- Email: [email protected]
Progressive decline in social attention in aging rats: An information-statistical method
Neurobiologyof Aging, Vol. 13, pp. 145-151. ©Pergamon Press plc, 1991. Printed in the U.S.A.
0197-4580/92 $5.00 + .00
Progressive Decline in Social Attention in Aging Rats: An Information-Statistical Method B E R R Y M. S P R U I J T
Division of Molecular Neurobiology, Institute of Molecular Biology and Medical Biotechnology University of Utrecht, Padualaan 8, 3584 CH Utrecht, The Netherlands R e c e i v e d 18 A u g u s t 1989; A c c e p t e d 23 M a y 1991 SPRUIJT, B. M. Progressive decline in social attention in aging rats: An information-statistical method. NEUROBIOL AGING 13(1) 145-151, 1992.--The purpose of this study was to assess the earliest signs of behavioral changes during aging. In a longitudinal setup the spontaneously displayed social behavior was observed and analyzed. Social interactions offer an appropriate paradigm for studying a gradual decrease in several categories of social behavior in aging animals. The frequencies and durations of 12 behavioral categories collected over 18 sessions yield a wealth of data, which is difficult to evaluate. Therefore, an information-statistical analysis was used, which compressed the frequencies and sequential organization of behavior into a very sensitive and meaningful index. This index expressing the predictability of behavior does not only show age-related changes but also yields an interpretation of the complex changes seen in frequency distributions. The applied analysis focuses on the predictability of behavior, which may be predominantly determined by the animal's own preceding behavior or by the behavior of its social partner. It is shown that the influence of the behavior of a partner on ongoing behavior declines age-dependently, whereas the influence of the animal's own preceding behavior increases. The simultaneous change in these two indices is interpreted as a progressive decline in social attention, which seems to be characteristic for aging rats. It is concluded that the analysis of the sequential organization of behavior offers a new tool, useful in longitudinal behavioral studies on aging. Aging
Sociability
Predictability of behavior
Rats
A variety of behavioral changes can be observed in aging rats, e.g., the performance in different learning tasks (13), sexual behavior (8) and grooming (17). Studies are predominantly focused on models for cognitive deficits as spatial orientation (14), visual recognition (12), classical conditioning (18), associative learning (2) and various avoidance paradigms (10). The preferential use of these cognitive models may be due to the fact that in humans changes in memory and learning are more prominent than changes in the structural organization of behavior. Nonetheless, a less efficient use of environmental cues, information seeking and a decrease in the organization of behavior in aged humans may underlie changes in cognitive functioning (1,7). Registration of multiple behavioral changes of two interacting individuals yields abundant complex information compared to the mere measurement of one index (e.g., latency, distance or frequency) of a few selected behaviors measured in classical learning paradigms. Measurement of a set of behavioral indices permits the analysis of complex behavioral characteristics, which are based on the quantification of relationships between different behavioral elements, i.e., the organization of behavioral elements. Spontaneously occurring behaviors such as social behavior may be especially useful for longitudinal studies, since learning paradigms, for methodological reasons, do not easily allow the measurement of performance at regular time points in a longitudinal setup. Although its nature--social exploratory versus more aggressive--may change as a consequence of previous experiences, characteristics such as the sequential organization of behavior and social interest can still be determined and used to assess the progressive nature of age-related features. An
information-statistical method was used to reveal gradually occurring changes in the organization of behavior of aging rats (3, 9, 11, 20, 21). The aim of this study was to investigate whether the information-statistical index applied to social behavior can be used as a marker for the aging process. This index implicates behavioral changes with respect to the degree of predictability of behavior, which is calculated per group for each time point in a longitudinal study and which allows statistical comparison between groups. METHOD
Housing and Social Interaction Test All animals were housed in a reversed day-night cycle: white light switched on at 8:00 p.m. and turned off at 8:00 a.m.; red light switched on at 8:00 a.m. and turned off at 8:00 p.m. The animals were kept in a temperature-controlled room, in which also the experiments were carried out. Thus the animals were never transported and kept under quiet and stable circumstances. Twenty-four hours before the social encounter the animals were housed in pairs in a Plexiglas observation cage (10 × 50 × 50 cm) with wooden shavings on the floor. However, the animals were separated by a removable partition which divided the cage in two compartments. To observe the social interaction the animals were taken from different cages, the partition was removed and subsequently the behavior of the animals was scored.
Behavioral Analysis From each animal 12 behavioral elements were observed
145
STRING OF BEHAVIORAL ELEMENTS OF • a-b-ef-d-a-of-e-b-a-f-a-c-b-c-d-
ANIMALS f-
2..
following animal
behaviors
I
i
c
O
j~
1
a
c
animal d
2 1
e
I 1
1
1 1
1
(..
~ 3.
FIG. 1. The behaviors scored are shown. The numbers correspond to: 1: approach; 2: social sniffing, licking and grooming; 3: anogenital sniffing; 4: running away, avoiding contact; 5: defensive behaviors such as lateral defensive, uptight defensive, sideways kicking with the hindpaws; 6: submissive behaviors, full submissive behaviors, freezing and crouch; 7: parry, the animal is sitting and carefully watching its partner, often in a comer of the cage; 8: autogrooming; 9: offensive behaviors such as lateral threat, uptight boxing, attack; 10: exploratory behaviors, sniffing, rearing; 11: all other occurring behaviors usually sitting, lying down, resting; 12: mounting. The outlines of the drawings are taken from (19).
(Fig. 1) 1: approach; 2: social sniffing, licking and grooming; 3: anogenital sniffing; 4: running away, avoiding contact; 5: defensive behaviors such as lateral defensive, upright defensive, sideways kicking with the hindpaws; 6: submissive behaviors, full submissive behaviors, freezing and crouch; 7: parry, the animal is sitting and carefully watching its partner, often in a corner of the cage; 8: autogrooming; 9: offensive behaviors such as lateral threat, upright boxing, attack; 10: exploratory behaviors, sniffing, rearing; 11: all other occurring behaviors usually sitting, lying down, resting; 12: mounting. For a detailed description of these behaviors see (5,19). Computer programs were used to register and analyze behavior (15). These programs--initially written for an Apple-II computer--now converted to an IBM PC, allow the registration and analysis of frequencies, durations and sequences of a stream of behavioral elements. Since an ongoing interaction was split up into 12 behavioral elements for each animal and over 300 behavioral elements per animal may
f
1
f
1
1
~.
2
1
i
i
I
1
1
FIG. 2. A transition matrix consisting of 4 matrices. Upper left panel contains transitions of behaviors displayed by animal 1. Lower tight panel shows transitions of behaviors displayed by animal 2. The upper tight panel contains the transitions of behaviors displayed by animal I followed by behavior of animal 2. The lower left panel shows the transitions of behaviors of animal 2 followed by behavior of animal I.
occur within a 30-min period, only experienced observers scored the behavior. For all three experiments, frequencies, durations and sequences of behavior were calculated. The procedure for the analysis of sequences will be shortly explained. The program denotes the sequences of behavior in so-called transition matrices [see Fig. 2; (15)], which formed the basis for further analysis. For an extensive explanation of these ethological methods one is referred to Van Hooff et al. (22). The transition matrices were summed over all animals per age class, yielding one transition matrix. The cells of such matrices contain the number of transitions between the behaviors identified by the row and column codes (see Fig. 2). In Fig. 2 an ethogram of each animal consists of 3 behavioral elements (a, b and c for animal 1 and d, e and f for animal 2). The total matrices were split up into smaller matrices representing, respectively, the transitions of behavioral elements displayed by animal 1 and animal 2 (transitions in behavior concerning the same animal are called "intra'" or "'self") and the transitions of behavior of animal 1 towards animal 2 (transitions of behavior of both animals are called "inter" or "other"). Transition matrices can be graphically represented by means of pathway diagrams. Combinations of behavior--transitions-which occur more often than can be expected according to a random distribution of transitions are depicted by arrows, which connect the preceding and following behavior. Statistical significance of transitions is based on p values (p<0.01) as is indicated by corresponding adjusted residuals (Z distributed). Differences between transitions are illustrated by relating the value of the adjusted residual to the thickness of the arrow. Thus in the present study adjusted residuals are used 1) for the selection of significant transitions according to a statistical criterium of p<0.01 and 2) for describing differences in probability of occurrence. For more details concerning the statistical and descriptive analysis of transitions one is referred to a previous study (15). When two behavioral elements are not connected this does not imply that those two behaviors were never combined, but that the occurrence of that combination did not significantly exceed the probability based on a random expectation. Furthermore, when a behavior is not connected with any other behavior
DECREASED SOCIAL ATTENTION IN AGING RATS
147
the occurrence of this particular behavior can be followed and or preceded by any other behavior; it does not mean that this behavior did not occur. Information statistics can be used to express the information in terms of predictability or uncertainty comprised by all transitions of one matrix in one index. For a more extensive description of the information statistics one is referred to the study of Van den Berken and Cools (21). The procedure is only shortly described: the inter- and intra-matrices are subjected to the information-statistical analysis. Information statistics on intra-matrices--either animal 1 or 2 in the example--calculates the predictability of each behavioral element based on the preceding element displayed by the animal itself. Information statistics on inter-matrices calculates the predictability of each element based on the preceding element of the other animal. The formulas for variability and covariability for rows (1) and columns (k) are: m
H(lt) = -=~1 p(li) * log2 P(li) = (I/N) * m
(N lo__g2N - E
n', log2 n',)
H (x)=,2.300 bits
1 .
