In the zebrafish, lamprey and frog tadpole the intrinsic function of the CPG providing alternation in a given segment is in principle understood [1]. The recent data from the El Manira laboratory has shown that there are three subtypes of glutamatergic excitatory interneurons (V2a-type), projecting directly to slow, intermediate and fast motoneurons [15,16,17,18]. The “slow” interneurons projecting to the slow motor neurons have the lowest threshold for activation from descending drive signals and a tendency to oscillate when depolarized, due to their intrinsic membrane properties (Figure 1 and Figure 2D). Within the slow interneuron population there is mutual excitation that reinforces and coordinates the activity in this subpopulation of interneurons, which provides the rhythmic excitatory drive to the slow motoneurons. A further increase in the descending drive signal will recruit the “intermediate” interneuron population. In addition, these interneurons receive excitation from the slow interneuron population, which allows for synchrony. The same applies with a further increased descending drive, when the “fast” interneuron population can be brought in to generate swimming at the highest speeds of locomotion. The burst termination is due to intrinsic membrane properties. Inhibitory interneurons with a commissural action are also brought in that inhibit premotor interneurons and motor neurons on the contralateral side and ascertain that when the CPG on one side is active the other side will be inhibited and vice versa. Previous less-detailed work on the lamprey and frog tadpole had shown a similar organization with populations of excitatory glutamatergic premotor interneurons responsible for providing the excitation of motoneurons and reciprocal inhibition [19,20]. The segmental lamprey network, as shown in Figure 2B, depends on pools of interacting excitatory interneurons (E in Figure 2B), which are activated by a descending drive, but also through mutual excitation between the E-interneurons (Figure 2E). The network has been simulated in great detail (Figure 2A) with the different neurons having properties similar to their biological counterparts and with biological variability introduced. The cellular mechanisms that determine burst termination are shown in Figure 2D. Essentially, during the burst the level of Ca²⁺ will increase intracellularly, which will activate calcium-dependent potassium channels, which will gradually hyperpolarize the E-interneurons. A segmental population of 100 E-interneurons generates burst activity (Figure 2F) when submitted to descending excitatory drive [12,13,21]. The alternation between the left and right side is achieved by inhibitor interneurons. The intersegmental coordination based on available connectivity has also been simulated as detailed in the references provided above.

During swimming a rostrocaudal wave is transmitted along the length of the body that pushes the animal forward through the water, and in the spinal cord there is a corresponding rostrocaudal wave of activity that results from an interaction between the segmental networks. As shown in the lamprey, a somewhat higher excitability in the lead-segment will result in a constant phase lag along the spinal cord (Figure 2F). This mechanism will produce forward swimming, but also backward swimming if the lead segment is located in a caudal position [12,13,22].

The mammalian CPG is more complex in that it needs to coordinate all the different muscles in the limb. Furthermore, the different CPGs of the four limbs can be coordinated in many different ways, as in walking, trotting and galloping. Let us now consider the CPG of a single limb. The hindlimb CPG has by far received the most attention experimentally over several decades, primarily in the cat and then in rodents (mice). The early work on the cat showed that the spinal cord could generate the complex motor pattern [7,8] even without any afferent input. The CPG could be released by activation of noradrenergic precursors [23,24,25]. To attempt to understand how the CPG is designed, recordings of rhythmically active interneurons were performed, and it was shown that inhibitory premotor neurons mediating the reciprocal inhibition between flexor and extensors, the so-called 1a interneurons, were rhythmically active and thus part of the CPG network [26,27,28,29]. Further, the Renshaw cells were active and receiving input from the CPG. However, the likelihood that one should be able to obtain sufficient rigorous knowledge of interneuronal connectivity was very low, particularly with the techniques available in the seventies. This was the reason why one of us (SG) shifted over from cat to lamprey many years ago, which allowed for paired recordings between pre- and postsynaptic neurons in the CPG.

The situation changed radically some decades ago, when it became possible to study the CPG in action in the isolated mouse spinal cord [30,31], combined with the fact that molecular techniques entered the scene. The definition of developmentally defined subtypes of interneurons V0 to V3, which included physiologically defined subtypes of inhibitory and excitatory interneurons and dorsal subtypes of precursor neurons, opened up a wealth of new possibilities [32,33,34,35,36,37]. Subtypes of interneurons could be deleted and/or optogenetically activated or inhibited, while the behavioral effects on the locomotor network could be investigated.

