Ultimate Source of the Neural Speed Signal: Motor or Sensory or Both?
The motor circuitry directly upstream of the MLR (Garcia-Rill, 1986) seems to contain robust speed coding as well (Fig. 2A). Rate codes for speed have been reported in the motor cortex (Leinweber et al., 2017), striatum (Kim et al., 2014) and substantia nigra pars compacta (Barter et al., 2015), whereas theta and gamma oscillatory temporal codes have been reported in both motor cortex (von Nicolai et al., 2014) and striatum (Masimore et al., 2005; von Nicolai et al., 2014; but see Lalla et al., 2017). Additionally, optogenetic activation of striatal populations encoding speed modulates downstream MLR signaling (Roseberry et al., 2016) as well as locomotion (Bartholomew et al., 2016; Roseberry et al., 2016) (Fig. 1C). Taken together, this evidence suggests that internally-generated motor commands give rise to the speed signals utilized in the hippocampus and MEC for spatial representation maintenance through an efference copy-like mechanism. Corroborating this hypothesis, removing motor and/or proprioceptive cues by passively moving an animal in a clear cart around an already-run track diminishes speed modulations of grid cell firing rate as well as MEC and hippocampal theta power and frequency (Terrazas et al., 2005; Winter et al., 2015). Moreover, motor cues might also be sufficient to some degree for the maintenance of theoretically speed-dependent internal spatial representations, as animals running on a running wheel (i.e., without a true ‘environment’ to navigate) have been reported to still show hippocampal place sequences.
However, the original premise of dead reckoning maintains two possible sources from which speed information can be derived: externally- or internally-generated cues. While efference copies might represent the latter, the former is most likely represented by sensory systems encoding information such as changes in optic or tactile flow. Logic and intuition thus demand that these types of informational streams should be seriously examined as an alternative origin of hippocampal-entorhinal speed signaling. Indeed, as discussed below, movement speed is directly encoded in the sensory systems.
Optic flow speed seems to be encoded by LGN and primary visual cortical cells (Roth et al., 2016; Saleem et al., 2013; Eriksen et al., 2014; but see Niell and Stryker, 2010), while specialized cells exist in rodent barrel cortex that encode the speed at which whiskers drag along the ground (Chorev et al., 2016). A preliminary study has also reported the presence of hippocampal-entorhinal-like “speed-responsive” interneurons in the barrel cortex (Long and Zhang, 2018), inviting further investigation of this possibility. Active functioning of sensory systems seems necessary for speed signal generation as well, as complete darkness has recently been shown to disrupt speed modulation of MEC theta and grid cell activity in addition to other features of the grid cell network (Chen et al., 2016). And while it remains less well-investigated than the motor circuitry discussed in this review, there also seems to be at least one possible circuit with consistently reported speed encoding that might be able to transmit sensory-derived speed information to the hippocampal-entorhinal complex: the visual cortical areas project to posterior parietal cortex (Wilber et al., 2017; Yang et al., 2017; Miller and Vogt, 1984), which projects to the postrhinal complex (Furtak et al., 2012; Burwell and Amaral, 1998), and onto the hippocampus and MEC (Burwell and Amaral, 1998; Agster and Burwell, 2009) (Fig. 2B).
If both sensory- and motor-derived estimates of speed are indeed required to eventually generate speed signaling in the hippocampal-entorhinal complex, the two informational streams must at some point interact and influence each other to give rise to a unified speed signal. Evidence for a kind of comparison or reconciliation process has already emerged in the early visual system: Studies investigating responses to incongruent visual and running speed have noted either mismatch-based (Keller et al., 2012; Roth et al., 2016) or integrative responses (Saleem et al., 2013; Roth et al., 2016), with implications for the downstream place cell network (Chen et al., 2013).
Despite these findings, a compelling argument can be made for a somewhat deterministic influence of the motor system over sensory information in speed signal generation. Locomotion influences general and speed-specific sensory cortical processing through an efference-copy-like mechanism (Niell and Stryker, 2010; Ayaz et al., 2013; Eriksen et al., 2014; Schneider et al., 2014; Zhou et al., 2014; Dadarlat and Stryker, 2017). The motor cortex and MLR have been implicated in mediating these changes by acting through direct innervation of sensory areas (Schneider et al., 2014; Leinweber et al., 2017) and ascending basal forebrain projections (Pinto et al., 2013; Fu et al., 2014; Lee et al., 2014), respectively. The MLR might in turn be dependent upon the basal ganglia or other higher motor planning regions to mediate these changes (Roseberry et al., 2016). Moreover, various classes of units throughout the visual system are tuned to running speed and remain so in the absence of visual input (Fu et al., 2014; Pakan et al., 2016; Roth et al., 2016; Eriksen et al., 2014; Saleem et al., 2013; Christensen and Pillow, 2017), while M2 axons in V1 have “predictive” activity ramp-ups that precede locomotion initiation, lead similar responses by V1 cells, and also scale with running speed (Leinweber et al., 2017). A similar M2 projection to auditory cortex has been shown to carry an efference copy that precedes locomotion and inhibits local responses to auditory stimuli (Schneider et al., 2014; Fig 2). Lastly, initial reports claim that layer V contains the highest share of speed-tuned neurons in V1, whereas layer IV had the smallest, suggesting that the visually-derived speed signal may derive more strongly from other cortical inputs rather than from raw sensory inputs coming into layer IV from the LGN (Christensen and Pillow, 2017). Together, this evidence strongly suggests that an efference copy of the motor-derived speed signal arrives in sensory cortices through multiple pathways before a sensory-derived speed estimate can be made and thus influences that sensory-based estimate.
It seems unlikely, however, that the motor system completely dominates the sensory system’s speed signal determination; instead, the speed signal that ends up in the hippocampal-entorhinal complex is probably derived from some combination of the two sources. Recent findings have begun to strongly support this more nuanced view. Predictive motor-related signals from M2 can be modified after locomotion onset to reflect visual flow or the expected changes in visual flow based on the visual scene before locomotion onset (Leinweber et al., 2017). Furthermore, the tuning of MEC speed cells is retained in the dark with reduced specificity (Perez-Escobar et al., 2016) and can more faithfully reflect either visual or locomotive inputs during bidirectional manipulation of the gain between visual flow and running speed in a virtual reality environment (Campbell et al., 2018). Lastly, in a recent experiment examining MEC spatial encoding in the vertical plane, both rate and temporal speed signals were altered, a finding the authors attributed to a likely change in both the incoming sensory input and efference copies (Casali et al., 2019). Further investigation is thus required to fully elucidate the mechanisms by which sensory and motor input combine to create a unified speed signal, while carefully tracking the precise prospective coding latency in each relevant brain region.