Clinical Biomechanics 15 (2000) 176±187

www.elsevier.com/locate/clinbiomech

Cruciate ligament forces in the human knee during rehabilitation

exercises
D.E. Toutoungi a, T.W. Lu d,1, A. Leardini c, F. Catani c, J.J. OÕConnor b,*
a
Cambridge Consultants Ltd., Science Park, Milton Rd, Cambridge, UK
b
Oxford Orthopaedic Engineering Centre, Nueld Orthopaedic Centre, University of Oxford, Windmill Road, Headington, Oxford OX3 7LD, UK
c
Movement Analysis Laboratory, Istituto Orthopedico Rizzoli, Bologna, Italy
d
School of Physical Therapy, China Medical College, Taichung 404, Taiwan, ROC
Received 29 July 1997; accepted 17 August 1999

Abstract
Objective. To determine the cruciate ligament forces occurring during typical rehabilitation exercises.
Design. A combination of non-invasive measurements with mathematical modelling of the lower limb.
Background. Direct measurement of ligament forces has not yet been successful in vivo in humans. A promising alternative is to
calculate the forces mathematically.
Methods. Sixteen subjects performed isometric and isokinetic or squat exercises while the external forces and limb kinematics
were measured. Internal forces were calculated using a geometrical model of the lower limb and the ``dynamically determinate one-
sided constraint'' analysis procedure.
Results. During isokinetic/isometric extension, peak anterior cruciate ligament forces, occurring at knee angles of 35±40°, may
reach 0.55 body-weight. Peak posterior cruciate ligament forces are lower and occur around 90°. During isokinetic/isometric
¯exion, peak posterior cruciate forces, which occur around 90°, may exceed 4 body-weight; the anterior cruciate is not loaded.
During squats, the anterior cruciate is lightly loaded at knee angles up to 50°, after which the posterior cruciate is loaded. Peak
posterior cruciate forces occur near the lowest point of the squat and may reach 3.5 body-weight.
Conclusions. For anterior cruciate injuries, squats should be safer than isokinetic or isometric extension for quadriceps
strengthening, though isokinetic or isometric ¯exion may safely be used for hamstrings strengthening. For posterior cruciate injuries,
isokinetic extension at knee angles less than 70° should be safe but isokinetic ¯exion and deep squats should be avoided
until healing is well-advanced.

Relevance
Good rehabilitation is vital for a successful outcome to cruciate ligament injuries. Knowledge of ligament forces can aid the
physician in the design of improved rehabilitation protocols. Ó 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Cruciate ligament forces; Rehabilitation; Squats; Isometrics; Isokinetics; Geometrical models

