Tải bản đầy đủ - 0trang
2 Anatomy and Connectivity of the Subthalamic Nucleus: From a Motor Relay Structure to a Limbic Structure
Fig. 20.6 (a) Conventional illustration of the control loop. The crucial mistake made by Cybernetics
and engineering control theory is in the assignment of the input and output of the system, as shown
in the basic control diagram (Wiener 1948). This way of illustrating the relationship between the
controller and the environment creates the appearance that the controller is some device that transforms error into output. It has led to the misunderstanding that the output is controlled, and the system
input is the command from the user. (b) Illustration of the control loop discussed in this chapter, in
which the controlled variable is the perceptual input derived from sensory receptors. So long as there
is effective control, the input signals will resemble the reference signal. The reference signal has to
specify a magnitude only, not what kind of perception is to be obtained. The kind of perception, its
relationship with the external world, is determined by the connectivity of the input function—i.e.,
whether it comes from auditory sensors or visual sensors, etc. The perceptual signal represents its
current magnitude; a separate reference signal speciﬁes how much of the perceptual signal is to be
reached. Because the system produces an output that, via sensory feedback, reduces the discrepancy
between what the input variable should be and what it is, it is capable of reaching the desired or referenced perception. Negative feedback does not mean that the sign of the feedback is negative. It
means that the feedback reduces the error. By contrast, positive feedback ampliﬁes error by increasing the discrepancy between reference and input. Here two levels are illustrated showing the principle
of hierarchical organization. The error signal from the comparator in the higher level (x + 1) is turned
into the reference signal of the lower level (x). The lowest level in the hierarchy is the ﬁnal common
path from motor neurons to muscles. The correct illustration of the control loop is shown in Fig. 20.6b.
Because output is produced by a comparison between some perception and some reference, a complete knowledge of the input will not allow one to predict the behavioral output of a closed loop
system. In a closed loop, because the input is affecting the output at the same time that the output is
affecting the input, the concept of cause or effect is simply not applicable
Neural Implementation of the Control Hierarchy
Control is here assumed to be the primary function of the nervous system. The
development of neural signaling, which provides rapid and long-range signaling as
well as analog computing using ﬁring rates, is critical for the implementation of
highly effective control systems for important biological variables. As a given controller can only control a single one-dimensional variable, to control multiple variables multiple controllers are needed. The nervous system can thus be viewed as a
collection of distinct controllers. The question is how these controllers are connected to each other. To answer this question, Powers proposed the hierarchical
control model, according to which each controller is a basic building block in a
control hierarchy (Powers et al. 1960). Each level in this hierarchy can have multiple controllers of the same type. The higher level in a hierarchy uses inputs that are
re-representations of lower level inputs. Their outputs do not directly affect effectors, but change the reference signal of the lower levels. The error signal from a
higher level can be transformed into a reference signal for the lower level, allowing
direct command of the lower levels. The higher level does not dictate how much
output the lower level should produce, but only how much input it is to obtain. To
the lower level controller itself, this amount of requested input has no particular
signiﬁcance. It is only signiﬁcant via some environmental feedback function that
links the ultimate effect of the output on the input of the higher order controller. The
order given to the lower controller has the effect of changing the value of the higher
level controlled variable in the right direction, reducing its error.
The hierarchical organization of the nervous system has long been recognized
(Fuster 1995; Hayek 1952), but the crucial error in previous models is the assumption of linear causation, rather than closed loop control (Yin 2013). For example, it
is often assumed that higher levels have cognitive functions, e.g., cognitive control
modiﬁes existing stimulus–response paths (Miller and Cohen 2001). By contrast,
here it is postulated that the higher levels send projections to the comparator function of lower controllers to alter their reference signals, thus specifying how much
input the lower level should obtain. If this basic postulate is correct, it follows that
all neurons can be classiﬁed according to their functional role in a control system—
namely input function, comparator, and output function, and that different brain
regions would correspond to different levels of the control hierarchy.
Muscle Tension Control and the Final Common Path
The cerebral cortex and the BG are situated at the top of the neural hierarchy.
Outputs from the BG do not reach the lower motor neurons directly, but they can set
the reference signals of lower levels, which ultimately result in muscle contraction.
To understand BG function, we must start from the lower levels, which interact with
the external environment. There are many types of effectors, including cells that
20 The Basal Ganglia and Hierarchical Control in Voluntary Behavior
secrete hormones and transport ﬂuids. For our purposes, by far the most important
effectors are the skeletal muscles. The ﬁnal common path, the projection from alpha
motor neurons to muscles (Sherrington 1906), is shared by virtually all behaviors.