.
.
.
.
.
.
.
.
.
H 1y)-2.100 bits FIG. 3. The amount of uncertainty expressed in H values at time t [ongoing behavior; H(y)] and at time t - 1 [preceding behavior; H(x)]. The total amount of uncertainty in preceding and following behavior is represented by H(x,y). The transmitted information from preceding to ongoing behavior is represented by T(x,y) and can easily be calculated from the H values, see the Method section.
(1)
I t represents the behavioral status of individual i on a certain moment t. 1i represents the behavioral element of individual i; m is the total number of behavioral elements, in this case 12; p (1i) = n.z/N indicates the probability that individual i performs element 1. This probability is calculated by dividing the frequency of behavioral (margin total) of individual i by the total frequency of all frequencies (grand total N). H(I,) expresses the amount of uncertainty (variability) in behavior at time t and is expressed in bits per behavioral category. Degrees of freedom = (1 - 1). m
H(It_l) = (l/N) * (Nlog2 N - E
nk'lOg2 nk')
(2)
k=l
n k is the margin total of column k; HOt 1) is the total individual variability (uncertainty) at time t - 1. m
H(Itlt_t) = (l/N) * (Nlog2 N - E
H (x, y)=3.100 bits
~
m
nk, lo__g2nk,)
(3)
nk~ is observed value of cell kl of the transition matrix, i.e., the total frequency of the combination of element k and 1. H(ItI t_ 1) represent the variability (uncertainty) of the combination of k and 1. For this index degrees of freedom = (k * 1 - 1); TE(Itlt-l) = H(It) + HOt l) - H(Itlt-~)
(4)
For TE values degrees of freedom = ( k - 1) * (l - 1). TE(ItI t_ 0 represents the autocovariability of combinations of preceding and following elements or, in other words, the reduction in uncertainty in predicting the next element, given the preceding one, expressed in bits. The higher the transmission efficiency (TE) value the more predictable the behavior is. TE values can be calculated over intra-transition matrices and over inter-transition matrices; the calculations are carded out similarly, Figure 3 illustrates this transmission of information coefficient graphically. H(x) represents preceding behavior, H(y) represents following behavior, H(x,y) the total amount of information. T(x,y), the transmitted information from preceding to following behavior is obtained by subtracting H(x) + H(y) H(x,y). As can be seen, the maximum information which can be transmitted from preceding to following behavior equals the smallest H value. This may lead to different maximum transmis-
sion coefficients for every matrix, which may interfere with the comparison of different TE values. By relating TE values to H(t 0, the amount of transmitted information is related the total amount of information of preceding behavior. Now it is possible to compare different TE values expressed as a percentage. No standard method is available to calculate means and variances of TE values, because one cannot calculate a TE value for every individual matrix, since individual matrices do not contain enough data (transitions). To avoid this problem, the method known as the Jack-knife technique (4) was applied to estimate variances. This method calculates the variance by subtracting every individual matrix from the total matrix and then calculating the according T values for this t o t a l - 1 number of matrices. This method also allows the calculation of 95% confidence intervals, which enables statistical comparison of TE values of different groups. Since TE values represent a rather abstract feature of a sequence of behavior, differences between young and aged animals in social interactions are also illustrated with more detailed pathway diagrams.
Experiment I The first experiment was carried out to assess changes in TE values in different age classes. Ten young (4 months) and 16 old (24 months) albino WAG/RIJ rats (related to Wistars) were obtained from TNO (Rijswijk, Netherlands) and housed in Makrolon cages with food and tap water ad lib. The social interaction test, in which 12 behaviors per animal were scored, lasted 20 min for these animals and was carried out as described above.
Experiment H For the longitudinal study 28 Brown Norway rats (3 months old) were purchased from TNO (Rijswijk, The Netherlands). These animals were housed in wire cages containing 8 animals each. After a month of adjustment to the housing conditions the animals have been observed 18 times at the following ages: 4, 5.5, 6.5, 8.5, 11.5, 12, 13, 14, 15, 16, 17,5, 18, 19, 20, 21, 22.5, 24, and 25 months. The social interaction test was similar to that of Experiment I. However, Brown Norway rats were observed for 30 rain, since these rats showed longer periods of social activity. The reason for using a second strain of rats was that the Brown Norway rats attain a higher age and have a lower incidence of tumors in the brain compared to Wistar and WAG/
-;P~l [ J i
148
]'ABLE 1 BEHAVIORAL CATEGORIES EXPRESSED IN FREQUENCY OF OCCURENCE DURING THE 30-MIN OBSERVATION PERIOD AT 5 DIFFERENT AGES OF BROWN NORWAY RATS
(a) 6.5 Months I. 2. 3. 4. 5. 6. 7. 8. 9. 10. I1. 12.
Approach Social sniffing Anogenital sniffing Avoid contact Defense Submission Parry Autogrooming Aggression Exploration Rest/sitting Mounting
36.92 121.42 17.91 36.72 24.46 2.65 22.29 19,71 15.23 62.37 16.36 28.99
-* 2.59 t''~''~ ± 5 . 9 2 b'~'d
_~. 1.96 b~J~ ± 2.89 a ± 2 . 1 8 b'c'd'e ± 0 . 4 8 b'c'd'c
-~ ± -+ -+
1.77 t''':'d 1.64 b 2,91 ~'~d" 2.91 bx'e
~
1 . 6 8 b'c'd
±
2 . 4 b'~'d'"
(b) 13 Months 56,75 100.25 5,80 41.01 13.54 0.33 1,42 34.62 3.75 121.99 5.63 11.08
(c) 17.5 Months
± 2,39" _+ 5.20 ~" ± 0.84 _+ 3.08 ± 1.36 -~ 0.10 -+ 0.36 ~ ± 1.74 "'d ± 0,87 ± 5.11 c ± (1,84~ _+ 1.38 a~
55.25 92,58 2.21 46.71 16.25 0.04 1.42 25,25 2,58 116.79 2.79 7.75
(d) 24 Months
± 3,87 e -+ 5.06 e ± 0.37 ~ ± 3.51 ± 1.93 ~ ± 0.01 _~ 0.38 ~ ± 1.54 +_ 0.50 ± 4.69 e ± 0.69 ~ ± 1.27
65.26 98.50 2,72 51.56 13.62 (L37 2,21 25.27 5.33 122,83 1,93 %43
~e) 3t Months
___ 4 . 3 r ± 5.53" = 0.44 ± 3.82" ± 1,67 ± 1).12 +- 0,67" + 2.06 -+ 1.04 _~ 4.88 ~" _~ 0.71' = 0.64
29,63 69,75 5.~'~3 34.13 9,63 0.63 12.63 26.63 2.75 58.63 17.88 475
_t: 4,51 ~ 9.48 _~:: 1.27 _~: 5,83 :: ~ 56 =: ().21 t: 2.04 * 298 ~ 0.75 .'.: 8,3I :~ 2 3N r !.!!
Mean ( - S E M I ; samples from observations at 6.5, 13, 17.5, 24 months (Experiment II) and 30 months (Experiment Ili) of age in Brown Norway rats. Frequencies were compared using the Mann-Whitney U-test; the level of significance (p<0.05) was corrected for the number of tests and adjusted to p<0.001. Significant differences are indicated by the letters b,c,d.e (superscript) referring to the age class.