1. Introduction content of rehabilitation programs can di€er consider-

ably: some physicians advocate early active mobilisat-
Knee ligament injuries are common and potentially ion [3], others recommend only passive motion in the
serious. A crucial part of the treatment process fol- early stages [4], some include single joint, resisted leg
lowing ligament injury is an exercise-based rehabilita- extension and ¯exion exercises (such as isometrics and
tion period. The importance of the rehabilitation phase isokinetics) relatively early [5], while others prefer
is widely acknowledged [1,2]. However, the exact multi-joint exercises (such as squats and leg-presses) [2].
One reason for the disagreement as to what constitutes
the best rehabilitation program for any particular case
* is the lack of information about the forces carried by
Corresponding author.
E-mail address: john.oconnor@eng.ox.ac.uk (J.J. OÕConnor).
the ligaments or ligament grafts during the various
1
Present address: Institute of Biomedical Engineering, National exercises. Knowledge of the ligament forces occurring
Taiwan University, 1 Sec. 1, Jen-Ai Road, Taipei 100, Taiwan. during di€erent exercises would aid the clinician in the
0268-0033/00/$ - see front matter Ó 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 2 6 8 - 0 0 3 3 ( 9 9 ) 0 0 0 6 3 - 7
 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187
design of a suitable rehabilitation programme for a
particular injury.
Determining the forces carried by the knee ligaments
during activity is a dicult task. A few authors have
attempted direct measurement of ligament forces
in vitro, under various load conditions. The methods
used have included attaching buckle transducers to the
ligaments [6±8] or attaching load cells to mechanically
isolated bone plugs containing the tibial attachments of
the anterior cruciate ligament (ACL) and posterior
cruciate ligament (PCL) [9,10]. The main problem with
in vitro studies is that it is very dicult to replicate the
loading environment that the knee joint experiences in
vivo during activity. Also, since the methods are invasive
and destructive, they cannot be used to make in vivo
force measurements. One method that has been applied
in vivo in goats, by Holden et al. [11], required subse-
quent in vitro calibration. Similar calibration is obvi-
ously not possible for the strain measurements made
in human knees by Beynnon et al. [12].
An alternative approach is to calculate ligament
forces mathematically, starting from non-invasive in vivo
experimental measurements. Several authors have used
this approach to determine muscle forces and joint shear
and compressive forces during various rehabilitation
exercises, including isometric extension and ¯exion
[13,14], isokinetic extension and ¯exion [15±17] and
squats [18,19]. All these studies, except that of Kaufman
[16] on isokinetic exercise, were two-dimensional. All
reported the tibio±femoral shear and compressive forces
but none proceeded to determine the magnitudes of the
resultant ligament forces.
Zavatsky and OÕConnor [20] carried out a theoreti-
cal study of isometric quadriceps contractions, in which
they calculated the values of the ligament forces. These
were given as proportions of an arbitrary restraining
load, for various di€erent points of application of the
load. They used a geometrical model of the knee joint
(including continuous extensible ligaments) that gave
the lines of action of all muscles, ligaments and contact
forces for any given knee con®guration. Imran and
OÕConnor [21] used the same model to demonstrate the
modest level of hamstring forces required to compen-
sate for a ruptured ACL. Ligament forces were also
calculated during gait in an experimental study by
Collins and OÕConnor [22,23], who recorded the
ground reaction force and limb kinematics continu-
ously through a gait cycle. These data were used to
determine the resultant forces and moments at the
lower limb joints. A similar model to that used by
Zavatsky and OÕConnor, though including the whole
lower limb, was then used as the basis of the calcula-
tion of the internal forces. The indeterminacy of the
lower limb was overcome using a method that they
called the limiting solutions approach. The values of
the ligament forces found in this study were thought to
177
be too large because the extensibility of the ligaments
was not taken into account [24].
The aim of the present study was to calculate the
forces carried by the cruciate ligaments during a number
of common rehabilitation exercises, in particular to de-
termine the peak ligament forces and the ranges of
motion over which each ligament was loaded. The ex-
ercises studied were isometrics, isokinetics and squats.
The methodology used was similar to that of Collins and
OÕConnor [22,23], except that the more realistic model
of the cruciate ligaments, developed by Zavatsky and
OÕConnor [25,26], was incorporated into the knee
model.
2. Experiments
Two sets of experiments were carried out: in Set 1,
isometric and isokinetic exercises were investigated, and
in Set 2, squats were investigated. For each set, eight
normal subjects were used. The subjects in Set 1 had
mean age 26.6 years (range 22±35) and mean weight
72 kg (range 60±89 kg). Those in Set 2 had mean age 29
years (range 23±35) and mean weight 78 kg (range 70±
89 kg). Ethical approval was obtained from the Central
Oxford Research Ethics Committee for all the experi-
ments carried out. All the subjects were healthy, active
people with no history of knee pathology. In all cases,
the right leg was tested.
2.1. Equipment
The isokinetic and isometric exercises were carried
out using a KinCom dynamometer (Chattecx, Chatta-
nooga, TN, USA). The dynamometer controlled the
angular velocity at which the subjectsÕ legs were allowed
to move. It also incorporated a uniaxial load cell, which
was used to measure the main component of the resist-
ing force, the component perpendicular to the dyna-
mometer arm in the sagittal plane. It was not possible to
measure the other components of the resisting force and
