10.1038/nature11601 118 Xu, S., Jiang, W., Poo, M.-m., & Dan, Y. (2012). Activity recall in a visual cortical ensemble [Number: 3 Publisher: Nature Publishing Group]. Nature Neuroscience, 15 (3), 449–455. https://doi.org/10.1038/nn.3036 Yamins, D. L. K., Hong, H., Cadieu, C. F., Solomon, E. A., Seibert, D., Dicarlo, J. J., & Designed, J. J. D. (2014). Performance-optimized hierarchical models predict neural responses in higher visual cortex. PNAS, 111 (23), 8619–8624. https://doi.org/10.1073/pnas.1403112111 Yi, G., Wang, J., Wei, X., & Deng, B. (2017). Action potential initiation in a two-compartment model of pyramidal neuron mediated by dendritic Ca2+ spike [Publisher: Nature Publishing Group]. Scientific Reports, 7 (1), 45684. https://doi.org/10.1038/srep45684 Young, H., Belbut, B., Baeta, M., & Petreanu, L. (2021). Laminar-specific cortico-cortical loops in mouse visual cortex (S. P. Brown & J. R. Huguenard, Eds.) [Publisher: eLife Sciences Publications, Ltd]. eLife, 10, e59551. https://doi.org/10.7554/eLife.59551 Zanto, T. P., Rubens, M. T., Thangavel, A., & Gazzaley, A. (2011). Causal role of the prefrontal cortex in top-down modulation of visual processing and working memory. Nature Publishing Group, 14. https://doi.org/10. 1038/nn.2773 Zhang, S., Xu, M., Kamigaki, T., Hoang Do, J. P., Chang, W.-C., Jenvay, S., Miyamichi, K., Luo, L., & Dan, Y. (2014). Long-range and local circuits for top-down modulation of visual cortex processing. Science, 345 (6197), 660–665. https://doi.org/10.1126/science.1254126 119 Zipser, K., Lamme, V. A. F., & Schiller, P. H. (1996). Contextual Modulation in Primary Visual Cortex. The Journal of Neuroscience, 16 (22), 7376–7389. https://doi.org/10.1523/JNEUROSCI.16-22-07376.1996 Zong, W., Obenhaus, H. A., Skytøen, E. R., Eneqvist, H., Jong, N. L. D., & Jorge, M. R. (2021). Large-scale two-photon calcium imaging in freely moving mice, 1–62. 120