RIJ rats (6). T h e data o f 4 animals, w h o s h o w e d infections at the e y e s at an age o f 11 m o n t h s were not included in o u r analysis. T h e n u m b e r o f a n i m a l s used for the study w a s 24 until an age o f 17.5 m o n t h s . After this age the n u m b e r o f a n i m a l s from w h i c h social b e h a v i o r could be recorded gradually declined. A n imals suffering f r o m infections (eye or m i d ear) were e x c l u d e d f r o m further e x p e r i m e n t a t i o n , w h i c h left us with ten a n i m a l s at an age o f 25 m o n t h s . F r e q u e n c i e s o f behavior at the age o f 6.5, 13, 17.5 and 24 m o n t h s h a v e been collected to identify c h a n g e s in types o f behavior in a g i n g rats. For a m o r e detailed description o f the behavior the transition matrix o f the 2 3 - m o n t h - o l d rats (N = 12) was c o m p a r e d with the transition matrix o f the 6.5 m o n t h old B r o w n N o r w a y rats (N = 24),
Experiment 111 T h e repeated observation o f the s a m e a n i m a l s m a y result in habituation to the social interaction test, w h i c h m a y partially account for a decline in social interest. Therefore naive aged B r o w n N o r w a y rats were subjected to a similar experimental procedure to a s s e s s if naive aged a n i m a l s differ f r o m regularly o b s e r v e d s e n e s c e n t a n i m a l s . T w e n t y - s e v e n B r o w n N o r w a y rats were p u r c h a s e d f r o m T N O Rijswijk, T h e Netherlands, at an age o f 24 m o n t h s . A f t e r o n e m o n t h o f a d j u s t m e n t to laboratory conditions they were subjected to the social interaction test 7 t i m e s at the ages o f 25, 26, 28, 29, 31, 32 and 33.5 m o n t h s . In addition to the frequencies o f behavior at the ages m e n t i o n e d above the f r e q u e n c i e s o f the 12 b e h a v i o r s e x p r e s s e d at an age o f 31 m o n t h s h a v e been a d d e d to Table 1.
ior displayed by their partner, as c o m p a r e d to the first 10 ± i n . Y o u n g a n i m a l s m a i n t a i n e d their level o f being influenced by the behavior o f their partner during the total 2 0 - m i n observation session. In E x p e r i m e n t II T E values altered a g e - d e p e n d e n t l y as can be seen in Fig. 5 A (left panel), in w h i c h both T E - o t h e r and TEself values are g i v e n from 4 until 25 m o n t h s o f age per 10-min time period. Until the age o f 12 m o n t h s T E - o t h e r exceed TEself values. After this age T E - s e l f is larger than TE-other and this difference increases age-dependently. T E - s e l f e x c e e d i n g T E - o t h e r was first and m o s t clearly seen in the last 10 m i n of the observation session. Variance as calculated by the Jack-knife m e t h o d w a s always b e t w e e n 0.25 and 1.30 with an average o f 1 - 2 % o f the T E value; thus, the variance per trial b e t w e e n the a n i m a l s is rather small, w h i c h underlines the within trial stability o f the proposed index. For statistical c o m p a r i s o n o f different points o f the curve the intervals o f confidence can be set to 2%.
sociability m aging rats
100-
[--"]intra
~inter
:¥
50
i
Fq
I r'-I
r---1
RESULTS I
Figure 4 s h o w s the T E coefficients for 4 - m o n t h and 24m o n t h - o l d W A G / R I J rats. D u r i n g the first 10 m i n o f the social encounter, T E - o t h e r values for y o u n g a n d aged rats were higher than T E - s e l f values. T h e b e h a v i o r o f both y o u n g and aged animals w a s m o r e d e t e r m i n e d by the preceding behavior o f their partner than by their o w n preceding behavior. D u r i n g the second 10 m i n , h o w e v e r , the T E - o t h e r values for aged a n i m a l s decreased. T h e s e n e s c e n t rat s e e m e d less susceptible to the behav-
1
21 old
I 1
21 young
FIG. 4. Comparison of the predictability of behavior of young (4 months) and old (23 months) rats. Predictability is based on preceding behavior of the animal itself (expressed in TE-self) and the behavior of its partner (represented by TE-other). The observation period was split into 2 blocks of 10 min: 1 and 2. An asterisk indicates levels of significance of p < 0 . 0 5 , as assessed by the Jack-knife method for calculating confidence intervals.
DECREASED SOCIAL ATTENTION IN AGING RATS
30-
sociabilit
A
in aging rats
149
B
10090-
28-
80-
26'
.~ 7o._>
24-
60-
22-
~.
2018-
:30-
30-
20
28-
b
4
'
~
'
1'2 '
116 '
210 '
2n4mnth
2624.
FIG. 6. The total number of the 12 behavioral categories displayed after a behavioral change of the partner have been summed in totals representing the total amount of behavioral reactivity. Squares, filled circles and open circles represent first, second and third ten-min block, respectively.
22. 20. 18. 30. 2826" 24. 222018. 16" 14. ~
n
t
h
25
27
29
31
33mnth
FIG. 5. Illustration of the age-dependent increase in the predictability of behavior based on the animal's own preceding behavior (TE-self, filled circles) and the age-dependent decrease in predictability of behavior based on the behavior of another rat (TE-other, open circles). These indices have been calculated for the first, second and third 10 min of every observation period for a group of rats observed from 4 to 25 months of age (panel A, left) and for a group of rats observed from 25 to 33 months of age (panel B, right). The division on the right side of panel B also concerns panel A.
The fluctuation of TE-other values between trials is not surprising, since circumstances and experience of the animals may change over an interval between successive observations. The total number of behavioral actions occurring after a behavioral change of the partner has been assessed and is depicted in Fig. 6. The values representing behavioral reactivity have been derived from parts of total matrices corresponding with behavior directed towards the other animal. Since both TE values and these totals are derived from the same matrices the variability must be in the same order of magnitude. It is evident that no linear age-dependent difference in the number of behavioral changes displayed after a behavioral change of the partner could be observed in any of the 10-min observation blocks (Fig. 6). The comparison of Fig. 6 and Fig. 5 yields two arguments for a dissociation between behavioral reactivity and predictability (TE values). First, an age-dependent decrease in behavioral reactivity is seen most prominently in the third 10-rain block, whereas the age-related changes in TE indices are present in all three blocks of 10 min. Second, the profile of the age-related changes in TE values does not correspond with the profile of the values representing behavioral reactivity. Thus, the decreased TE-other values, as shown in Fig. 5A and B, appear related to a gradual diminishment in specific action-reaction combinations between two animals rather than related to a decrease in behavioral reactivity. Pathway diagrams of young and aged rats have been composed and are compared in Fig. 7 to illustrate which specific combinations attribute to the decreased predictability of social behavior in aging rats. The main differences between the path-
way diagrams representing combinations of preceding and following behaviors of young animals (left panel) and the pathway diagram of aged animals (right panel) are indicated by black arrows (Fig. 7). In each pathway diagram preceding behaviors are drawn at the left side and following behaviors at the right side. Young animals show more specific combinations of behaviors, i.e., more arrows, than old animals. Social sniffing [2], aggressive behavior [9], rest/sitting [11] and mounting [12] are followed by social sniffing [2], defensive [5] and aggressive behavior [9], autogrooming [8] or rest/sitting [11], respectively. Exploration [10] as displayed by young animals is not considered as a social behavior and seems more performed at the same time rather than induced by exploratory behavior of the partner; therefore this " a r r o w " is not made black. For similar reasons the mutual dependency of rest/sitting in old rats is not highlighted by a black arrow. Figure 5B (right panel) shows the TE-self and TE-other values for rats from 25 to 33.5 months of age. The TE-other values decline and the TE-self values increase rapidly in all three blocks of 10 min. Only during the first observation TE-other exceeds TE-self in all 10-min blocks. In Experiment II this feature changes after 6 trials in the first 10 min of the observation session; in the senescent, unexperienced rat the change in the ratio TE-other versus TE-self occurs already after one trial (compare Fig. 5A and B, first 10-min panels). In addition, this change was seen in Experiment II after an inter-trial interval of 3 months (8.5 to 11.5 months). Since behavior is usually expressed in frequencies and our analysis is based on the sequential ordering of frequencies, the " r a w " frequencies of the twelve behavioral categories scored during 30 min at 5 ages are given in Table 1. Thirteen-monthold animals show significantly more approach, self-grooming, exploration and less social sniffing, anogenital sniffing, defense, parry, aggression, sitting and mounting than 6.5-month-old animals. From 6.5 to 13 months there seems to be a change from agonistic behavior towards more social sniffing and exploratory behaviors. From an age of 13 to 24 months the frequencies appear rather similar. The senescent rats of 31 months display a significant decline in the frequencies of approach, social sniffing, avoid, defense, exploration and an increase in parry, rest/ sitting as compared to the age of 24 months. DISCUSSION
From behavioral elements of an ongoing stream of behavior the probability of occurrence of each behavioral element, given
t5n
:'~PR[qt ~ pathway diagram of social behavior
/~'~
young
@ ///
@
®/,/
® ®
FIG. 7. Transitions in behavior between the 12 behaviors of young (6.5 months, left panel) and old animals (25 months, right panel) are depicted as arrows connecting the two behaviors. An arrow must be interpreted as "is followed by." The thickness of the arrow refers to the adjusted residual value of the particular cell of the transition matrix. Adjusted residuals indicate the probability of occurrence assessed by comparing the observed number of transitions with a random expectation of transitions. Only those transitions are depicted as arrows when reaching a level of significance of p<0.01. For more explanation see text and Spruijt and Gispen (15). Black arrows represent behavioral combinations only displayed by young animals, whereas white arrows represent behavioral combinations occurring in both, young and aged rats.