the e€ect that neglecting these components may have
had on the results is discussed later (Section 5.3). The
KinCom recorded the force data at a rate of 100 Hz.
The squat exercises were carried out with one foot on an
AMTI force platform (Advanced Medical Technology,
MA, USA), which was used to measure the ground re-
action force acting on the foot. In both sets of experi-
ments, a VICON370 optometric motion analysis system
(Oxford Metrics, Oxford, UK) was used to record the
trajectories of re¯ective markers attached to the subjectsÕ
limb segments at a sampling rate of 50 Hz. Four
markers, placed over key anatomical landmarks, were
used on each of the pelvis, thigh, shank and foot (i.e., a
total of 16 markers on the limb). For the isokinetic and
isometric experiments, two markers were also placed on
178 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187
the rotating arm of the dynamometer, one level with the
load cell.
In the squat experiments, electromyography was used
to record the activity of six muscles: rectus femoris, bi-
ceps femoris, medial head of gastrocnemius, gluteus
maximus, tibialis anterior and tensor fasciae latae. The
MA100 EMG system (Motion Lab Systems, Tampa,
FL, USA) was used for this purpose. The electrodes
were Biotrace Bio-Adhesive Ag/AgCl electrodes, which
are ¯exible, pre-gelled circular electrodes with a diame-
ter of 1 cm. Two electrodes were placed over each
muscle belly with their centres 2 cm apart. The skin was
prepared by rubbing with alcohol prior to axing the
electrodes. The EMG was sampled at a rate of 1000 Hz.
It was not possible to record EMG during the isokinetic
experiments as the necessary equipment could not be
moved to the building where the dynamometer was
located.
2.2. Experimental protocol
2.2.1. Isometrics and isokinetics
The subjects were seated on the dynamometer. The
back rest was inclined at approximately 30° to the ver-
tical. The pelvis and distal end of the thigh were strap-
ped ®rmly down. The subject was positioned with the
right lateral epicondyle aligned with the axis of the dy-
namometer arm and the lower leg was attached to the
dynamometer arm via a cu€ fastened ®rmly just above
the ankle. For the isokinetic trials the subjects were
instructed to extend and ¯ex as hard and as fast
as possible. Each subject carried out continuous con-
centric isokinetic extension and ¯exion at 60°/s, 120°/s
and 180°/s. For each speed, the subjects were allowed a
warm-up period to accustom themselves to the exercise.
Then, after a few minutes of rest, two trials were
recorded. Each trial consisted of 5 cycles of ¯exion and
extension. The order in which the speeds were
performed was randomised.
Next, the isometric trials were carried out. The knee
¯exion angles selected were 15°, 30°, 45°, 60° and 75°.
The subjects were instructed that in each position they
should extend or ¯ex as hard as possible for 3 s and then
relax. Two trials of ¯exion and two of extension were
recorded. The order of ¯exion and extension was ran-
domised.
Finally, a few frames of VICON data were recorded
with the subject standing in a relaxed, upright posture.
The data from this static trial were used when it was
necessary to reconstruct the position of a marker that
had been obscured during the exercise trials.
2.2.2. Squats
First, a manual voluntary maximum contraction
(MVM) test was carried out on each muscle from which
EMG was being collected. This was to provide data used
for normalising the EMG data collected during the ex-
ercise trials. The MVM tests were performed according
to standard testing procedure, described in clinical
textbooks [27].
The subjects were then asked to stand with their feet
parallel and shoulder-width apart, and with one foot
only on the force platform. They were instructed to
perform squats by bending the knees while keeping the
back straight, and to bend the knees no further than was
comfortable. Each subject performed two types of two-
legged squat: one with the heels remaining in contact
with the ground (referred to as a heel-on-ground (HG)
squat), and one allowing the heels to lift o€ the ground
(referred to as a heel-o€ (HO) squat). In addition, those
subjects who were able also performed a one-legged
(OL) squat, with heel on the ground. The subjects were
allowed to hold their arms out in front or to the side to
aid balance. A number of practice squats were per-
formed, then data from two trials of each type of squat
were collected, in a randomised order. Each trial con-
sisted of a single down±up movement. A static trial was
also collected, as during the isokinetic experiments.
3. Internal force calculation
The process by which ligament forces were calculated
can be broken down into two steps. The ®rst step was to
calculate the resultant intersegmental forces and mo-
ments at each joint using inverse dynamics. The second
step was to determine how these resultant forces and
moments were distributed among the force-bearing
structures at the joint.
3.1. Calculation of resultant intersegmental forces and
moments
The lower limb was taken to comprise four linked
rigid body segments: the pelvis, thigh, shank and foot.
The positions of the joint centres, about which moments
were calculated, relative to each segment were deter-
mined from the geometrical model (see below). The re-
sultant intersegmental forces and moments about the
ankle, knee and hip joint centres [23] were calculated by
applying Newtons' equations of motion in two dimen-
sions. For these calculations, the external forces acting
on each body segment, the mass and moment of inertia
of each segment and linear accelerations of the centres
of mass and angular accelerations of each segment must
be known. The external forces were known from the
measurements described above. The segment masses and
moments of inertia, and the co-ordinates of the centres
of mass relative to their respective body segments were
derived from measurements of body mass and segment
lengths, using anthropometric data tables given by
Winter [28]. From the positions of the markers, the
 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187 179

