The aims of this project were threefold; first, we wanted to examine the influence of thixotropic conditioning on each of three commonly used methods of measuring human position sense at the elbow joint and see whether it influenced position errors. Secondly, we wanted to compare position errors over the full working range of forearm movement, looking for changes in the distribution of errors at different muscle lengths. Thirdly, we wanted to compare the degree of preservation of thixotropic patterns in the position errors observed with each method.

Thixotropic conditioning alters the maintained rates of discharge in passive spindles. Thixotropy-dependent position errors are an expression of those changes in rates (Gregory et al. 1988). A first question was, did position errors generated with all three methods show evidence of thixotropy-related errors, that is, input from spindles? The answer was, “yes!”. This was an important conclusion since we don’t have any other, non-spindle explanation, for the generation of position sense (but see Gandevia et al. 2006). The second question was, “is the size of thixotropy-dependent errors always the same with each method of measurement? Here the answer was, “no!”. Why might that be? Our interpretation is that the measured value of position errors depends on two factors, the input provided by spindle discharges and the central processing of that information. Our conclusion was that with the three methods used here to study position sense, there were significant differences in the amount of central processing the position signal had undergone.

Muscle conditioning effects on position errors for two-arm matching

This is the first study of position sense in a matching task, measured over the full working range of a joint, covering 120° of forearm movement at the elbow. From inspection of the pooled data in Fig. 3a, it is apparent that for the three mid-range angles, (35°, 65°, 95°), opposite conditioning produced errors of about ± 6°, representing an error range of 12° at each test angle. Furthermore, when the reference arm was flexion conditioned (125°) and the indicator extension conditioned (5°), the matching errors always lay in the direction of forearm extension; after extension conditioning of the reference and flexion conditioning of the indicator, errors lay in the direction of flexion. The systematic changes in the direction of the errors, after the two forms of conditioning, argue in support of thixotropic influences from muscle spindles as responsible for that distribution (Gregory et al. 1988).

Early observations of what turned out to be thixotropic behaviour, referred to “post contraction sensory discharge” (Hutton et al. 1973). This was the increase in spindle discharge following a muscle contraction. We now know that this happens because of the take-up of slack in the muscle and its spindles by the contraction (Proske and Gandevia 2018). Since muscle spindles are stretch receptors signalling muscle length (Matthews 1988), in a position sense experiment, the higher spindle discharge rate after a conditioning contraction is perceived by the subject as a longer muscle; for elbow flexors, this meant a more extended forearm and for elbow extensors a more flexed forearm. It is an illusory effect, similar to that produced by vibration, since after the contraction there was the perception of a displaced forearm, yet the arm had not moved.

The regular occurrence of post-contraction sensory discharge means that in any study of position sense in a passive limb, if limb muscles are left unconditioned, measured values of position sense will be biased towards perception of a shorter, less stretched muscle than is actually the case. The extent to which this happens depends on the immediate history of contraction and length changes of the muscle. This is a frequently overlooked issue in studies of proprioception and it is known that to avoid such effects, standing torque levels of 5–10% of maximum are necessary (Jahnke et al. 1989).

We are proposing that, in the absence of muscle conditioning, the brain makes use of an established spindle discharge rate—muscle length (joint angle) relation. That is, a given forearm position is attributed to a particular, maintained rate of spindle discharge. This relationship is laid down during development, based on a young animal viewing its arm movements and perceiving the accompanying sensations generated by the movements (Held and Bauer 1967). In the adult, at a given test angle, if spindle discharge rates are raised above the expected, calibrated value for that angle by a muscle contraction, this will be misinterpreted by the brain as a longer, more stretched muscle (Banks et al. 2021, Fig. 10).

There is some evidence in support of the existence of a calibrated spindle discharge: muscle length relation. Two independent studies have reported that the increase in spindle discharge evoked by vibration can lead to perception of joint angles beyond the anatomical limit of movement at a joint (Craske 1977; Lackner and DiZio 1992). These observations suggest that as a muscle is stretched to long lengths, there is no information contained within the spindle signal alerting the brain of the approaching limit. When vibration evokes a rate of spindle discharge that implies a length beyond the limit, the discharge rate: muscle length relation operating within the limits of limb movement, is extrapolated by the brain to determine the anatomically impossible value (Craske 1977). Presumably, it is left to joint receptors to signal the approaching limits of movement at a joint (Fuentes and Bastian 2010; Proske 2023).