Despite these new possibilities and a dedicated effort of many prominent research groups such as those of Martyn Goulding, Ole Kiehn, Sam Pfaff, Tom Jessel, Rob Brownstone and many more, it has not yet been possible to crack the code of the limb CPG. The periodic inhibition of flexors is mediated by V1 interneurons and of extensors via V2b interneurons. A simultaneous inactivation of both groups abolishes the flexor–extensor alternation and makes them discharge in synchrony, but the rhythmic bursting still continues. This indicates that the burst generation is not crucially dependent on inhibition, which experiments with blockade of glycinergic inhibition had also suggested [38]. In the frog tadpole, recent experiments have indicated the burst generation requires commissural inhibition [39], however, in contrast to the situation in lamprey [40].

A wealth of information has been gathered concerning different subtypes of excitatory interneurons that when interfered with may either lower or make the rhythm more irregular. For instance, V2a interneurons provide excitation of motoneurons but is thought not to initiate the rhythmic burst activity as they do in the zebrafish. A population of non-V2a interneurons that expresses Shox2 has recently come in focus, which is composed by electrically coupled excitatory interneurons [41,42] and they may contribute to the generation of rhythmic bursting and provides excitation of V2a interneurons. An excellent and more detailed account for the activity in different subtypes of interneurons is provided by Rancic and Gosgnach [43] that are part of this volume.

The overall conclusion so far is that we have information on many subtypes of interneurons, but no overall understanding of the integrated CPG network that coordinates the whole limb with its complex motor pattern, except that V1 and V2b interneurons are required for the generation of the flexor–extensor alternation. More knowledge is required regarding connectivity between interneurons, synaptic properties and membrane properties. There is also little knowledge of the interaction between the premotor interneurons located in different segments along the spinal cord that are engaged in the control of different muscle groups.

One important question to consider is if the rhythm-generation in the isolated spinal cord of the mouse locomotor activity is located in one or two segments, or if it is distributed along the entire lumbosacral spinal cord. Locomotor activity can be induced pharmacologically, by stimulation of descending pathways or optogenetically (channelrhodopsin (ChR2)) by activation of glutamatergic interneurons, including not only V2a interneurons. Hägglund et al. [44] could show that shining light on the entire spinal cord elicited coordinated locomotor activity, while light limited to only one side of the spinal cord produced coordinated locomotor bursting only on the side exposed to light. Furthermore, when the light was limited to only one segment burst activity was produced in only that segment. Furthermore, by recording activity in single flexor and extensor nerves, they could show that exclusive flexor or extensor activity could be induced separately. Thus, discrete bursting could be induced in either a flexor or an extensor. The ability to generate locomotor-related burst activity is thus distributed along the spinal cord and not limited to one part of the spinal cord. Burst activity can be generated independently in either flexors or extensors.

These findings agree with earlier findings reported in the cat [10] showing that the spinal cord could be divided into two parts, both of which could be made to generate burst activity, and that flexor bursting could occur with tonic activity in extensors or vice versa. This led Grillner [10,45] to suggest a conceptual scheme, the unit burst generator model (UBG-model), in which each group of synergists at the different joints would be by a “unit burst generator”. Each unit CPG would be comprised by the interneurons that are presynaptic to one group of close of synergists such as the hip extensors or the ankle flexor. The different unit CPGs would interact by reciprocal inhibition as between the hip flexor UBG and that of the hip extensors, or by mutual excitation between the UBGs of hip and knee extensors. Moreover, the more complex patterns of muscle activity that occur in knee-flexors and toe extensors at the transition between main flexors and extensors at the hip could also be accounted for. We will discuss simulations of the UBG further below.

The focus of this brief account is the mammalian limb CPG; although not understood in all aspects, we feel that it is important to briefly summarize facts regarding the different parts of the CPG jigsaw puzzle—which parts are in place, and which are undefined. Below, we will consider all interneurons being presynaptic to the motoneurons or upstream that contribute to the generation of the motor pattern as part of the locomotor CPG. Within the family of interneurons some may be more important for generating the frequency of bursting and others for inhibiting or activating a specific muscle group. We will not subdivide the CPG in one layer concerned with rhythm generation and a second downstream layer providing the motor pattern, because they operate, as far as I can see, as an integrated whole.