the preceding behavioral events, can be calculated. The TE index representing the predictability of an event is either based on preceding behavior of the animal itself (TE-self) or based on the preceding behavior of another animal (TE-other). This index gives insight in how far ongoing behavior is determined by the animal's own preceding behavior (TE-self) or by the behavior of another animal (TE-other). Thus, a high TE-other value parallels a high level of communication and a low TE-other value indicates little specific responsiveness to the behavior of a social partner. A high TE-self value represents a rigid behavioral pattern, which is governed by the own preceding behavior of the animal. It appears that young animals--both WAG/RIJ rats and Brown Norway rats--are more influenced by the behavior of their partners than by their own preceding behavior, as TE-other values exceed TE-self values. However, in the senescent rat this relationship is reversed: now the own preceding behavior determines ongoing behavior more than preceding behavior of another rat. This dramatic decline of specificity of following behaviors after a certain preceding behavior of the other rat progresses gradually in aging rats, as is demonstrated by the results of this longitudinal study. The decline in TE-other values is parallelled by increased TE-self values. This reversed TE-self/TE-other ratio cannot be explained by a general decrease in the total number of acts displayed after a behavioral change of the partner, i.e., the degree of reactivity. Changes in the frequencies of the 12 behaviors perhaps suggest a change in the nature of the interaction occurring between 6.5 and 13 months and after 24 months, but do not reveal changes in the period from 13 to 24 months. The nature of the interaction reflected in frequencies appears rather constant in the time period from 13 to 24 months, whereas in this period the
FE-other values gradually decline, The frequencies ot me ~ar~ ous behaviors are difficult to interpret: some behaviors iapproach, avoid contact, self-grooming and explorationi increase in frequency, whereas others (social sniffing, defense, aggces sion, mount) decrease in 24-month-old animals as compared to 6.5-month-old animals. One might speculate that contact behav iors, with the exception of approach, generally decrease, wherea,~ nonsocial behaviors increase in frequency. These changes can be reconciliated with more overt changes in the pathway diagram and TE values. However, the mere change in frequency ,,3 a certain behavior is dissociated from a particular preceding or following behavior of another animal. Approach is not associated at all with any particular behavior of the partner. In the senescent rat social sniffing is only preceded by defense, and exploration, self-grooming and aggression are not preceded by a particular behavior despite the increase in frequency for approach, self-grooming and exploration. Likewise, changes m the, TE values are not parallel to changes in total frequencies of behavior. The alteration in TE values occurs more gradual and in a more relevant period for aging (13.5 to 24 months~. Another explanation for a gradual decline in performance in a longitudinal setup may be the repeated exhibition to a similar test situation (habituation). Therefore, the social behaviors of senescent rats, which had never been exposed to such social interaction test before, were studied; initially, they showed larger TE-other values, but rapidly they reached the level of the aged animals of Experiment II. Apparently, the novelty o1 a first exposure to a social interaction test induces in aged rats a q~;-other/TE-self ratio similar to the ratio seen in younger rats. However. already during the first exposure the difference between TE-self and TE-other is larger in younger rats than the difference seen in the aged rat. After the first test the decline in TE-other values and the increase in TE-self values occurs quicker than in younger animals. Thus the degree of social responsiveness in aged rats seems to be partly influenced b~ previou~ experience, but i~ mainly age-dependent. Other studies suggested that alterations m the structure ol behavior underlie age-dependent changes in behavior as has been demonstrated for sexual behavior (16) and grooming behavior (17). In the latter studies changes were illustrated with complex diagrams representing the combination of behaviors studied. In the present study such complex behavioral alterations are reduced to one index. This method offers the possibility to define characteristics such as changes in flexibility or rigidity which are also known to occur in aging humans (1). Changes in the organization of behavior and the seeking of information may underlie age-related cognitive deficits (2). Changes in social behavior assessed with an information-statistical method have also been demonstrated in other animals such as the Java monkeys (Macaca fascicutaris). Manipulation of the nucleus caudatus induced changes in communication patterns, which were expressed in changes in TE values in socially housed monkeys (21). This method has also been applied in our laboratory to assess the behavioral reflection of the onset of aging and to investigate the effects of ACTH-Iike neuropeptides on such age-related behavioral changes. The advantage of the present index compared to the registration of mere frequencies and durations is that in one TE value all the information concerning the specificity of combinations of the behavior of two interacting animals is compressed. Furthermore, this index is more appropriate for the interpretation of complex behavioral changes as is illustrated by the comparison with a change in the distribution of various frequencies. In the present social interaction test I suggest that this index reflects very adequately the level of social attention (TE-other) in aging rats.
D E C R E A S E D S O C I A L A T T E N T I O N IN A G I N G RATS
151
ACKNOWLEDGEMENTS The authors wish to express their gratitude to Marlou Josephy for her advanced biotechnical assistance. This study was supported by a Constantijn and Christiaan Huygens Career Development Award to Berry Spruijt received from the Netherlands Organization for the Advancement of Pure Research (N.W.O.).
REFERENCES 1. Botwinnick, J. Neuropsychology of aging. In: Filskov, S. B.; Boll, T. J., eds. Handbook of clinical neuropsychology. New York: Wiley; 1981:135-171. 2. Buchanan, S. L.; Powell, D. A. Age-related changes in associative learning: studies in rabbits and rats. Neurobiol. Aging 9:523-534; 1988. 3. Chatfield, Ch.; Lemon, R. E. Analyzing sequences of behavioural events. J. Theor. Biol. 29:427-445; 1970. 4. Fagen, R. M. Information measures: statistical confidence limits and inferences. J. Theor. Biol. 73:61-79; 1978. 5. Grant, E. C. An analysis of the social behavior of the male laboratory rat. Behaviour 2 1:260-281 ; 1963. 6. Hollander, C. F. Current experience using the laboratory rat in aging studies. Lab. Animal Sci. 26:320-328; 1976. 7. Jolles, J. Neuropeptides and the treatment of cognitive deficits in aging and dementia. Prog. Brain Res. 70:429-441; 1986. 8. Larsson, K. Sexual activity in senile male rats. J. Gerontot. 13:136-139; 1958. 9. Losey, G. S. Information theory and communication. In: Colgan, P. W., ed. Quantitative ethology. New York: John Wiley & Sons; 1979:43-78. 10. Martinez, J. L.; Schulteis, G.; Janak, P. H.; Weinberger, S. B. Behavioral assessment of forgetting in aged rodents and its relationship to peripheral sympathetic function. Neurobiol. Aging 9:697708; 1988. 11. Miller, G. A.: Madow, W. G. On the maximum likelihood estimate of the Shannon-Wiener measure of information. Air Force Cambridge, Research Center Technical Report AFCRC-TR-54-75, 1954: 1-22. 12. Moss, M. B.; Rosene, D. L.; Peters, A. Effects of aging on visual recognition memory in the rhesus monkey. Neurobiol. Aging 9:495-
506; 1988. 13. Neurobiology of Aging. Special issue on experimental models of age-related memory dysfunction and neurodegeneration. 9:445-765; 1988. 14. Olton, D. S. Aging and spatial cognition. Neurobiol. Aging 9:469474; 1988. 15. Spruijt, B. M.; Gispen, W. H. Behavioural sequences as an easily quantifiable parameter in experimental studies. Physiol. Behav. 32: 707-710; 1984. 16. Spruijt, B. M.; Meyerson, B. Sequential analysis of the copulatory behavior of young and aged rats. Neurosci. Res. Commun. 1:3138; 1987. 17. Spruijt, B. M.; Welbergen, P.; Brakkee, J.; Gispen, W. H. An ethological analysis of excessive grooming in young and aged rats. Ann. NY Acad. Sci. 525:89-100; 1988. 18. Thompson, R. F. The rosetta stone for brain substrates of age-related deficits in learning and memory? Neurobiol. Aging 9:547-548; 1988. 19. Timmermans, P. J. A. Social behavior in rats, Ph.D. thesis. Nijmegen, The Netherlands, 1978. 20. Van den Bercken, J. H. L.; Cools, A. R. Information-statistical analysis of social interaction and communication: an analysis of variance approach. Anim. Behav. 28:172-188; 1980. 21. Van den Bercken, J. H. L.; Cools, A. R. Information-statistical analysis of factors determining ongoing behaviour and social interaction in Java monkeys (Macaca fascicularis). Anim. Behav. 28: 189-200; 1980. 22. Van Hooff, J. A. R. A. M. Categories and sequences of behaviour: methods of description and analysis. In: Ekman, P.; Scherer, K., eds. Handbook of non-verbal communication research. New York: Cambridge University; 1982:362--439.