co-ordinates of the segment centres of mass relative to

the laboratory and the segment angles were calculated.
The accelerations of the segments were determined by
dierentiation, using a splining algorithm based on the
routines published by Woltring [29]. The options used
were for quintic splines, calculated using the least-
squares ®t mode with a prior given variance of 0.02.

3.2. The geometrical lower limb model

The various components of the lower limb model

used for the calculation of internal forces have been
described in detail in several publications. The basic
lower limb model was that used by Collins [23] and Lu
et al. [30]. It is a two-dimensional sagittal plane model
consisting of the four rigid body segments mentioned
above. The segments are linked by models of the hip,
knee and ankle joints. The hip and ankle are modelled as
pin-joints, with the ¯exion axes passing through the
centre of the femoral head and the centre of curvature of
the articular surface of the talus (modelled as part of a
circle), respectively. The knee joint model is a single
degree of freedom joint with a moving ¯exion axis,
based on a planar four-bar linkage [25,31±33]. The lower
limb model includes seven muscles: gluteus maximus,
iliopsoas, quadriceps, hamstrings, gastrocnemius, soleus
and tibialis anterior. Rectus femoris and the vastii were
assumed to act simultaneously, with a force ratio of 1:3.
The patello-femoral joint model developed by Gill and
OÕConnor [34] was included. Muscles were represented
as straight lines running from origin to insertion, except
that changes in direction caused by wrapping around the
bones were taken into account. The two cruciate liga-
ments were modelled as continuous arrays of extensible
®bres. A diagram of the lower limb model is given in
Fig. 1.
3.3. Force distribution
The intersegmental forces and moments are the re-
sultants of the internal forces transmitted by the mus-
cles, ligaments and articular surfaces of the joints. A
series of equipollence equations relating the internal
forces to the resultant forces and moments were written.
However, due to the large number of structures able to
transmit internal forces that are present in the lower
limb, the system is indeterminate and the equations
cannot be solved directly. The use of the geometrical
model reduces the level of indeterminacy considerably,
by de®ning the lines of action and moment arms of all
the forces, leaving only the magnitudes of the forces to
be determined.
In order to overcome the indeterminacy, a method
called the DDOSC (dynamically determinate one-sided
constraint) method [30] (originally known as the limiting
solutions method [22]) has been developed.
Fig. 1. Representation of the two-dimensional geometrical lower limb
model used in the study.
The method has been described in detail in previous
publications [22,23,30]. Brie¯y, if the system contains n
unknowns, but is only described by m equations, then all
possible combinations of m out of the n unknowns were
created 2 and the equations solved for each combination
in turn. Only physiologically possible solutions, with
muscle and ligament forces tensile and contact forces
compressive, were accepted. The admissible solutions
were then compared with phasic EMG data and the
solution most closely matching the actual combination
of inactive and active muscles at the instant was selected.
This procedure was repeated for each frame of data.
A modi®cation of the procedure was used in this
study. The extra degree of freedom at the knee (a±p
tibial translation) allowed by extension of the ligaments
introduced an extra unknown, the tibial displacement d
relative to its passive ¯exion position at a given knee
¯exion angle. This quantity cannot easily be measured
both because it is such a small displacement and because
it is dicult in practice to de®ne the passive ¯exion
position relative to which d is measured. Instead, d was
calculated, using an iterative process, as follows:
1. For each combination of forces in turn, an initial val-
ue for d was chosen and the cruciate ligament force
(magnitude and line of action) and the lines of action
2
Subject to the constraint that there were at least three force
components at each joint.
180 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187
of all other forces corresponding to this value were
determined.
2. A new position for the centre of the knee joint was
de®ned, at the intersection of the resultant ligament
force and the tibio±femoral contact force, and the re-
sultant knee moment recalculated.
3. The equipollence equations were solved for that par-
ticular combination.
4. There were then two values for the ligament force:
one determined directly from the ligament model
and one determined from the solution of the equip-
ollence equations. These two values were compared.
5. When the two values di€ered by more than 2%, the
initial estimate of d was increased or decreased as nec-
essary and the process repeated until the two values
for the ligament force agreed. When this occurs, all
the equations describing the system are satis®ed and
the solution is unique.
This process was repeated for all combinations in-
volving a cruciate ligament force. Other combinations
were treated according to the original method (equiva-
lent to letting d equal 0). As before, the admissible so-
lutions for each data frame were compared with EMG
data in order to select the most realistic solution. When
no exact match with the active muscles was found, the
closest possible solution was selected. When there were
several equally close matches, the one which predicted
the highest value for the ligament force was chosen. This
meant that the ligament forces were not underestimated.
In the analysis of isokinetics and isometrics, it was
only possible to calculate the internal forces at the ankle
and knee joints, as there were unknown forces acting on
the thigh (the reaction with the dynamometer seat). In
this case, there were 10 unknowns and six equations
(n ˆ 10; m ˆ 6), rather than 14 unknowns and nine
equations (n ˆ 14; m ˆ 9) in the full leg model used to
analyse the squats.
3.3.1. EMG processing
In the isokinetic/isometric experiments, it was not
possible to collect simultaneous EMG data. Therefore
data collected in an earlier series of experiments on
isokinetics [35] were used as a guide to muscle activity
Fig. 2. Graphical view of where during the range of motion the selected DDOSC solutions occurred for the di€erent exercises. IK ˆ isokinetic,
IM ˆ isometric. The force components involved in each solution are labelled as follows: A ˆ ACL force, P ˆ PCL force, T ˆ tibialis anterior force,
G ˆ gastrocnemius force, S ˆ soleus force, Q ˆ quadriceps force, H ˆ hamstrings force, C ˆ contact force magnitude, D ˆ contact force direction.
patterns. The EMG data collected during the squat ex-
periments were processed by ®rst rectifying and ®ltering
using a fourth order digital low-pass, zero-phase But-
terworth ®lter with a cut-o€ frequency of 10 Hz [28]. The
resulting signal envelope is termed the mean absolute
value (MAV). The same process was applied to the data
collected during the MVM contraction tests and then
the MAVs from each of these trials were averaged over
0.5 s of the test to give a single threshold value. The
time-varying MAVs determined from the squat trials for
each muscle were then compared with the appropriate
threshold value, and the muscle was taken to be active
when its MAV exceeded 5% of the threshold value [36].
3.4. Data analysis
For each exercise, data from the trial with the most
complete data set were selected for analysis. For the
isokinetic trial, one representative ¯exion-extension cy-
cle was then selected. In the other exercises, each trial
consisted of only one cycle. Internal forces were calcu-
lated at 2.5° increments of knee ¯exion angle. At each
angle, the mean values and standard errors in the mean
over all subjects were calculated. In each case, the mean
peak ligament forces were also calculated. Two-factor
ANOVAs were used to determine whether there were
signi®cant di€erences between the peak forces occurring
at di€erent isokinetic speeds or during di€erent types of
squat.
4. Results
4.1. DDOSC solutions selected
Fig. 2 gives a graphical display of DDOSC solutions
found to be the best match at each point during the
various exercises. During isokinetic extension the
combination of tibialis anterior, ankle contact force
(and direction), quadriceps, ACL and knee contact
force (combination denoted TCD±QAC) best matched
the EMG, except at ¯exion angles greater than around
70°, where the PCL, rather than the ACL, was in
 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187
Fig. 3. Cruciate ligament forces during isokinetic exercise. Bold line shows mean of eight subjects, shading indicates one standard error from the
mean. Solid lines show ACL forces, dashed lines show PCL forces. Left: ¯exion, right: extension. Top to bottom: 60°/s, 120°/s, 180°/s.
tension (TCD±QPC). During isokinetic ¯exion, the best
match was tibialis anterior, ankle contact force (and
direction), hamstrings, PCL and knee contact force
(TCD±HPC). The same combinations were selected
during isometrics.
These extension solutions were an exact match with
the EMG. The ¯exion solutions were a close match,
except that they did not include gastrocnemius, which
EMG data (collected in a previous study [37]) showed to
be active. This means that the PCL forces presented here
for isokinetics and isometrics give an upper bound on
the actual PCL force, since gastrocnemius works syn-
ergistically with the PCL.
During squats, there were three phases of solutions.
With the leg nearly straight, the best match with EMG
consisted of tibialis anterior, gastrocnemius, ankle
contact force, quadriceps, ACL, knee contact force,
hamstrings and hip contact force (and direction)
(TGC±QAC±HCD). There was then usually a short
phase at knee angles around 40±70° where the solution
was the same except that the PCL, rather than the
ACL, was in tension (TGC±QPC±HCD). Finally, at
high ¯exion angles, the solution selected was the same
as in the previous phase, but with a soleus rather than
gastrocnemius force (TSC±QPC±HCD). The true situ-
ation at the high ¯exion angles, according to the EMG
181
data, was that both soleus 3 and gastrocnemius were
acting at the ankle and that gluteus maximus, as well as
hamstrings was active at the hip. The above solution
was chosen because it was the closest match possible
that did not underestimate the PCL force (additional
gluteus maximus activity, which reduces the hamstrings
force, or gastrocnemius activity would both tend to
reduce the PCL force). Therefore, during squats, the
reported PCL forces are also upper bounds on the
actual forces.
4.2. Ligament forces
The mean cruciate ligament forces occurring during
the three exercises are shown in Fig. 3 (isokinetics), Fig. 4
(isometrics) and Fig. 5 (squats). During isokinetic ex-
tension, the ACL was loaded at knee angles from 0° to
80°, with the PCL loaded at higher knee angles. The
peak ACL force tended to occur at knee angles of 35±
40° and the peak PCL force at around 90°. During
isokinetic extension, the PCL was loaded throughout the
3
EMG data were recorded from soleus using wire electrodes during
a separate series of squat exercises performed by one of the authors
(DT).
182 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187

Fig. 4. Cruciate ligament forces during isometric exercise. Bold line shows mean of six subjects, shading indicates one standard error from the mean.
Solid lines show ACL forces, dashed lines show PCL forces. Left: ¯exion, right: extension.