The force generated by muscle contraction is proportional to alpha motor neuron
output. The resulting tension is sensed by Golgi tendon organs, which are in series
with the extrafusal muscle ﬁbers and produce a net inhibitory input to the alpha
motor neuron that is proportional to the amount of contraction (Houk and Henneman
1967; Houk and Rymer 2011). This control system therefore uses muscles as its
output function and tension sensors as its input function (Powers 1973a; Yin 2014c).
The comparison function is implemented by the alpha motor neuron. The error signal is the difference between tension reference dictated by the signals arriving at the
alpha motor neuron and sensed tension. Normally this control system maintains the
muscle tone speciﬁed by the sum of the descending tension reference signals. An
unexpected contraction elicits muscle relaxation, i.e., the inverse myotatic reﬂex.
Variability in output is the key feature of all control systems. There is no consistent one-to-one mapping between muscle activity and behavior. That is, from a
measure of muscle activity (e.g. EMG) we cannot tell exactly what the animal is
“doing.” Repeating output will not repeat the behavior, because the output is not the
only force responsible for observable behavior (Bernstein 1967). For example,
because the environmental disturbances change (e.g., wind), the muscle force
needed to raise one’s arm can be different every time. The output varies according
to disturbances, as deﬁned by deviations from internal references.
There is therefore a fundamental ambiguity if we only measure the output from
the nervous system. Based on output, blinking and winking might be similar, but they
differ in the variables being controlled. Behaviors can only be classiﬁed according to
their purpose or reference signals, not by their outputs. The same outputs can be used
to serve different purposes, yet different outputs can be used to serve the same purpose. Muscle tension only reﬂects the reference setting of the tension controller. It
varies in order to reach the reference value of some other variable. With the exception
of the inverse myotatic reﬂex, tension control usually serves higher purposes.
Muscle Length Control
The level just above the tension controller is a muscle length controller. Muscle
length and muscle tension are independently controlled variables. Both can vary at
the same time, but there is an intrinsic hierarchical relationship between these two
variables: tension regulation is the means by which muscle length control can be
achieved (Yin 2014c).
The length error signal is conveyed by the Ia primary afferent from the muscle
spindle. This signal is traditionally viewed as a sensory signal conveying information about muscle length, because the muscle spindle, being in parallel with the
extrafusal ﬁber, is activated whenever the extrafusal muscle is elongated. In fact, the
Ia afferent signal is independent of actual muscle length. It can be produced either
as a result of the extrafusal muscle ﬁbers in parallel with the spindle, or as a result
of the contraction of intrafusal muscles due to gamma neuron activity. It reﬂects the
difference between desired muscle length and actual length—the error signal in the
length controller. The spindle acts as a mechanical comparator, as a stretch detected
by muscle spindle is compared with the net reference signal for muscle length. The
discrepancy or error is then sent to the alpha motor neuron to produce muscle contraction. At the same time, however, the alpha motor neurons can also be commanded directly by corticospinal projections. Such direct adjustment of force
reference seems particularly important for the movement of distal digits.
The descending signals for length reference come from the gamma motor neurons, which in turn receive inputs from the brainstem and the cerebral hemispheres.
When gamma motor neurons are activated, the contractile parts of the spindle (intrafusal ﬁbers) attempt to shorten. A pull is generated at the equatorial region of the
spindle that results in Ia afferent activity, even though the spindle length does not
change signiﬁcantly because it is anchored at both ends. Consequently, alpha motor
neurons are activated.
Position Control: Joint Angle and Body Configuration
To change a joint angle, the lengths of multiple muscles must be changed simultaneously. In turn, a body conﬁguration consists of a set of joint angles. In each case,
the relevant perceptual input is represented as a one-dimensional signal. The controlled variable is a conﬁguration of lower order proprioceptive inputs, and the
output function can reach a group of muscles that work together to produce the
appropriate net effect.
The control of body conﬁgurations is a type of position control. In position control, the controlled variable represents some position coordinate, and output is generated by computing the difference between the reference or desired position coordinate
and the input signal reporting the actual position. For motion with multiple degrees
of freedom, the actual position vector is determined by the action of multiple orthogonal controllers. In the brainstem, for example, there is evidence for distinct position
controllers for vertical and horizontal movements (Deliagina et al. 2012; King et al.
1981; Luschei and Fuchs 1972; Masino 1992; Masino and Knudsen 1990).
The key neural substrates for posture control are the reticulospinal, vestibulospinal, and rubrospinal pathways (Deliagina et al. 2008; Foreman and Eaton 1993;
Peterson et al. 1979). For example, stimulation of the reticulospinal pathway can
produce coordinated changes in joint angles: depending on stimulation location,
ipsilateral ﬂexion and contralateral extension or the opposite pattern of ipsilateral
extension and contralateral ﬂexion can be produced (Sprague and Chambers 1954).
The reticulospinal pathway receives direct and indirect projections from the SNr.
These descending projections are assumed to alter the reference signals for body