0197-4580/92 $5.00 + .00
Progressive Decline in Social Attention in Aging Rats: An Information-Statistical Method B E R R Y M. S P R U I J T
Division of Molecular Neurobiology, Institute of Molecular Biology and Medical Biotechnology University of Utrecht, Padualaan 8, 3584 CH Utrecht, The Netherlands R e c e i v e d 18 A u g u s t 1989; A c c e p t e d 23 M a y 1991 SPRUIJT, B. M. Progressive decline in social attention in aging rats: An information-statistical method. NEUROBIOL AGING 13(1) 145-151, 1992.--The purpose of this study was to assess the earliest signs of behavioral changes during aging. In a longitudinal setup the spontaneously displayed social behavior was observed and analyzed. Social interactions offer an appropriate paradigm for studying a gradual decrease in several categories of social behavior in aging animals. The frequencies and durations of 12 behavioral categories collected over 18 sessions yield a wealth of data, which is difficult to evaluate. Therefore, an information-statistical analysis was used, which compressed the frequencies and sequential organization of behavior into a very sensitive and meaningful index. This index expressing the predictability of behavior does not only show age-related changes but also yields an interpretation of the complex changes seen in frequency distributions. The applied analysis focuses on the predictability of behavior, which may be predominantly determined by the animal's own preceding behavior or by the behavior of its social partner. It is shown that the influence of the behavior of a partner on ongoing behavior declines age-dependently, whereas the influence of the animal's own preceding behavior increases. The simultaneous change in these two indices is interpreted as a progressive decline in social attention, which seems to be characteristic for aging rats. It is concluded that the analysis of the sequential organization of behavior offers a new tool, useful in longitudinal behavioral studies on aging. Aging
Sociability
Predictability of behavior
Rats
A variety of behavioral changes can be observed in aging rats, e.g., the performance in different learning tasks (13), sexual behavior (8) and grooming (17). Studies are predominantly focused on models for cognitive deficits as spatial orientation (14), visual recognition (12), classical conditioning (18), associative learning (2) and various avoidance paradigms (10). The preferential use of these cognitive models may be due to the fact that in humans changes in memory and learning are more prominent than changes in the structural organization of behavior. Nonetheless, a less efficient use of environmental cues, information seeking and a decrease in the organization of behavior in aged humans may underlie changes in cognitive functioning (1,7). Registration of multiple behavioral changes of two interacting individuals yields abundant complex information compared to the mere measurement of one index (e.g., latency, distance or frequency) of a few selected behaviors measured in classical learning paradigms. Measurement of a set of behavioral indices permits the analysis of complex behavioral characteristics, which are based on the quantification of relationships between different behavioral elements, i.e., the organization of behavioral elements. Spontaneously occurring behaviors such as social behavior may be especially useful for longitudinal studies, since learning paradigms, for methodological reasons, do not easily allow the measurement of performance at regular time points in a longitudinal setup. Although its nature--social exploratory versus more aggressive--may change as a consequence of previous experiences, characteristics such as the sequential organization of behavior and social interest can still be determined and used to assess the progressive nature of age-related features. An
information-statistical method was used to reveal gradually occurring changes in the organization of behavior of aging rats (3, 9, 11, 20, 21). The aim of this study was to investigate whether the information-statistical index applied to social behavior can be used as a marker for the aging process. This index implicates behavioral changes with respect to the degree of predictability of behavior, which is calculated per group for each time point in a longitudinal study and which allows statistical comparison between groups. METHOD
Housing and Social Interaction Test All animals were housed in a reversed day-night cycle: white light switched on at 8:00 p.m. and turned off at 8:00 a.m.; red light switched on at 8:00 a.m. and turned off at 8:00 p.m. The animals were kept in a temperature-controlled room, in which also the experiments were carried out. Thus the animals were never transported and kept under quiet and stable circumstances. Twenty-four hours before the social encounter the animals were housed in pairs in a Plexiglas observation cage (10 × 50 × 50 cm) with wooden shavings on the floor. However, the animals were separated by a removable partition which divided the cage in two compartments. To observe the social interaction the animals were taken from different cages, the partition was removed and subsequently the behavior of the animals was scored.
Behavioral Analysis From each animal 12 behavioral elements were observed
145
STRING OF BEHAVIORAL ELEMENTS OF • a-b-ef-d-a-of-e-b-a-f-a-c-b-c-d-
ANIMALS f-
2..
following animal
behaviors
I
i
c
O
j~
1
a
c
animal d
2 1
e
I 1
1
1 1
1
(..
~ 3.
FIG. 1. The behaviors scored are shown. The numbers correspond to: 1: approach; 2: social sniffing, licking and grooming; 3: anogenital sniffing; 4: running away, avoiding contact; 5: defensive behaviors such as lateral defensive, uptight defensive, sideways kicking with the hindpaws; 6: submissive behaviors, full submissive behaviors, freezing and crouch; 7: parry, the animal is sitting and carefully watching its partner, often in a comer of the cage; 8: autogrooming; 9: offensive behaviors such as lateral threat, uptight boxing, attack; 10: exploratory behaviors, sniffing, rearing; 11: all other occurring behaviors usually sitting, lying down, resting; 12: mounting. The outlines of the drawings are taken from (19).
(Fig. 1) 1: approach; 2: social sniffing, licking and grooming; 3: anogenital sniffing; 4: running away, avoiding contact; 5: defensive behaviors such as lateral defensive, upright defensive, sideways kicking with the hindpaws; 6: submissive behaviors, full submissive behaviors, freezing and crouch; 7: parry, the animal is sitting and carefully watching its partner, often in a corner of the cage; 8: autogrooming; 9: offensive behaviors such as lateral threat, upright boxing, attack; 10: exploratory behaviors, sniffing, rearing; 11: all other occurring behaviors usually sitting, lying down, resting; 12: mounting. For a detailed description of these behaviors see (5,19). Computer programs were used to register and analyze behavior (15). These programs--initially written for an Apple-II computer--now converted to an IBM PC, allow the registration and analysis of frequencies, durations and sequences of a stream of behavioral elements. Since an ongoing interaction was split up into 12 behavioral elements for each animal and over 300 behavioral elements per animal may
f
1
f
1
1
~.
2
1
i
i
I
1
1
FIG. 2. A transition matrix consisting of 4 matrices. Upper left panel contains transitions of behaviors displayed by animal 1. Lower tight panel shows transitions of behaviors displayed by animal 2. The upper tight panel contains the transitions of behaviors displayed by animal I followed by behavior of animal 2. The lower left panel shows the transitions of behaviors of animal 2 followed by behavior of animal I.
occur within a 30-min period, only experienced observers scored the behavior. For all three experiments, frequencies, durations and sequences of behavior were calculated. The procedure for the analysis of sequences will be shortly explained. The program denotes the sequences of behavior in so-called transition matrices [see Fig. 2; (15)], which formed the basis for further analysis. For an extensive explanation of these ethological methods one is referred to Van Hooff et al. (22). The transition matrices were summed over all animals per age class, yielding one transition matrix. The cells of such matrices contain the number of transitions between the behaviors identified by the row and column codes (see Fig. 2). In Fig. 2 an ethogram of each animal consists of 3 behavioral elements (a, b and c for animal 1 and d, e and f for animal 2). The total matrices were split up into smaller matrices representing, respectively, the transitions of behavioral elements displayed by animal 1 and animal 2 (transitions in behavior concerning the same animal are called "intra'" or "'self") and the transitions of behavior of animal 1 towards animal 2 (transitions of behavior of both animals are called "inter" or "other"). Transition matrices can be graphically represented by means of pathway diagrams. Combinations of behavior--transitions-which occur more often than can be expected according to a random distribution of transitions are depicted by arrows, which connect the preceding and following behavior. Statistical significance of transitions is based on p values (p<0.01) as is indicated by corresponding adjusted residuals (Z distributed). Differences between transitions are illustrated by relating the value of the adjusted residual to the thickness of the arrow. Thus in the present study adjusted residuals are used 1) for the selection of significant transitions according to a statistical criterium of p<0.01 and 2) for describing differences in probability of occurrence. For more details concerning the statistical and descriptive analysis of transitions one is referred to a previous study (15). When two behavioral elements are not connected this does not imply that those two behaviors were never combined, but that the occurrence of that combination did not significantly exceed the probability based on a random expectation. Furthermore, when a behavior is not connected with any other behavior
DECREASED SOCIAL ATTENTION IN AGING RATS
147
the occurrence of this particular behavior can be followed and or preceded by any other behavior; it does not mean that this behavior did not occur. Information statistics can be used to express the information in terms of predictability or uncertainty comprised by all transitions of one matrix in one index. For a more extensive description of the information statistics one is referred to the study of Van den Berken and Cools (21). The procedure is only shortly described: the inter- and intra-matrices are subjected to the information-statistical analysis. Information statistics on intra-matrices--either animal 1 or 2 in the example--calculates the predictability of each behavioral element based on the preceding element displayed by the animal itself. Information statistics on inter-matrices calculates the predictability of each element based on the preceding element of the other animal. The formulas for variability and covariability for rows (1) and columns (k) are: m
H(lt) = -=~1 p(li) * log2 P(li) = (I/N) * m
(N lo__g2N - E
n', log2 n',)
H (x)=,2.300 bits
1 .