range of motion, with the peak force occurring again at
around 90°. There was no force in the ACL during
¯exion. The patterns for isometric extension and ¯exion
were similar to those during isokinetics. During squats,
the ACL experienced small forces when the knee angle
was less than around 50° (these do not show on the
graphs in Fig. 5 due to the scale of the y-axis). At greater
knee angles, the PCL was loaded, with the peak force
occurring near the lowest point of the squat.
The mean peak cruciate ligament force values are
shown in Table 1. During isokinetic extension, the peak
ACL force decreased signi®cantly (A N O V A , p < 0.005)
with increasing exercise speed. The decrease in peak
PCL force with increasing speed during isokinetic ¯ex-
ion was also signi®cant (A N O V A , p < 0.01). For the
squats, the di€erence in peak PCL force between the
three types of squat was signi®cant (A N O V A , p < 0.05)
during the descending phase (knee ¯exion angle in-
creasing), but not during the ascending phase (knee
¯exion angle decreasing). All other di€erences were not
signi®cant.
5. Discussion
A knowledge of ligament forces during activity could
be useful for a number of purposes. As well as aiding the
design of rehabilitation programs, such information

Fig. 5. Posterior cruciate ligament forces during squats. Bold line shows mean of eight subjects (six for OL squats), shading indicates one standard
error from the mean. Left: sinking down phase, right: rising up phase. Top to bottom: heel o€ squat, heel-on-ground squat, one-legged squat.
 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187 183

Table 1
Mean peak ACL and PCL forces occurring during the exercisesa

Exercise type Peak ACL force Peak PCL force

(N) (BW) (N) (BW)

Isokinetics
Extension 60°/s 349 (110) 0.48 (0.15) 74 (72) 0.10 (0.10)
120°/s 325 (72) 0.45 (0.10) 59 (61) 0.08 (0.08)
180°/s 254 (91) 0.35 (0.13) 55 (42) 0.08 (0.06)
Flexion 60°/s ± ± 2701 (719) 3.8 (1.0)
120°/s ± ± 2394 (775) 3.3 (1.1)
180°/s ± ± 1952 (731) 2.7 (1.0)
Isometrics
Extension 396 (106) 0.55 (0.15) ± ±
Flexion ± ± 3330 (1060) 4.6 (1.5)
Squats
HO squat Descent 95 (40) 0.12 (0.05) 2113 (217) 2.7 (0.3)
Ascent 49 (57) 0.06 (0.07) 2222 (300) 2.8 (0.38)
HG squat Descent 26 (31) 0.03 (0.04) 2432 (819) 3.1 (1.1)
Ascent 28 (36) 0.03 (0.05) 2704 (805) 3.5 (1.0)
OL squat Descent 117 (85) 0.15 (0.11) 1912 (665) 2.5 (0.9)
Ascent 142 (67) 0.18 (0.09) 2246 (659) 2.9 (0.8)
a
Values are expressed in Newtons (N) and as a fraction of body-weight (BW). Standard errors in the mean are given in brackets.

might help with the prevention of injury, or be taken
into account in the design of arti®cial ligament grafts or
the choice of tissue for a biological graft. As discussed
earlier, direct measurement of ligament forces in vivo is
very dicult and data from measurements made in vitro
cannot easily be extrapolated to the in vivo situation.
Therefore there is a need for an alternative means of
determining ligament forces during activity.
A procedure involving non-invasive experimental
measurements combined with geometrical modelling of
the limb has been used in this study. It di€ers from
methods used in previous experimental studies of reha-
bilitation exercises [13±19] in that it allows the calcula-
tion of the resultant ligament force magnitudes and
directions, rather than just the tibio±femoral shear and
compressive forces. The shear force does relate to the
cruciate ligament force, since the cruciates are the pri-
mary structures responsible for resisting the shearing
induced by muscle action or external loads. Thus the
shear force, provided the contribution of muscles has
been deducted, can indicate which ligament is loaded,
and give a rough indication of the magnitude of the
load. However, the ratio of resultant cruciate ligament
force to shear force changes through the range, because
the direction of pull of the ligament changes with ¯exion
angle and tibial displacement (which is dependent on the
applied load). Therefore, peak ligament forces are not
necessarily simultaneous with peak shear forces, nor
proportional to them.
Although there have been a few previous studies in
which ligament forces have been calculated from in vivo
data [22,38], these studies were primarily concerned
with muscle force magnitudes, and included only
simple representations of the ligaments as one or two
line elements. In this study, the ligaments are repre-
sented more realistically as continuous arrays of ex-
tensible ®bres.
5.1. Comparison with previous studies
Since none of the previous studies of these exercises
determined ligament force magnitudes, direct compari-
son of the ligament force results with published work is
not possible. However, the shear forces reported in those
studies should be of a similar order of magnitude,
though somewhat smaller than the calculated ligament
forces, as the shear force is equal to the ligament force
times the cosine of its angle of inclination to the tibial
plateau. In Table 2, the shear forces reported by other
authors in the exercises studied here are shown. An
anterior shear force here means that the tibia tends to
move anteriorly relative to the femur, thus loading the
ACL. The opposite is true for a posterior shear force.
These values can be compared with the values given in
Table 1 for the peak ligament force values found in the
present study.
The peak ACL forces during isokinetic extension are
in rough agreement with the reported anterior shear
forces. Although the values reported by Nisell et al. [17]
and Baltzopolous [15] are higher than the peak ACL
forces in this study, their knee moments were also
higher, requiring higher muscle and anterior shear
forces. The peak PCL forces during isokinetic ¯exion
also agree well with the published posterior shear forces,
when the angle of inclination of the ligament is taken
into account.
184 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187