After co-conditioning of the reference arm in a flexed position (125°), it is moved in the direction of extension to the test angle. Its flexor muscles have been stretched by the extension movement and they are, therefore, generating high levels of spindle activity at the test angle, higher than normal, leading the subject to believe that their arm is more extended than is the case. The indicator arm comes from the opposite direction (5°) to make a match, and this time, it is the extensors which are stretched by the movement, making the subject think their arm is more flexed. At the test angle, therefore, the reference is perceived as overextended by about 3°, the indicator overflexed by 3°. In making the match, the subject stops their indicator arm too early, 6° short of the actual test angle, as a result of influences coming from both arms. The same argument can be applied if the reference is coming from 5° and the indicator from 125°, but the direction of the errors will be reversed and lie 6° in the direction of flexion.

Such reversals of errors have been observed previously. In a study of position sense measured with arm movements in the horizontal plane, with the reference arm flexion conditioned and the indicator extension conditioned, errors of 11.6° into extension were observed (Allen et al. 2007). Reversing conditioning led to 9.5° errors into flexion. Therefore, here, the total error range was 20.1°. In the present study, where position sense was measured in the sagittal plane, when the reference arm was moved from 125° to the test angle, for the three mid-range angles, the mean error was 5.9° into extension; when it was coming from 5°, it was 7.6° into flexion. This gave a total error range of 13.5°. The larger errors in the earlier study are attributed to the fact that position sense was measured in the horizontal plane, in a gravity-neutral posture. In the present experiments, we opted for the more natural situation where the subject had to bear the weight of their forearm themselves. The accompanying muscle activity (5% of maximum, Winter et al. 2005) may have led to some uptake of slack in elbow muscles, thereby reducing the size of thixotropic errors.

When an arm, conditioned at 125°, is moved into extension, its flexors are stretched; at the same time, its extensors will be shortened and, therefore, fall slack, their spindles becoming desensitised. Do the shortened extensors contribute, in any way, to matching errors? Muscle spindles are stretch receptors and they signal muscle lengthening, not shortening (Capaday and Cooke 1981, 1983; Inglis and Frank 1990; Inglis et al. 1991; see also Di Giulio et al. 2009). Therefore, when the arm is moved into extension, flexor spindles will provide the position signal, when it is moved into flexion, extensor spindles will provide the position signal, with no contributions from the slack antagonists.

While matching errors for the three mid-range angles were large, for the two extreme angles, 5° and 125°, errors were small and for 5° insignificant. This was probably because here, in the matching process, one of the arms had not moved, leaving spindles in both of its antagonists sensitised after conditioning. Movement of the other arm to make a match sensitised spindles in one of its muscles and this could be accurately matched with spindles in the stationary arm.

Identical conditioning of the two arms was used as a form of control (Fig. 3b). Here, errors were very much smaller, 2°–3°. This was to be expected since after conditioning, both arms were likely to be in a near identical thixotropic state at each test angle. It is interesting that the small size of the control errors is maintained over the full range of angles tested, including the extreme angles (Fig. 3b). There is no evidence of a change between mid-range and extreme angles. The observation supports the view that for the experiment using opposite conditioning of the two arms (Fig. 3a), a reduction in position errors at the extreme angles (5° and 125°) was, at least in part, attributable to thixotropic effects.

Muscle conditioning and position errors in a pointing task

First, it is worthwhile to recapitulate what is known about pointing and matching as two separate methods of measuring position sense. We have proposed that indicating the position of one limb by placement of the other in a matching task involves muscle spindles from muscles of both limbs (Proske and Chen 2021). Alignment of the limbs uses the frequency code of the afferents, where an increase in impulse rate is interpreted as a longer muscle and, accordingly, a more extended or flexed joint. It is an accurate mechanism, where the brain determines the degree of alignment of the arms based on differences in afferent signals between them. The mechanism does not appear to involve vision since normally the task is carried out blindfolded and visual distortions presented to the subject before the measurements do not lead to additional matching errors (Velay et al. 1989).

A second means of determining position sense is by pointing to a hidden body part. Most studies claiming a pointing task used, for example, a finger of one hand pointing to the perceived position of the equivalent finger, or other landmarks, on the hidden hand (Longo and Haggard 2010; Ingram et al. 2019). It is thought that vision of the pointing finger is an important contributor to the task. Our own work comes from a background of studies of two-arm matching using alignment of the forearms. For pointing, we wanted to measure the perceived position of the hidden forearm under conditions where there was no opportunity for proprioception in the other arm to be able to make a meaningful contribution. For placement of the pointer paddle, the position information arising from the hidden forearm was presumed to be converted to a visual frame of reference that allowed the subject to align the pointer.

The work of Velay et al. (1989) has shown that the pointing mechanism is susceptible to errors from distortions of the visual field presented immediately before a measurement. Therefore, vision is likely to play a role in the generation of this sense. In addition, there is the suggestion of a memory component in perception of the position of the hidden arm (Velay et al. 1989; Chen et al. 2021). In contrast, there is no evidence that memory plays a role in two-limb matching (Horch et al. 1975; Tsay et al. 2014).