.
.
.
.
.
.
.
.
.
H 1y)-2.100 bits FIG. 3. The amount of uncertainty expressed in H values at time t [ongoing behavior; H(y)] and at time t - 1 [preceding behavior; H(x)]. The total amount of uncertainty in preceding and following behavior is represented by H(x,y). The transmitted information from preceding to ongoing behavior is represented by T(x,y) and can easily be calculated from the H values, see the Method section.
(1)
I t represents the behavioral status of individual i on a certain moment t. 1i represents the behavioral element of individual i; m is the total number of behavioral elements, in this case 12; p (1i) = n.z/N indicates the probability that individual i performs element 1. This probability is calculated by dividing the frequency of behavioral (margin total) of individual i by the total frequency of all frequencies (grand total N). H(I,) expresses the amount of uncertainty (variability) in behavior at time t and is expressed in bits per behavioral category. Degrees of freedom = (1 - 1). m
H(It_l) = (l/N) * (Nlog2 N - E
nk'lOg2 nk')
(2)
k=l
n k is the margin total of column k; HOt 1) is the total individual variability (uncertainty) at time t - 1. m
H(Itlt_t) = (l/N) * (Nlog2 N - E
H (x, y)=3.100 bits
~
m
nk, lo__g2nk,)
(3)
nk~ is observed value of cell kl of the transition matrix, i.e., the total frequency of the combination of element k and 1. H(ItI t_ 1) represent the variability (uncertainty) of the combination of k and 1. For this index degrees of freedom = (k * 1 - 1); TE(Itlt-l) = H(It) + HOt l) - H(Itlt-~)
(4)
For TE values degrees of freedom = ( k - 1) * (l - 1). TE(ItI t_ 0 represents the autocovariability of combinations of preceding and following elements or, in other words, the reduction in uncertainty in predicting the next element, given the preceding one, expressed in bits. The higher the transmission efficiency (TE) value the more predictable the behavior is. TE values can be calculated over intra-transition matrices and over inter-transition matrices; the calculations are carded out similarly, Figure 3 illustrates this transmission of information coefficient graphically. H(x) represents preceding behavior, H(y) represents following behavior, H(x,y) the total amount of information. T(x,y), the transmitted information from preceding to following behavior is obtained by subtracting H(x) + H(y) H(x,y). As can be seen, the maximum information which can be transmitted from preceding to following behavior equals the smallest H value. This may lead to different maximum transmis-
sion coefficients for every matrix, which may interfere with the comparison of different TE values. By relating TE values to H(t 0, the amount of transmitted information is related the total amount of information of preceding behavior. Now it is possible to compare different TE values expressed as a percentage. No standard method is available to calculate means and variances of TE values, because one cannot calculate a TE value for every individual matrix, since individual matrices do not contain enough data (transitions). To avoid this problem, the method known as the Jack-knife technique (4) was applied to estimate variances. This method calculates the variance by subtracting every individual matrix from the total matrix and then calculating the according T values for this t o t a l - 1 number of matrices. This method also allows the calculation of 95% confidence intervals, which enables statistical comparison of TE values of different groups. Since TE values represent a rather abstract feature of a sequence of behavior, differences between young and aged animals in social interactions are also illustrated with more detailed pathway diagrams.
Experiment I The first experiment was carried out to assess changes in TE values in different age classes. Ten young (4 months) and 16 old (24 months) albino WAG/RIJ rats (related to Wistars) were obtained from TNO (Rijswijk, Netherlands) and housed in Makrolon cages with food and tap water ad lib. The social interaction test, in which 12 behaviors per animal were scored, lasted 20 min for these animals and was carried out as described above.
Experiment H For the longitudinal study 28 Brown Norway rats (3 months old) were purchased from TNO (Rijswijk, The Netherlands). These animals were housed in wire cages containing 8 animals each. After a month of adjustment to the housing conditions the animals have been observed 18 times at the following ages: 4, 5.5, 6.5, 8.5, 11.5, 12, 13, 14, 15, 16, 17,5, 18, 19, 20, 21, 22.5, 24, and 25 months. The social interaction test was similar to that of Experiment I. However, Brown Norway rats were observed for 30 rain, since these rats showed longer periods of social activity. The reason for using a second strain of rats was that the Brown Norway rats attain a higher age and have a lower incidence of tumors in the brain compared to Wistar and WAG/
-;P~l [ J i
148
]'ABLE 1 BEHAVIORAL CATEGORIES EXPRESSED IN FREQUENCY OF OCCURENCE DURING THE 30-MIN OBSERVATION PERIOD AT 5 DIFFERENT AGES OF BROWN NORWAY RATS
(a) 6.5 Months I. 2. 3. 4. 5. 6. 7. 8. 9. 10. I1. 12.
Approach Social sniffing Anogenital sniffing Avoid contact Defense Submission Parry Autogrooming Aggression Exploration Rest/sitting Mounting
36.92 121.42 17.91 36.72 24.46 2.65 22.29 19,71 15.23 62.37 16.36 28.99
-* 2.59 t''~''~ ± 5 . 9 2 b'~'d
_~. 1.96 b~J~ ± 2.89 a ± 2 . 1 8 b'c'd'e ± 0 . 4 8 b'c'd'c
-~ ± -+ -+
1.77 t''':'d 1.64 b 2,91 ~'~d" 2.91 bx'e
~
1 . 6 8 b'c'd
±
2 . 4 b'~'d'"
(b) 13 Months 56,75 100.25 5,80 41.01 13.54 0.33 1,42 34.62 3.75 121.99 5.63 11.08
(c) 17.5 Months
± 2,39" _+ 5.20 ~" ± 0.84 _+ 3.08 ± 1.36 -~ 0.10 -+ 0.36 ~ ± 1.74 "'d ± 0,87 ± 5.11 c ± (1,84~ _+ 1.38 a~
55.25 92,58 2.21 46.71 16.25 0.04 1.42 25,25 2,58 116.79 2.79 7.75
(d) 24 Months
± 3,87 e -+ 5.06 e ± 0.37 ~ ± 3.51 ± 1.93 ~ ± 0.01 _~ 0.38 ~ ± 1.54 +_ 0.50 ± 4.69 e ± 0.69 ~ ± 1.27
65.26 98.50 2,72 51.56 13.62 (L37 2,21 25.27 5.33 122,83 1,93 %43
~e) 3t Months
___ 4 . 3 r ± 5.53" = 0.44 ± 3.82" ± 1,67 ± 1).12 +- 0,67" + 2.06 -+ 1.04 _~ 4.88 ~" _~ 0.71' = 0.64
29,63 69,75 5.~'~3 34.13 9,63 0.63 12.63 26.63 2.75 58.63 17.88 475
_t: 4,51 ~ 9.48 _~:: 1.27 _~: 5,83 :: ~ 56 =: ().21 t: 2.04 * 298 ~ 0.75 .'.: 8,3I :~ 2 3N r !.!!
Mean ( - S E M I ; samples from observations at 6.5, 13, 17.5, 24 months (Experiment II) and 30 months (Experiment Ili) of age in Brown Norway rats. Frequencies were compared using the Mann-Whitney U-test; the level of significance (p<0.05) was corrected for the number of tests and adjusted to p<0.001. Significant differences are indicated by the letters b,c,d.e (superscript) referring to the age class.