Table 2 5.2. Comparison of ACL and PCL forces

Tibio±femoral shear forces reported in previous studies of rehabilita-
tion exercises
The highest ACL forces occurring during these ex-
Author Exercise Peak Peak ercises were of approximately 0.55BW, or around
type anterior posterior 400 N. This is about 25% of the ultimate strength of a
shear shear young ACL [39]. In contrast, the highest PCL forces
(BW) (BW)
were around 4.6BW (3330 N), which is almost 85% of
Isokinetics the estimated ultimate strength of the young PCL [40]. It
Kaufman Extension 60°/s 0.3 0.5 must be remembered that, except during isokinetic ex-
et al. [16] tension, the PCL forces were upper bound estimates, as
180°/s 0.2 0.6 the DDOSC solutions chosen were not an exact match
Flexion 60°/s 1.7
180°/s 1.4
with the EMG data. Also, recent comparison of model
Nisell et al. Extension 30°/s 0.9 calculations with experimental measurements by Mar-
[17] kolf et al. [41] of antero-posterior sti€ness showed that
180°/s 0.7 while the anterior sti€ness (which depends on the ACL)
Baltzopolous Extension 30°/s 0.9 agreed reasonably well with experiment, the model
[15]
210°/s 0.8
posterior sti€ness (determined by the PCL) was signi®-
cantly larger than experiment (A. Imran, Personal
Isometrics Communication, 1997). For these reasons, the PCL
Smidt [13] Extension 0.4 0.4
Flexion 1.8
forces reported here are likely to be somewhat greater
than the actual values. Nevertheless, even when these
Squats factors are taken into account, it seems probable that
Dahlkvist Descent 3.66
et al. [18]
the PCL forces are still considerably higher than the
Ascent 2.75 ACL forces.
This di€erence is primarily a result of the di€erence in
angle of inclination of the patellar tendon and ham-
strings, though it is also a€ected by the angles of incli-
The main di€erence for isokinetic exercise is that the nation of the ACL and PCL and the moment arms of
posterior shear forces during extension reported by the muscles. The patellar tendon, whose anterior
Kaufman et al. [16] are much higher than in this study. shearing action is resisted by the ACL, is inclined at not
However, Nisell et al. also found low or negligible less than 65° to the tibial plateau, therefore the ratio of
posterior shear forces during extension in their study. anterior shear force to patellar tendon force is at most
The peak ACL and PCL forces during isometric exten- 0.4:1. The hamstrings, however, at about 90° ¯exion lie
sion and ¯exion, respectively, are also in relatively good parallel to the tibial plateau. Therefore the posterior
agreement with the published data. shear force which the PCL resists can be equal to the
Very few studies have reported tibio±femoral shear total hamstrings force. These geometrical considerations
forces during squats, and apart from Dahlkvist et al. explain the relative di€erences in peak ACL and peak
[18], have not separated the e€ect of muscle forces from PCL force.
the passive shear forces. DalhkvistÕs results are included
in Table 2. Their shear forces are larger than the peak 5.3. Limitations of the method
PCL forces in the present study. However, the combi-
nation of muscles they took to be active was di€erent to The forces calculated using this method agree rea-
those selected in this study, and this can have a con- sonably well with the limited published data that is
siderable e€ect on the ligament force magnitudes. De- available. In addition, various aspects of the model have
spite this, the values are of a similar order of magnitude previously been validated [30,37,42]. However, there are
and suggest that, although the PCL forces reported in a number of important limitations to the method.
this study represent an upper limit on the PCL forces, Firstly, the model is two-dimensional. This means that it
they are not excessively high. is only suitable to use for the analysis of exercises that
One of the quantities calculated in this study was the involve mainly sagittal plane movements [30]. Even in
tibial displacement, d, relative to the passive ¯exion these cases, there is likely to be some degree of out-of-
con®guration. Due to lack of space, it is not possible to plane load or movement, therefore some errors will be
present the calculated tibial displacements in this paper. introduced by the use of this model. However, the out-
However, elsewhere [37] they have been compared with of-plane components are relatively small: in squats the
experimental measurements of tibial displacement peak medio-lateral component of the measured ground
during exercise and shown to agree well with those reaction force was typically around 3% of the peak
results. vertical force. We were unable to measure the medio-
 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187
lateral component in the isokinetic experiments, but
data from the literature [43] indicate that the peak
medio-lateral component is around 5% of the total ap-
plied force.
A second potential source of error in the model is the
fact that, in this study, the same generalised set of model
parameters was used for all subjects. This was un-
avoidable, as it was not feasible to measure many of the
parameters on living subjects. A parameter sensitivity
study was carried out to assess how much this may have
a€ected the results [37]. It was found that, while most
model parameters have only a small e€ect on the peak
ligament forces, there are a few key parameters which
can have a signi®cant e€ect. These parameters include
those de®ning the geometry of the patello±femoral joint
and those a€ecting the moment arms of the hamstrings
about the knee. The results of the parameter sensitivity
study indicated that the peak ligament forces could be
up to 20% greater than the calculated values. In the
future, it may be possible to devise scaling methods so as
to allow the parameters to be adjusted to a particular
subject based on external measurements. However,
further work is needed before such a method is proven.
Another possible limitation of the method is the dif-
®culty of selecting a suitable solution combination from
all the admissible DDOSC solutions, in cases where
relatively large number of such solutions occur. This
tends to be the case with exercises where many muscles
