BRAIN,

BEHAVIOR,

AND

IMMUNITY

5, i?o&218

(1991)

Thymic Peptides, Stress, and Depressive Symptoms in Older Men: A Comparison of Different Statistical Techniques for Small Samples CAROLYN M. ALDWIN' Human

Development

and Family Studies, of California,

Department qf Applied Behar~ioral Davis, Culifornia 95616

AVRON SPIRO III ANDGEORGE Normutive

Aging

Study, Boston Veteruns Administrution Health, Bnston Uni,,ersity. Boston.

Sciences.

University

CLARK

Outpatient Massachusetts

Clinic und School 02215

of Public

AND NICHOLAS HALL Department

of Psychiatry

und Behuvioral Medicine. Tumpu. Floridu 33620

University

qf South

Florida,

Thymic peptides play an important role in aging and immune regulation, but little is known about their relationship to psychosocial factors. One thymic fraction. thymosin-cu, (TSN-n,) may be of particular interest given its hypothesized role in the differentiation of immature T cells into functional. mature T cells. We examined the relationships among stress. psychological symptoms, and TSN-cur levels in two conditions: before and after a glucose challenge test. The sample consisted of 18 men. aged 48-80. participants in the Normative Aging Study. While none of the correlations reached significance in the baseline condition, life events and depressive symptoms were significantly correlated with TSN-o, in the postchallenge condition (r’s = .57, and .61. respectively). Hierarchical regression analyses with cross-product interaction terms suggested that individuals who were high in both life events and depression showed the highest levels of postchallenge TSN-IX,. with the psychosocial variables accounting for 65% of the variance. Given the small sample size, we replicated these analyses using jackknife and bootstrap techniques, which generally confirmed these findings. Thus. these preliminary results suggest that psychosocial factors may be related to abnormal TSN-u, responses to a challenge.

INTRODUCTION Psychoneuroimmunology

in Older

Populatiorls

Literally hundreds of studies have been conducted on the impact of stress on neuroendocrine and immune functioning in both animals and humans in the past decade (for reviews, see Ader, 1981: Baker, 1987; Melnechuk, 1988). However, relatively few studies have examined psychosocial factors and immune compe’ To whom

correspondence

should

be addressed. 206

0889-1591191

$3.00

TIHYMIC

PEPTIDES,

STRESS,

AND

DEPRESSIVE

SYMPTOMS

207

tence in older adults (Biondi & Pancheri, 1987). The elderly have higher rates of both cancers and autoimmune factors, suggesting that they might have poorer regulation of both “internal” and “external” immune functions (e.g., Hausman & Weksler, 1985). A major limitation to studying psychoneuroimmunology in the elderly is that older individuals often have chronic illnesses; the impact of illness on immune functioning may overwhelm more subtle psychosocial effects. Given the difficulty and cost of immune assays (cf. Kiecolt-Glaser & Glaser, 1989), and finding healthy older populations, the samples in such studies tend to be very small. Thus, it is not surprising that the findings concerning the relations between psychosocial and immune factors in older samples have been mixed, varying by both the psychosocial factor assessed and the immune outcome. For example, one study found tlhat social support was modestly correlated with total lymphocyte counts in older people (Thomas, Goodwin, & Thomas, 1985). However KiecoltGlaser and her colleagues (1985) did not find an impact of social support on natural killer (NK) cell activity in older women, but they did report that both loneliness and the use of relaxation techniques affected immune competence. Other studies, though, have not confirmed this link between loneliness and NK cell activity in older women (Mishra, Colby, Milanesi, Cesario, & Yousefi, 1990). While stressful life events do not appear to be correlated with NK cell activity in older samples (Mishra, Aldwin, Colby, & Oseas, 1991; Mishra et al., 1990). several studies have found that psychological distress in response to both major and minor stresses may particularly affect several immune parameters in older populations, including NK cell activity, T-helper/T-suppressor cell ratios, IgE and IgM (Irwin, Daniels, Bloom, Smith, & Weiner, 1987; Mishra et al., 1990b; Zautra, Okun, Robinson, Lee, Roth. & Emmanual, 1989). These studies suggest that it may be important to examine the interaction between stress and psychological distress when studying the relationship between psychosocial factors and immune functioning, especially in older populations. Further, investigating the link among stress, regulatory hormones, and immunity may help clarify the relationship between stress and immune system functioning. Thymic Peptitles The thymus gland is central to immunocompetence, especially in the maturation of T cells. It normally becomes involuted with age (e.g., it becomes progressively infiltrated with fat), which has been correlated with a decline in some measure of immunologic function (Steinman, 1986). Plasmic levels of thymic peptides have also been found to decline with age (Koninkx, Schreurs, Penninks. & Seinen. 1984: Lewis, Twomey, Bealmeer, Goldstein, & Good, 1978: Steinman, 1986), although these changes are not always correlated with a decrease in immunocompetence. Epithelial cells of the thymus secrete several peptides with hormone-like factors. They maty be secreted by different regions of the thymus gland and may be biochemically distinct (Goldstein, Low, Zatz. Hall, & Naylor, 1983). While the thymus is the primary source of thymic peptides. many of them are found in extra thymic sites (Spangelo, Hall, & Goldstein, 1987). The first thymic peptide to be