RIJ rats (6). T h e data o f 4 animals, w h o s h o w e d infections at the e y e s at an age o f 11 m o n t h s were not included in o u r analysis. T h e n u m b e r o f a n i m a l s used for the study w a s 24 until an age o f 17.5 m o n t h s . After this age the n u m b e r o f a n i m a l s from w h i c h social b e h a v i o r could be recorded gradually declined. A n imals suffering f r o m infections (eye or m i d ear) were e x c l u d e d f r o m further e x p e r i m e n t a t i o n , w h i c h left us with ten a n i m a l s at an age o f 25 m o n t h s . F r e q u e n c i e s o f behavior at the age o f 6.5, 13, 17.5 and 24 m o n t h s h a v e been collected to identify c h a n g e s in types o f behavior in a g i n g rats. For a m o r e detailed description o f the behavior the transition matrix o f the 2 3 - m o n t h - o l d rats (N = 12) was c o m p a r e d with the transition matrix o f the 6.5 m o n t h old B r o w n N o r w a y rats (N = 24),
Experiment 111 T h e repeated observation o f the s a m e a n i m a l s m a y result in habituation to the social interaction test, w h i c h m a y partially account for a decline in social interest. Therefore naive aged B r o w n N o r w a y rats were subjected to a similar experimental procedure to a s s e s s if naive aged a n i m a l s differ f r o m regularly o b s e r v e d s e n e s c e n t a n i m a l s . T w e n t y - s e v e n B r o w n N o r w a y rats were p u r c h a s e d f r o m T N O Rijswijk, T h e Netherlands, at an age o f 24 m o n t h s . A f t e r o n e m o n t h o f a d j u s t m e n t to laboratory conditions they were subjected to the social interaction test 7 t i m e s at the ages o f 25, 26, 28, 29, 31, 32 and 33.5 m o n t h s . In addition to the frequencies o f behavior at the ages m e n t i o n e d above the f r e q u e n c i e s o f the 12 b e h a v i o r s e x p r e s s e d at an age o f 31 m o n t h s h a v e been a d d e d to Table 1.
ior displayed by their partner, as c o m p a r e d to the first 10 ± i n . Y o u n g a n i m a l s m a i n t a i n e d their level o f being influenced by the behavior o f their partner during the total 2 0 - m i n observation session. In E x p e r i m e n t II T E values altered a g e - d e p e n d e n t l y as can be seen in Fig. 5 A (left panel), in w h i c h both T E - o t h e r and TEself values are g i v e n from 4 until 25 m o n t h s o f age per 10-min time period. Until the age o f 12 m o n t h s T E - o t h e r exceed TEself values. After this age T E - s e l f is larger than TE-other and this difference increases age-dependently. T E - s e l f e x c e e d i n g T E - o t h e r was first and m o s t clearly seen in the last 10 m i n of the observation session. Variance as calculated by the Jack-knife m e t h o d w a s always b e t w e e n 0.25 and 1.30 with an average o f 1 - 2 % o f the T E value; thus, the variance per trial b e t w e e n the a n i m a l s is rather small, w h i c h underlines the within trial stability o f the proposed index. For statistical c o m p a r i s o n o f different points o f the curve the intervals o f confidence can be set to 2%.
sociability m aging rats
100-
[--"]intra
~inter
:¥
50
i
Fq
I r'-I
r---1
RESULTS I
Figure 4 s h o w s the T E coefficients for 4 - m o n t h and 24m o n t h - o l d W A G / R I J rats. D u r i n g the first 10 m i n o f the social encounter, T E - o t h e r values for y o u n g a n d aged rats were higher than T E - s e l f values. T h e b e h a v i o r o f both y o u n g and aged animals w a s m o r e d e t e r m i n e d by the preceding behavior o f their partner than by their o w n preceding behavior. D u r i n g the second 10 m i n , h o w e v e r , the T E - o t h e r values for aged a n i m a l s decreased. T h e s e n e s c e n t rat s e e m e d less susceptible to the behav-
1
21 old
I 1
21 young
FIG. 4. Comparison of the predictability of behavior of young (4 months) and old (23 months) rats. Predictability is based on preceding behavior of the animal itself (expressed in TE-self) and the behavior of its partner (represented by TE-other). The observation period was split into 2 blocks of 10 min: 1 and 2. An asterisk indicates levels of significance of p < 0 . 0 5 , as assessed by the Jack-knife method for calculating confidence intervals.
DECREASED SOCIAL ATTENTION IN AGING RATS
30-
sociabilit
A
in aging rats
149
B
10090-
28-
80-
26'
.~ 7o._>
24-
60-
22-
~.
2018-
:30-
30-
20
28-
b
4
'
~
'
1'2 '
116 '
210 '
2n4mnth
2624.
FIG. 6. The total number of the 12 behavioral categories displayed after a behavioral change of the partner have been summed in totals representing the total amount of behavioral reactivity. Squares, filled circles and open circles represent first, second and third ten-min block, respectively.
22. 20. 18. 30. 2826" 24. 222018. 16" 14. ~
n
t
h
25
27
29
31
33mnth
FIG. 5. Illustration of the age-dependent increase in the predictability of behavior based on the animal's own preceding behavior (TE-self, filled circles) and the age-dependent decrease in predictability of behavior based on the behavior of another rat (TE-other, open circles). These indices have been calculated for the first, second and third 10 min of every observation period for a group of rats observed from 4 to 25 months of age (panel A, left) and for a group of rats observed from 25 to 33 months of age (panel B, right). The division on the right side of panel B also concerns panel A.
The fluctuation of TE-other values between trials is not surprising, since circumstances and experience of the animals may change over an interval between successive observations. The total number of behavioral actions occurring after a behavioral change of the partner has been assessed and is depicted in Fig. 6. The values representing behavioral reactivity have been derived from parts of total matrices corresponding with behavior directed towards the other animal. Since both TE values and these totals are derived from the same matrices the variability must be in the same order of magnitude. It is evident that no linear age-dependent difference in the number of behavioral changes displayed after a behavioral change of the partner could be observed in any of the 10-min observation blocks (Fig. 6). The comparison of Fig. 6 and Fig. 5 yields two arguments for a dissociation between behavioral reactivity and predictability (TE values). First, an age-dependent decrease in behavioral reactivity is seen most prominently in the third 10-rain block, whereas the age-related changes in TE indices are present in all three blocks of 10 min. Second, the profile of the age-related changes in TE values does not correspond with the profile of the values representing behavioral reactivity. Thus, the decreased TE-other values, as shown in Fig. 5A and B, appear related to a gradual diminishment in specific action-reaction combinations between two animals rather than related to a decrease in behavioral reactivity. Pathway diagrams of young and aged rats have been composed and are compared in Fig. 7 to illustrate which specific combinations attribute to the decreased predictability of social behavior in aging rats. The main differences between the path-
way diagrams representing combinations of preceding and following behaviors of young animals (left panel) and the pathway diagram of aged animals (right panel) are indicated by black arrows (Fig. 7). In each pathway diagram preceding behaviors are drawn at the left side and following behaviors at the right side. Young animals show more specific combinations of behaviors, i.e., more arrows, than old animals. Social sniffing [2], aggressive behavior [9], rest/sitting [11] and mounting [12] are followed by social sniffing [2], defensive [5] and aggressive behavior [9], autogrooming [8] or rest/sitting [11], respectively. Exploration [10] as displayed by young animals is not considered as a social behavior and seems more performed at the same time rather than induced by exploratory behavior of the partner; therefore this " a r r o w " is not made black. For similar reasons the mutual dependency of rest/sitting in old rats is not highlighted by a black arrow. Figure 5B (right panel) shows the TE-self and TE-other values for rats from 25 to 33.5 months of age. The TE-other values decline and the TE-self values increase rapidly in all three blocks of 10 min. Only during the first observation TE-other exceeds TE-self in all 10-min blocks. In Experiment II this feature changes after 6 trials in the first 10 min of the observation session; in the senescent, unexperienced rat the change in the ratio TE-other versus TE-self occurs already after one trial (compare Fig. 5A and B, first 10-min panels). In addition, this change was seen in Experiment II after an inter-trial interval of 3 months (8.5 to 11.5 months). Since behavior is usually expressed in frequencies and our analysis is based on the sequential ordering of frequencies, the " r a w " frequencies of the twelve behavioral categories scored during 30 min at 5 ages are given in Table 1. Thirteen-monthold animals show significantly more approach, self-grooming, exploration and less social sniffing, anogenital sniffing, defense, parry, aggression, sitting and mounting than 6.5-month-old animals. From 6.5 to 13 months there seems to be a change from agonistic behavior towards more social sniffing and exploratory behaviors. From an age of 13 to 24 months the frequencies appear rather similar. The senescent rats of 31 months display a significant decline in the frequencies of approach, social sniffing, avoid, defense, exploration and an increase in parry, rest/ sitting as compared to the age of 24 months. DISCUSSION
From behavioral elements of an ongoing stream of behavior the probability of occurrence of each behavioral element, given
t5n
:'~PR[qt ~ pathway diagram of social behavior
/~'~
young
@ ///
@
®/,/
® ®
FIG. 7. Transitions in behavior between the 12 behaviors of young (6.5 months, left panel) and old animals (25 months, right panel) are depicted as arrows connecting the two behaviors. An arrow must be interpreted as "is followed by." The thickness of the arrow refers to the adjusted residual value of the particular cell of the transition matrix. Adjusted residuals indicate the probability of occurrence assessed by comparing the observed number of transitions with a random expectation of transitions. Only those transitions are depicted as arrows when reaching a level of significance of p<0.01. For more explanation see text and Spruijt and Gispen (15). Black arrows represent behavioral combinations only displayed by young animals, whereas white arrows represent behavioral combinations occurring in both, young and aged rats.