are active simultaneously. Although the number of ad-
missible solutions is found to be small compared to the
number of possible combinations (at any instant, the
number of admissible solutions is no more than 6% of
the total number of potential solutions for both isoki-
netics/isometrics and squats), it can be di€cult to choose
between several alternative solutions when the EMG
data indicate that many muscles are active at that time.
This was more of a problem for the squats than for the
isokinetics, but was not an insuperable hurdle. It is
unlikely that other two-dimensional exercises would be
any harder to analyse than squats.
There are a number of other simpli®cations in the
model, such as the exclusion of collateral ligaments and
the assumption that the articular surfaces are non-de-
formable. However, all these simpli®cations lead to an
overestimation of the peak cruciate ligament forces. In a
clinical context, this means that there is e€ectively a
safety factor included in the results ± the actual peak
forces generated in a particular exercise should be no
greater than those forces reported here.
5.4. Clinical relevance
Eventually, it may be possible to use calculated liga-
ment force values such as these to determine relatively
precisely when it is safe to use a particular exercise for
rehabilitation. Before this is possible, however, it will be
185
necessary to learn more about the change in strength of
ligament grafts over the course of the healing process.
Meanwhile, it is possible to compare the ligament forces
occurring during di€erent exercises so as to gain an idea
of which exercises are preferable to use at which stage of
the rehabilitation process for di€erent injuries. Note that
the isokinetic and isometric force values calculated in
this study apply to maximal e€ort contractions. Sub-
maximal exercises will clearly lead to lower ligament
forces. The results presented here should therefore be
used as a guide to the maximum forces that could be
expected in the worst case.
The results of this study indicate that isokinetic ex-
tension is inadvisable at early stages of healing following
an ACL reconstruction. Although the peak ACL forces
are well below the ultimate strength of a healthy ACL,
or of most common graft materials, work by Butler et al.
[44], among others, has indicated that the graft strength
in animals is very low compared to the strength of a
normal ACL over the ®rst few weeks or months of
healing. However, isokinetic ¯exion is safe as no ACL
forces are produced at all. Increasing the exercise speed
does appear to reduce the peak ligament forces ± the
reduction is statistically signi®cant but relatively small in
clinical terms. Isometric exercise produces similar liga-
ment force patterns to isokinetics, and can be considered
as a special case of isokinetics (with an exercise speed of
0°/s). Squats do produce forces in the ACL, albeit rel-
atively small ones, and therefore should probably be
avoided for a few weeks post-reconstruction. However,
after that this appears to be an exercise that could more
safely be used for quadriceps strengthening than resisted
leg extension type exercises, provided there was no
reason to avoid the higher tibio±femoral and patello±
femoral contact forces that occur during squats.
In the case of PCL reconstruction or partial rupture,
the situation is more problematic. Isokinetic extension
would appear to be reasonably safe to use for
strengthening the quadriceps after a few weeks of heal-
ing, especially if high knee ¯exion angles are avoided.
However, both squats and isokinetic ¯exion ± exercises
which might strengthen the hamstrings ± lead to very
high PCL forces. This suggests that they should be
avoided until healing is well-advanced. Shallow knee
bends (to knee ¯exion angles of 50° or less) could be
introduced somewhat earlier.
6. Conclusions
During isokinetic or isometric extension exercises, the
ACL is loaded when the knee angle is less than 80° and
the PCL is loaded at higher knee angles. The peak ACL
forces, which occur at knee angles of around 35±40°, can
be greater than 0.5BW, but the PCL forces are much
lower. During isokinetic or isometric ¯exion, only the
186 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187
PCL is loaded. The peak forces may reach over 4BW,
occurring at knee angles of around 90°. In isokinetics,
there is a signi®cant decrease in peak ligament force with
increasing exercise speed.
During squats, relatively low ACL forces (less than
0.2BW) occur when the knee angle is less than around
50°. At larger knee angles, the PCL is loaded, with the
force increasing as the knee angle increases. The peak
PCL forces may be up to 3.5BW. The di€erences in
peak force occurring in di€erent types of squat were of
limited signi®cance.
The results suggest that, for rehabilitation of the
ACL, squats are a safer exercise than isokinetic exten-
sion to use for quadriceps strengthening, and that iso-
lated isokinetic ¯exion should be safe to use for
hamstrings strengthening. For rehabilitation of the
PCL, isokinetic extension may be used for quadriceps
strengthening if the range of motion is restricted to
¯exion angles of 80° or less. Shallow squats may be used
to keep the hamstrings exercised, with the knee ¯exion
angle initially kept below 50°. Deep squats and isoki-
netic ¯exion should be avoided until the graft is close to
fully healed.
Acknowledgements
D. Toutoungi was supported by the UK EPSRC and
by additional funding from Howmedica, Smith and
Nephew Healthcare and Corin Medical.
References
[1] Muller W. The knee: form, function, and reconstruction. Berlin:
Springer, 1983.
[2] Paulos L. Knee rehabilitation after ligament injury or surgery. In:
Norman Scott W, editor. The knee. St Louis: Mosby, 1994:943±
56.
[3] Reid D. Current concepts in rehabilitation of the anterior cruciate
de®cient knee. Current Orthop 1993;7:101±5.
[4] Sherman M, Warren R, Marshall J, Savatsky G. A clinical and
radiographical analysis of 127 anterior cruciate insucient knees.
Clin Orthop 1988;227:229±37.
[5] Shelbourne K, Nitz P. Accelerated rehabilitation after anterior
cruciate ligament reconstruction. Am J Sports Med
1990;18(3):292.
[6] Ahmed A, Hyder A, Burke D, Chan K. In vitro ligament tension