208

ALDWIN

ET

AL.

isolated was thymosin fraction 5 (TF5). A purified and widely studied component of TFS is thymosin-c-u, (TSN-a,). TSN-o, has been hypothesized to play a key role in the differentiation of immature T cells and the stimulation of various immune cell functions. Administration of TSN-a, has been reported to reverse the effects of immune suppression due to elevated levels of cortisol in rir~ and in vitro. Thus, TSN-(-Y, may be a central regulator of immune function, although controversy still exists concerning its role and measurement (for reviews, see Clark, Hall, Aldwin, Goldstein, & Steiner, 1990; Goldstein et al., 1983; Spangelo et al., 1987). Psychological stress has been hypothesized to affect thymic function. Cortisol has been shown to increase thymic involution in rats and suppress thymic functioning (cf. Borysenko & Borysenko. 1982). An extensive review of the literature, however, found only one TSN-a, study which examined whether psychosocial factors were related to thymic peptide levels in humans. Hoon, Hoon, Rand, Johnson, Hall, and Edwards (1991) found that psychosocial stress was not directly related to TSN-a, levels in a sample of university students, but clearly additional work on this relationship is needed. Present Study This study has two purposes. The first is to present preliminary data on the relations among TSN-(Y,, stress, and psychological symptoms in a sample of healthy older men. Given that life events were assessed over the course of a year, while hassles and psychological symptoms were assessed in the prior 3 months, we expect that both hassles and psychological symptoms will be more related to TSN-a, than life events. We will also explore the interaction effects of stress and symptoms on TSN-o,. The second purpose is to explore the relevance to psychoneuroimmunology of statistical methods developed for small-sample studies. In such studies, the standard errors (and thus the statistical significance) can be unduly influenced by outliers or inadequate sampling. This is especially a problem when psychosocial effects on physiological processes tend to be subtle and hard to demonstrate. A time-honored way of controlling for nonnormal distributions is to use nonparametric statistics such as Spearman’s p or the Mann-Whitney U test. Newer, more sophisticated techniques that involve multiple resampling, such as the jackknife (Hinkley, 1978) and bootstrap techniques (Efron & Gong, 1983). may provide more accurate tests of significance for small samples. Thus, we will contrast findings from both parametric and nonparametric analyses. METHOD

Sample The sample consists of 18 men (mean age = 67.6, SD = 8.9, range = 48-80), participants in the Normative Aging Study (NAS). The NAS is a longitudinal, biomedical study of aging initiated in 1963 (Bosse’, Ekerdt, & Silbert, 1984). The original selection criteria for the men included absence of major disease, blood pressure below 140/90, and geographic stability, as evidenced by extensive social networks and stated intentions to remain in the area. The 18 men in the present