the preceding behavioral events, can be calculated. The TE index representing the predictability of an event is either based on preceding behavior of the animal itself (TE-self) or based on the preceding behavior of another animal (TE-other). This index gives insight in how far ongoing behavior is determined by the animal's own preceding behavior (TE-self) or by the behavior of another animal (TE-other). Thus, a high TE-other value parallels a high level of communication and a low TE-other value indicates little specific responsiveness to the behavior of a social partner. A high TE-self value represents a rigid behavioral pattern, which is governed by the own preceding behavior of the animal. It appears that young animals--both WAG/RIJ rats and Brown Norway rats--are more influenced by the behavior of their partners than by their own preceding behavior, as TE-other values exceed TE-self values. However, in the senescent rat this relationship is reversed: now the own preceding behavior determines ongoing behavior more than preceding behavior of another rat. This dramatic decline of specificity of following behaviors after a certain preceding behavior of the other rat progresses gradually in aging rats, as is demonstrated by the results of this longitudinal study. The decline in TE-other values is parallelled by increased TE-self values. This reversed TE-self/TE-other ratio cannot be explained by a general decrease in the total number of acts displayed after a behavioral change of the partner, i.e., the degree of reactivity. Changes in the frequencies of the 12 behaviors perhaps suggest a change in the nature of the interaction occurring between 6.5 and 13 months and after 24 months, but do not reveal changes in the period from 13 to 24 months. The nature of the interaction reflected in frequencies appears rather constant in the time period from 13 to 24 months, whereas in this period the
FE-other values gradually decline, The frequencies ot me ~ar~ ous behaviors are difficult to interpret: some behaviors iapproach, avoid contact, self-grooming and explorationi increase in frequency, whereas others (social sniffing, defense, aggces sion, mount) decrease in 24-month-old animals as compared to 6.5-month-old animals. One might speculate that contact behav iors, with the exception of approach, generally decrease, wherea,~ nonsocial behaviors increase in frequency. These changes can be reconciliated with more overt changes in the pathway diagram and TE values. However, the mere change in frequency ,,3 a certain behavior is dissociated from a particular preceding or following behavior of another animal. Approach is not associated at all with any particular behavior of the partner. In the senescent rat social sniffing is only preceded by defense, and exploration, self-grooming and aggression are not preceded by a particular behavior despite the increase in frequency for approach, self-grooming and exploration. Likewise, changes m the, TE values are not parallel to changes in total frequencies of behavior. The alteration in TE values occurs more gradual and in a more relevant period for aging (13.5 to 24 months~. Another explanation for a gradual decline in performance in a longitudinal setup may be the repeated exhibition to a similar test situation (habituation). Therefore, the social behaviors of senescent rats, which had never been exposed to such social interaction test before, were studied; initially, they showed larger TE-other values, but rapidly they reached the level of the aged animals of Experiment II. Apparently, the novelty o1 a first exposure to a social interaction test induces in aged rats a q~;-other/TE-self ratio similar to the ratio seen in younger rats. However. already during the first exposure the difference between TE-self and TE-other is larger in younger rats than the difference seen in the aged rat. After the first test the decline in TE-other values and the increase in TE-self values occurs quicker than in younger animals. Thus the degree of social responsiveness in aged rats seems to be partly influenced b~ previou~ experience, but i~ mainly age-dependent. Other studies suggested that alterations m the structure ol behavior underlie age-dependent changes in behavior as has been demonstrated for sexual behavior (16) and grooming behavior (17). In the latter studies changes were illustrated with complex diagrams representing the combination of behaviors studied. In the present study such complex behavioral alterations are reduced to one index. This method offers the possibility to define characteristics such as changes in flexibility or rigidity which are also known to occur in aging humans (1). Changes in the organization of behavior and the seeking of information may underlie age-related cognitive deficits (2). Changes in social behavior assessed with an information-statistical method have also been demonstrated in other animals such as the Java monkeys (Macaca fascicutaris). Manipulation of the nucleus caudatus induced changes in communication patterns, which were expressed in changes in TE values in socially housed monkeys (21). This method has also been applied in our laboratory to assess the behavioral reflection of the onset of aging and to investigate the effects of ACTH-Iike neuropeptides on such age-related behavioral changes. The advantage of the present index compared to the registration of mere frequencies and durations is that in one TE value all the information concerning the specificity of combinations of the behavior of two interacting animals is compressed. Furthermore, this index is more appropriate for the interpretation of complex behavioral changes as is illustrated by the comparison with a change in the distribution of various frequencies. In the present social interaction test I suggest that this index reflects very adequately the level of social attention (TE-other) in aging rats.
D E C R E A S E D S O C I A L A T T E N T I O N IN A G I N G RATS
151
ACKNOWLEDGEMENTS The authors wish to express their gratitude to Marlou Josephy for her advanced biotechnical assistance. This study was supported by a Constantijn and Christiaan Huygens Career Development Award to Berry Spruijt received from the Netherlands Organization for the Advancement of Pure Research (N.W.O.).
REFERENCES 1. Botwinnick, J. Neuropsychology of aging. In: Filskov, S. B.; Boll, T. J., eds. Handbook of clinical neuropsychology. New York: Wiley; 1981:135-171. 2. Buchanan, S. L.; Powell, D. A. Age-related changes in associative learning: studies in rabbits and rats. Neurobiol. Aging 9:523-534; 1988. 3. Chatfield, Ch.; Lemon, R. E. Analyzing sequences of behavioural events. J. Theor. Biol. 29:427-445; 1970. 4. Fagen, R. M. Information measures: statistical confidence limits and inferences. J. Theor. Biol. 73:61-79; 1978. 5. Grant, E. C. An analysis of the social behavior of the male laboratory rat. Behaviour 2 1:260-281 ; 1963. 6. Hollander, C. F. Current experience using the laboratory rat in aging studies. Lab. Animal Sci. 26:320-328; 1976. 7. Jolles, J. Neuropeptides and the treatment of cognitive deficits in aging and dementia. Prog. Brain Res. 70:429-441; 1986. 8. Larsson, K. Sexual activity in senile male rats. J. Gerontot. 13:136-139; 1958. 9. Losey, G. S. Information theory and communication. In: Colgan, P. W., ed. Quantitative ethology. New York: John Wiley & Sons; 1979:43-78. 10. Martinez, J. L.; Schulteis, G.; Janak, P. H.; Weinberger, S. B. Behavioral assessment of forgetting in aged rodents and its relationship to peripheral sympathetic function. Neurobiol. Aging 9:697708; 1988. 11. Miller, G. A.: Madow, W. G. On the maximum likelihood estimate of the Shannon-Wiener measure of information. Air Force Cambridge, Research Center Technical Report AFCRC-TR-54-75, 1954: 1-22. 12. Moss, M. B.; Rosene, D. L.; Peters, A. Effects of aging on visual recognition memory in the rhesus monkey. Neurobiol. Aging 9:495-
506; 1988. 13. Neurobiology of Aging. Special issue on experimental models of age-related memory dysfunction and neurodegeneration. 9:445-765; 1988. 14. Olton, D. S. Aging and spatial cognition. Neurobiol. Aging 9:469474; 1988. 15. Spruijt, B. M.; Gispen, W. H. Behavioural sequences as an easily quantifiable parameter in experimental studies. Physiol. Behav. 32: 707-710; 1984. 16. Spruijt, B. M.; Meyerson, B. Sequential analysis of the copulatory behavior of young and aged rats. Neurosci. Res. Commun. 1:3138; 1987. 17. Spruijt, B. M.; Welbergen, P.; Brakkee, J.; Gispen, W. H. An ethological analysis of excessive grooming in young and aged rats. Ann. NY Acad. Sci. 525:89-100; 1988. 18. Thompson, R. F. The rosetta stone for brain substrates of age-related deficits in learning and memory? Neurobiol. Aging 9:547-548; 1988. 19. Timmermans, P. J. A. Social behavior in rats, Ph.D. thesis. Nijmegen, The Netherlands, 1978. 20. Van den Bercken, J. H. L.; Cools, A. R. Information-statistical analysis of social interaction and communication: an analysis of variance approach. Anim. Behav. 28:172-188; 1980. 21. Van den Bercken, J. H. L.; Cools, A. R. Information-statistical analysis of factors determining ongoing behaviour and social interaction in Java monkeys (Macaca fascicularis). Anim. Behav. 28: 189-200; 1980. 22. Van Hooff, J. A. R. A. M. Categories and sequences of behaviour: methods of description and analysis. In: Ekman, P.; Scherer, K., eds. Handbook of non-verbal communication research. New York: Cambridge University; 1982:362--439.