pattern in the ¯exed knee in passive loading. J Orthop Res
1987;5:217±30.
[7] Lew W, Lewis J. The e€ect of knee-prosthesis geometry on
cruciate ligament mechanics during ¯exion. J Bone Jt Surg Am
1982;64A:734±9.
[8] Paulos L, Noyes F, Grood E, Butler D. Knee rehabilitation after
anterior cruciate ligament reconstruction and repair. Am J Sports
Med 1981;9(3):140±9.
[9] Markolf K, Gorek J, Kabo M, Shapiro M. Direct measurement
of resultant forces in the anterior cruciate ligament. J Bone Jt
Surg Am 1990;72A(4):557±67.
[10] Markolf K, Wascher D, Finerman G. Direct in vitro measure-
ment of forces in the cruciate ligaments. Part II: the e€ect of
section of the posterolateral structures.. J Bone Joint Surg Am
1993;75-A(3):387±94.
[11] Holden J, Grood E, Korvick D, Cummings J, Butler D, Bylski-
Austrow D. In vivo forces in the anterior cruciate ligament: direct
measurements during walking and trotting in a quadruped. J
Biomech 1994;27(5):517±26.
[12] Beynnon B, Fleming B, Johnson R, Nichols C, Renstrom P, Pope
M. Anterior cruciate ligament strain behaviour during rehabili-
tation exercises in vivo. Am J Sports Med 1995;23(1):24±34.
[13] Smidt G. Biomechanical analysis of knee ¯exion and extension. J
Biomech 1973;6:79±92.
[14] Yasuda K, Sasaki T. The force exerted on the tibia by the
separate isometric contractions of the quadriceps and hamstrings.
Clin Orthop 1987;220.
[15] Baltzopolous V. Muscular and tibiofemoral joint forces during
isokinetic concentric knee extension. Clin Biomech
1995;10(4):208±14.
[16] Kaufman K, An K, Litchey W, Morrey B, Chao E. Dynamic
joint forces during knee isokinetic exercise. Am J Sports Med
1991;19(3).
[17] Nisell R, Ericson M, Nemeth G, Ekholm J. Tibio±femoral joint
forces during isometric knee extension. Am J Sports Med
1989;17(1):49±54.
[18] Dahlkvist N, Mayo P, Seedhom B. Forces during squatting and
rising from a deep squat. Engrg Med 1982;11(2):69±76.
[19] Nisell R, Nemeth G, Ohlsen H. Joint forces in extension of the
knee: analysis of a mechanical model. Acta Orthop Scand
1986;57:41±6.
[20] Zavatsky A, OÕConnor J. Ligament forces at the knee during
isometric quadriceps contractions. Proc Inst Mech Engs Part H, J
Engng Med 1993;207:7±18.
[21] Imran A, OÕConnor J. Control of knee stability after ACL injury
or repair: interaction between hamstrings contraction and tibial
translation. Clin Biomech 1998;13(3):153±62.
[22] Collins J, OÕConnor J. Muscle±ligament interactions at the knee
during walking. Proc Inst Mech Engs Part H, J Engng Med
1991;205:11±8.
[23] Collins JJ. The redundant nature of locomotor optimization laws.
J Biomech 1995;28(3):251±67.
[24] Wilson DR, Zavatsky AB, OÕConnor JJ. Cruciate ligament forces
at the knee in gait: parameter sensitivity and e€ects of ligament
elasticity. In: Proceedings of the International Society of Biome-
chanics 1993:1466±7.
[25] Zavatsky A, OÕConnor J. A model of human knee ligaments in
the sagittal plane: part I. response to passive ¯exion. Proc Inst
Mech Engs Part H, J Engng Med 1992;206:125±34.
[26] Zavatsky A, OÕConnor J. A model of human knee ligaments in
the sagittal plane part II. ®bre recruitment under load. Proc Inst
Mech Engs Part H, J Engng Med 1992;206:135±45.
[27] Lacote M, Chevalier A, Miranda A, Bleton J, Stevenin P. Clinical
Evaluation of Muscle Function. Churchill Livingstone, 1987.
[28] Winter D. Biomechanics and motor control of human movement.
2nd ed. New York: Wiley±Interscience, 1990.
[29] Woltring HJ. A fortran package for generalized, cross-validatory
spline smoothing and di€erentiation. Adv Engrg Software
1986;8:104±13.
[30] Lu T, OÕConnor J, Taylor S, Walker P. Validation of a lower limb
model with in vivo femoral forces telemetered from two subjects.
J Biomech 1998;31(2):129±34.
[31] OÕConnor J, Shercli€ T, FitzPatrick D, Biden E, Goodfellow J.
Mechanics of the knee. In: Daniel D, Akeson W, OÕConnor J,
editors. Knee ligaments: structure, function, injury, and repair.
New York: Raven Press, 1990:201±38.
[32] OÕConnor J, Shercli€ T, FitzPatrick D, Bradley J, Daniel D, Biden
E, Goodfellow J. Geometry of the knee. In: Daniel D, Akeson W,
 D.E. Toutoungi et al. / Clinical Biomechanics 15 (2000) 176±187
OÕConnor J, editors. Knee ligaments: structure, function, injury,
and repair. New york: Raven Press, 1990:163±200.
[33] OÕConnor J, Shercli€ T, Biden E, Goodfellow J. The geometry of
the knee in the sagittal plane. Proc Inst Mech Engs Part H, J
Engng Med 1989;203:223±33.
[34] Gill H, OÕConnor JJ. A bi-articulating two-dimensional computer
model of the human patellofemoral joint. Clin Biomech
1996;11(2):81±9.
[35] Toutoungi D, Giampaoli R, Leardini A, Catani F, OÕConnor J. A
method for the calculation of knee ligament loads during
rehabilitation following injury or reconstruction. In: Power H,
Hart R, editors. Computer Simulations in Biomedicine. Com-
puter Mechanics Publications 1995:393±400.
[36] Harris G, Wertsch J. Procedures for gait analysis. Arch Phys Med
Rehab 1994;75:216±25.
[37] Toutoungi D. The mechanics of rehabilitation exercises for the
cruciate-ligament injured knee. PhD. Thesis, University of
Oxford, 1997.
[38] Morrison J. The mechanics of the knee in relation to normal
walking. J Biomech 1970;3(1):51±61.
187
[39] Noyes F, Butler D, Grood E, Zernicke R, Hefzy M. Biomechan-
ical analysis of human ligament grafts used in knee±ligament
repairs and reconstructions. J Bone Joint Surg Am 1984;66A:344±
52.
[40] Race A, Amis A. The mechanical properties of the two bundles of
the human posterior cruciate ligament. J Biomech 1994;27(1):13±
24.
[41] Markolf K, Mensch J, Amstutz H. Sti€ness and laxity of the knee
± the contributions of the supporting structures. A quantitative in
vitro study. J Bone Joint Surg Am 1976;58A:585±93.
[42] Lu T, OÕConnor J. Lines of action and moment arms of the major
force-bearing structures crossing the human knee joint: compar-
ison between theory and experiment. J Anatomy 1996;189:575±
85.
[43] Kaufman K, An KN, Chao E. A comparison of intersegmental
joint dynamics to isokinetic dynamometer measurements. J
Biomech 1995;28(10):1243±56.
[44] Butler D, Kay M, Stou€er D. Comparison of material properties
in fascicle±bone units from human pateller tendon and knee
ligaments. J Biomech 1986;19:425±32.