THYMIC

PEPTIDES,

STRESS,

AND

DEPRESSIVE

SYMPTOMS

209

study had completed data from two studies: the first a joint study of anthropometrics and diabetes (Sparrow, Borkan, Gerzof, Wisniewski, & Silbert, 1986) and of neuroendocrine and immune factors (Clark, Hall, Aldwin, Harris, & Srinivasan, 1988), and the second a mail survey on stress and mental health (Bosse, Aldwin, & Levenson, 1987). Both studies were conducted in the late spring and early fall of 1985. The men had been screened for absence of chronic disease before participating in the first study. Measures

and Procedures

The men re,ported for their examination following an overnight fast. Blood was drawn at 8:00 AM (baseline condition) and again 2 h later following a 100-g glucose challenge test (postchallenge condition). TSN-a, was measured by radioimmunoassay using an antibody made against a synthetic form of peptide and an I”‘labeled tracer (McClure, Lameris, Wara, & Goldstein, 1981). This assay is sensitive enough to detect thymic peptide levels in both aged humans and thymectomized mice (McGillis, Hall, & Goldstein, 1983). Two measures of stress were used. Major life events in the past year were assessed using the Elders Life Stress Inventory (ELSI; Aldwin, 1990). The ELSl is a 3I-item inventory assessing events of particular relevance to older adults, such as a child’s divorce, institutionalization of spouse, and taking major responsibility for a parent. Events are rated on a scale of 1 to 5, with I indicating that an event was not at all stressful, and 5 indicating that it was extremely stressful. Since a simple count of life events correlated as well with outcomes as did the summed stress ratings, the former was used. Hassles occurring in the past 3 months were assessed in five different areas: health, marital, work/retirement, social, and financial. Each area was rated on a scale from 0 to 7, with 0 indicating that no event occurred, and 7 indicating that it was highly stressful. Scores were summed and divided by the number of valid domains (see Aldwin, Levenson, Spiro. & Bosse, 1989). Note that only 16 of the 18 men had valid data on the hassles scale. Psychological symptoms were measured using the SCL-90-R. a standard index of psychologi’cal distress (Derogatis, 1983). The respondents indicated on a fivepoint scale how distressed they were by each of 90 symptoms during the past 3 months. The SCL-90-R yields an overall measure of mental health, the Global Severity Index (GSI). and nine scales of more specific symptom categories. The mean GSI score of this sample (.29) was similar to published norms (.26). The means. standard deviations, skewness, and kurtosis for all variables are presented in Table I. Analyses

Pearson correlations were computed between TSN-a, under the two conditions (baseline and after the glucose challenge test) and stress and psychological symptoms. Given the limitations of parametric statistics for small samples (N < 25) and the sensitivity of Pearson correlations to nonnormality (e.g., Edge11 & Noon, 1984; Kowalski, 1972), we replicated the analyses using a nonparametric correlation (Spearman’s rank order correlation). In addition, resampling methods (the

210

ALDWIN

Descriptive Variable TSN-a, Baseline Postchallenge Stressor Life events Hassles” Psychological symptoms GSI Somatization Depression Anxiety Phobic anxiety Obsessive-compulsive Paranoid ideation Interpersonal sensitivity Hostility Psychoticism

Statistics

ET AL.

TABLE I for All Variables

(N

= 18)

Mean

SD

1887.00 2578.22

722.3 I 901.54

.I5 .98

2.33 10.56

2.43 7.34

1.88 .83

5.18 .56

.29 .40 .35 .14 .03 .46 .23 .24 .25 .I6

.32 5’.45 .3l .ll .55 .26 .28 .40 .25

I .70 2.17 1.83 3.62 3.40 1.57 .59 1.50 2.16 I .52

2.50 4.40 3.49 14.07 12.59 2.37 - .08 1.71 5.04 1.26

Skewness

n N = 16.