We have previously reported two features of the position signal in pointing: it appeared to be insensitive to thixotropic conditioning of muscles and, for the forearm, pointing errors lay consistently in the direction of arm extension (Tsay et al. 2016; Chen et al. 2021). In the present study, examination of the pooled data for 11 subjects who carried out the pointing task (Fig. 4) shows that at the intermediate test angles there were differences in errors, depending on whether the arm had been conditioned at 125° or at 5°, errors which reached significance at 35°, 65° and 95°. These findings are different from previous observations on pointing (Tsay et al. 2016). In Tsay et al., measurements were made at a single test angle (40°–50°) and the arm had been conditioned at either 90° or 0°, giving a conditioning: test angle range of approximately 45°. The smaller movement range to the test angle may have contributed to the lack of significance in the distribution of position errors in our earlier study. In addition, in the present study, making the comparison between positions of the arm conditioned at opposite ends of its movement range was likely to maximise thixotropic effects. We assume that this was responsible for bringing out thixotropy-dependent errors that we had not seen previously.

In pointing, when the arm was conditioned at 125°, values tended to lie further in the direction of extension compared with after conditioning at 5°. This was a similar pattern to that seen with two-arm matching. The average pointing error into extension for the three mid-range test angles after conditioning at 125° was + 11.9°, which was larger than in the matching study (+ 5.9°). When the arm was conditioned at 5°, the average pointing error was + 2.1°. This was quite different from that for matching where the error was − 7.6°. Therefore, in the mid-range of test angles, there were differences in the ranges of the errors: 13.5° for matching and 9.8° for pointing. It meant that the outcome of the pointing experiment in the present study showed some elements of our previous observations; in pointing differences in errors between the two opposite forms of conditioning tended to be smaller. Perhaps, this was due to the fact that proprioceptive signals from only one arm were involved during pointing. This is supported by the finding that errors attributed to conditioning were similar between matching and pointing if we compare trials where only the reference arm conditioning was manipulated (Fig. 3 vs. Fig. 4b, red triangles and blue circles). In addition, pointing values lay either on the line of equality or above it, in the direction of extension, with no values lying in the direction of flexion, as had been seen in matching (Fig. 3a).

Our earlier studies of pointing suggested that in the mid-range of elbow angles, error values lay superimposed on an offset, in the direction of extension, of between 4° and 10° (Tsay et al. 2016; Chen et al. 2021). We suggest that a similar offset is present in the current pointing data and it is this which accounts for the absence of position errors below zero. The assumption implicit in such an interpretation is that the offset always lies in the direction of extension regardless of the direction of the thixotropic errors.

We do not know why such an offset is present, but it may relate to the volume of spindle afferent traffic generated in forearm antagonist muscles. A higher level of flexor activity would bias the perceived position of the forearm in the direction of extension. In the only available count of spindle numbers in human elbow muscles, flexors contained 20% more spindles than extensors (Voss 1971). Observations supporting the existence of a flexor-biased afferent signal at the forearm come from the illusory responses to vibration; vibrating elbow flexors produced illusions into extension several times larger than illusions into flexion during vibration of elbow extensors (Craske 1977; Lackner and DiZio 1992). A similar offset in perceived arm position could potentially be present in two-arm matching, but since what is measured is the difference in position of the two arms (Proske and Chen 2021), any offset in perceived position would be subtracted out in the matching process.

To summarise, while in pointing the distribution of errors showed evidence of conditioning dependent effects, differences in errors for the two forms of conditioning were smaller and errors lay further in the direction of extension when compared to matching. Here, it must be kept in mind that pointing involved afferent signals from only one arm, while matching involved both arms.

Muscle conditioning effects on position errors in a repositioning task

The main objective of the present study of position sense using the method of repositioning was to try to determine whether thixotropy played a role at all. At the outset, we had assumed that determining position sense by repositioning, where the subject was asked to remember a given test angle, involved a large memory component. If so, it suggested that the position signal we were dealing with was likely to be a more processed one than for matching or pointing. We, therefore, hypothesised that in repositioning, if there was an influence of thixotropy on position errors, it was likely to be smaller than in matching or pointing. That prediction was fulfilled.

Plots of repositioned against remembered angles (Fig. 5a, b) showed that all values of remembered angles lay close to the line of equality, no matter whether the arm was coming from the direction of flexion or extension. Furthermore, differences in values between the three conditions were small. A little more information was provided by the display of errors (Fig. 6a, b).

