MAIN TEXT
Reliable and adaptable navigational capabilities are essential for
nearly all animal species. Animals often must take complicated paths
through their environments and move at a wide range of speeds. Despite
this, most species are remarkably successful at navigating complex
environments while simultaneously perceiving sensory stimuli that might
alert them to rewards or predators. Contemplating how animals might
possess these impressive navigational abilities, Darwin suggested a
strategy he termed “dead reckoning”. The theory proposed that by
combining internal and external motion cues to continuously estimate
speed and direction, animals could adequately track their current
position relative to a starting point (Darwin, 1873; Barlow 1964). Dead
reckoning is now commonly referred to as path integration and has taken
on a somewhat more restricted definition, focused primarily on the use
of internally generated (idiothetic) neural signals (Whishaw et al.,
2001; Whishaw and Wallace, 2003; Etienne and Jeffery, 2004; Buzsáki
2005; McNaughton et al., 2006; Buzsáki and Moser, 2013; Chrastil 2013;
Geva-Sagiv et al., 2015; Finkelstein et al., 2016; Igarashi 2016;
Grieves and Jeffery, 2017; Moser et al., 2017). Mammals were first
confirmed to utilize path integration in navigation nearly forty years
ago (Mittelstaedt and Mittelstaedt, 1980), and multiple brain regions
have since been implicated in this function (McNaughton et al., 1996;
2006; Whishaw et al., 1997; 2001; Whishaw and Wallace, 2003; Etienne and
Jeffery, 2004; Parron and Save, 2004; Nitz 2006; Wolbers et al., 2007;
Moser et al., 2008; 2017; Whitlock et al., 2012; Wilber et al., 2017).
How do neurons computationally represent direction and speed, as
required by path integration theories? In rodents, Taube and colleagues
have found head direction cells: assemblies of neurons, residing in many
navigationally important regions. These neurons integrate vestibular,
proprioceptive and other meaningful input to fire only when the animal’s
head points in a preferred orientation (Taube et al., 1990a; b; Stackman
et al., 2002; Peyrache et al., 2015). A number of reviews cover head
direction in exquisite detail (Sharp et al., 2001; Taube 2007; Yoder and
Taube, 2014; Grieves and Jeffery, 2017; Moser et al., 2017; Campbell and
Giocomo, 2018), and here we will focus on the neural representation and
control of linear running speed. Neural activity patterns associated
with locomotion have been studied in a variety of mammals and brain
regions for decades (e.g., Green and Arduini, 1954), yielding a
multitude of observations that can sometimes be difficult to reconcile.
The present review aims to synthesize these wide-ranging findings with
the goal of providing a clearer understanding of the mechanisms
underlying mammalian speed encoding. We also highlight some of the
critical questions that still need to be answered to paint a
comprehensive picture of how neural codes for running speed enable
successful spatial navigation.
Running speed plays a central role in broader theories of spatial
cognition. The known circuitry of the brain’s so-called ‘cognitive map’
is formed most prominently by two cell types: hippocampal place cells
and entorhinal grid cells. Place cells are pyramidal cells in areas CA1
and CA3 of the hippocampus that selectively fire in one (or sometimes
two) locations within a given environment (O’Keefe and Dostrovsky, 1971;
Wilson and McNaughton, 1993; Moser et al., 2008; Grieves and Jeffery,
2017). Medial entorhinal cortical (MEC) grid cells (stellate and
pyramidal cells in layers 2 & 3) fire in a similar but repeating manner
such that their firing fields produce a tessellating geometric grid over
a given environment (Fyhn et al., 2004; Hafting et al., 2005; Moser et
al., 2008; Grieves and Jeffery, 2017). For spatially invariant
representations to be continuously updated in a manner consistent with
the subject’s movement, the place cell-grid cell network must have
access to speed information among other self-motion metrics (Moser et
al., 2008; 2017; McNaughton et al., 2006). We begin the present review
with a discussion of how speed information appears to be encoded in
these two structures before shifting to an examination of the upstream
circuitry and computations that may provide this network with
speed-modulated inputs.