jackknife and bootstrap) were used to provide nonparametric estimates of standard errors for both correlation and regression coefficients. Both the jackknife and bootstrap estimators are obtained by resampling the observed data and are relatively computer-intensive. These nonparametric estimators are versatile in that they can be used in multivariate as well as bivariate analyses. All analyses were done using PC SAS (SAS, 1987).’ Results were considered valid only if they were consistent across all of the analytic techniques used in this study. The jackknife estimate (Hinkley, 1978) is obtained by computing the estimator in question N times, where N is the number of observations in the data. In each computation, one successive observation is omitted. The results of the N computations are then combined (for details, see Hinkley, 1978). and the resultant standard error is less biased than the usual normal-theory estimate and provides a more accurate measure of the actual variability in the data. The bootstrap estimate (Efron & Gong, 1983) proceeds in a similar fashion. However, rather than systematically deleting each observation in turn, the bootstrap proceeds by sampling, with replacement, from the observed data and computing the statistics for each sample. If a “large” number of samples is drawn (e.g., in our case, we used 400 for the regressions), the observed sampling distribution of the statistics provides an estimate of the parameter’s standard error. In effect, the bootstrap substitutes sampling from a known empirical distribution for sampling from an unknown, theoretical one (Lunneborg, 1985). In both the jackknife and bootstrap estimation of the correlation coefficients, the correlations were transformed by Fisher’s z prior to averaging over samples, ’ Sample

programs

for the bootstrap

and jackknife

procedures

are available

from

the second

author.

THYMIC

PEPTIDES,

STRESS,

AND

DEPRESSIVE

SYMPTOMS

211

and the resulting average was then back-transformed to a correlation (e.g., Lunneborg, 1985; Silver & Dunlap, 1987). In the bootstrap computation of correlations, we used 250 repetitions. To examine the interaction effects between stress and specific psychological symptoms on postchallenge TSN-a, levels, we used hierarchical regression analysis. As suggested by Cohen and Cohen (19X3), the terms comprising the interaction were centered at their respective means prior to computing the interaction, in order to reduce possible collinearity. Each hierarchical analysis involved generating four regression equations, each adding a variable in successive steps. Baseline TSN-aU, was entered first. to control for individual differences in thymic peptide levels. The two main effects, stress and the psychological symptom score, were added in the next two equations, and finally the stress x symptom interaction was entered. The regression coefficients (bs) reported below are from the full model and indicate the effect of each variable controlling for all others. However, the R’ increments (AR’) are from the successive equations and indicate the contribution of that variable to the total R’. controlling only for the variables entered earlier in the model. All regressions were done first using ordinary least squares (OLS), then repeated using the jackknife and bootstrap procedures. RESULTS

The average level of TSN-a, at baseline was 1887.0 ng/nl, SD = 722.31; in postchallenge condition it was 2168.22 nginl, SD = 901.54. This increase did reach significance in this sample, r( 17) = - I .39, ns. The correlation between baseline and postchallenge TSN-a, levels was .46 (p < .lO). TSN-(w, was associated witrh age at each time (r = .04 and .13. respectively). Bivariate

the not the not

Analyses

Table 2 presents the correlations of the baseline assessment of TSN-(r, with the two measures of stress and the 10 indices of psychological symptoms. In the baseline condition, none of the Pearson correlations reached significance, ranging from - .I0 to .28. This finding was confirmed by the three nonparametric correlation analyses. Thus, psychosocial factors do not appear to be related to baseline measures of TSN-a,. However, in the postchallenge condition, a very different picture emerged (see Table 3). Exalmining the Pearson correlation coefficients, the relation between overall symptom levels (GSI) and TSN-a, levels approached significance (r = .46, p = .057). The Pearson correlations achieved significance for three of the subscales: depression (r = .60), interpersonal sensitivity (r = .62), and psychoticism (r = .54). Life events were also significantly correlated with TSN-a, in the postchallenge condition (r = .57). while the correlation with hassles approached significance (r = .44, p < .lO). All three nonparametric correlations, Spearman’s p, the jackknife, and the bootstrap, confirmed the relations between postchallenge TSN-a, and the psychological symptom scales (GSI, depression, interpersonal sensitivity, and psychoticism). However, the nonparametric correlations between life events and TSN-a, were not significant, suggesting that the parametric statistics may have

212

ALDWIN

Correlations

between

ET

AL.