When the arm was coming from 125° (Fig. 6a), the error values for “none” were equal to or lay above those for “after” and “both”. Presumably for “none” a memory of forearm position was laid down, based on the perceived level of afferent activity at the test angle coming from the stretched, unconditioned, elbow flexor muscles. For the condition “after”, at the remembered angle, the level of afferent activity would have been higher because of the conditioning contraction (post contraction sensory discharge). Therefore, with the intention of reproducing the muscle length corresponding to the remembered, lower spindle discharge rate, for “after” the subject repositioned the arm at a more flexed angle, where flexor discharges were lower, leading to errors in the direction of flexion compared with “none”. Essentially, the same argument applies to “both”. Here, the subject had already incorporated into their memory the higher level of afferent activity from the conditioning contraction. Therefore, the repositioning error was similar to that for “after” and it remained different from “none”.

If the arm was coming from 5° (Fig. 6b), the muscles undergoing stretch during the movement to the test angle were the extensors. Here, the errors for “after”, would be expected to lie in the direction of extension compared with “none”, since the higher extensor activity after conditioning would indicate a more flexed angle. There was a hint of this at the most flexed angle, but the effect was weak.

There remain unexplained aspects of the error distributions in repositioning. For both the 125° and 5° starting positions, for the more flexed test angles, error values for “none”, “after” and “both”, all tended to lie in the direction of extension and they reversed in the mid-range to lie in the direction of flexion for the more extended angles. Since errors for all three conditions did this, it was unlikely that an explanation involved thixotropy.

To conclude, while evidence for thixotropic effects on position errors in the repositioning task was weak, certainly weaker than in matching or pointing, statistical analysis supported the presence of some influence of thixotropy on the errors, particularly when the starting angle was 125°. This raises the possibility that after conditioning at 125° the influence on repositioning errors from stretched flexors was greater than that from stretched extensors after conditioning at 5° (see above).

There is some evidence in the literature for an influence of muscle spindle signals on repositioning errors (Larish et al. 1984). In a forearm repositioning task in the horizontal plane, repositioning errors were larger if, during the interval between remembering and reproducing the test angle, elbow flexors were vibrated. It was concluded that vibration, a stimulus known to be selective for the primary endings of spindles, was able to interfere with the repositioning mechanism.

Wider considerations

In everyday life do we ever match the positions of our two arms? Whenever we work with both hands, we bring them together as we manipulate objects and work with tools. To be able to align the arms accurately, bringing the hands to face each other, we are likely to make use of the matching mechanism. However, we do not consciously align our arms to know where they are. If asked, without looking, we always know where each arm is separately.

In their study, Chen et al. (2021) carried out a pointing experiment where the subject indicated the position of the arm hidden behind a screen, not by pointing with the other arm, but by verbally reporting which of a series of lines drawn on the screen lay closest to the perceived position of the arm. The resultant distribution of the errors was similar to that from a standard pointing task.

Such a result leads to two conclusions; one, that in the pointing experiment, signals coming from the arm doing the pointing are not involved in locating position of the hidden arm. Secondly, if we cannot see our arm we just have to think about where it is and we know its position with reasonable accuracy. Here, presumably, at each test angle, the afferent signals coming from forearm muscles are converted in central sensory areas into muscle length: joint angle information. This information would then be forwarded to a central map indicating arm position relative to the rest of the body, as well as contributing to the sense of body ownership (Butler et al. 2017). In addition, the proprioceptive information has to be converted to a visual frame of reference to allow the subject to identify the appropriate line or move the pointer paddle to the perceived angle. All of this suggests that during a standard two-arm matching task position information from more than one source may become available at the same time; combined signals coming from the two arms during the matching, as well as information about each arm separately, as indicated in pointing.

In considering the different methods of measurement of proprioception, recently a broader view has been taken by Heroux et al. (2022). They proposed that proprioceptive assessments should be considered as low-level or high-level judgements, low-level with a single frame of reference and high-level with multiple frames of reference. According to this scheme, two-arm matching would be a low-level task involving a single, direct comparison between signals coming from the two arms. One-arm pointing would be a high-level task where the centrally recorded information from the muscles of one arm is converted to a visual frame of reference, to allow subjects to place their pointer. Repositioning would also be a high-level task: the muscle length: joint angle information is acquired and stored in memory. The remembered information has to be recalled and compared with that generated during repositioning. Therefore, we have tested three methods of measurement, each with a different level of judgement. The sizes and distribution of the observed thixotropic errors approximately follows this classification.

The present study has pointed out that if the aim of a method of measurement of position sense is to try to draw inferences about the central processing of the afferent signals, it will be important to state which method has been used. What might be the meaning of differences in expression of thixotropic errors in measurements of position sense? It seems that the central conversion of spindle impulses into position sensations can be more or less direct; direct in two-arm matching and less direct in pointing and repositioning. It presumably means that as transmission of the afferent signals progresses centrally, the position information it contains can be accessed at different points, dependent on the requirements of the method. This must be kept in mind when drawing any conclusions.