TABLE 7 Thymosin-u, (TSN-(r,) and Psychological Older Men (N = 18)

Baseline

Baseline

Psychosocial factors Stress Life events Hassles Psychological symptoms Somatization Depression Anxiety Phobic anxiety Obsessive-compulsive Paranoid ideation Interpersonal sensitivity Hostility Psychoticism GSI

Pearson

Spearman

~ .?O

~ .24 -.oo

-.07

.22

.23

.23

.I9

.31 .30

.07

.Ol

.32 p.10 .22

.30 p.09

.22

.25

.28

LI No correlation is significant at or below the IO level. b For the bootstrap, the correlation shown is the usual Pearson cance test is based on the bootstrap estimate of its standard error.

in

TSN-ol,”

Jackknife

.28 .31

.28 .28

Symptoms

(SE)

Bootstrap

-.I9

(.17)

-.20

-.08

(.27)

-.07

.21 f.13) .22 (.17) .lI (.lO) .16 1.21) .2l f.19) - .lO (20) .19 (20)

(SE)h

C.16) C.24)

.22 (.13) .23 C.19) .I9 C.14) .07 f.16)

.?2 f.18) ~ .I0 (23) .‘2 (20)

.23 (.I91

.28 C.21)

.26 t.20) .23 (.14)

.28 (21) .25 C.17)

correlation:

however.

the signifi-

resulted from a Type II error. Again, hassles were not significantly related to TSN-a, levels. The Spearman rank-order correlations suggested that three additional SCL90-R subscales, anxiety, hostility, and obsessive-compulsive, were correlated with TSN-a,, but the jackknife and bootstrap estimates did not confirm this. We consider these relations to be artifactual, given that they emerged with only one method. We examined the possibility that the significant correlations in the glucose challenge condition were an artifact of changes in blood sugar level. However, none of the correlations between these levels and measures of stress, psychological symptoms, and TSN-IX, even approached significance (results not shown), which suggested that the relations with TSN-a, were not artifactual. Multivariate

Analyses

As mentioned earlier, we used hierarchical regression analysis to determine whether there was a significant interaction effect between stress and symptoms on TSN-aU,. As we had no specific hypotheses as to which types of symptoms would interact with stress to affect TSN-a, levels, we decided to use depressive symptoms for illustrative purposes for two reasons. First, the relation between depressive symptoms and TSN-a, might be of more theoretical interest, given the widespread use of depressive symptoms in the psychoneuroimmunology literature (cf. Stein, Keller, & Schleifer, 1985). Second, depression had the strongest empirical relation to TSN-a, in this sample. As mentioned earlier. we examined the main and interaction effects of life

THYMIC

Correlations

between

PEPTIDES,

STRESS,

AND

DEPRESSIVE

Postchallenge

TABLE 3 Thymosin-a, (TSN-u,) Older Men (N = 18)

and Psychological

Postchallenge

Psychosocial factors Stress Life events Hassles Psychological symptoms Somatization Depression Anxiety Phobic anxiety Obsessive-compulsive Paranoid ideation Interpersonal sensitivity Hostility Psychoticism GSI

213

SYMPTOMS

Symptoms

in

TSN-a,

Pearson

Spearman

.57+*

.I8 .27

.70 f.65) .42 (261

.57 C.43) .44 c.28)

28

.I6 .54 .I9 .09 .28 .40 .62 .I7 .50 .38

.73 .60 .26 .32 .35 .3X .62 .25 .54 .46

.44 .23 .60*** .26 .32 .35 .38

.67*** i .48** .36 .58** .26 .43* .41* .44* .55**

,63*** .‘5 .54** .46”

“ For the bootstrap, the correlation cance test is baseld on the bootstrap * .I0 G p G .os: **p G .05; ***p

Jackknife

shown estimate

is the usual Pearson of its standard error.

(SE)

Bootstrap

(26) (.18)*** (.54) (.60) C.21) I .32) (.4I)* C.15) (.31)* (?“)* --

correlation;

however.

(SE)”

(20) (.32)*** t.33) C.39) (.?I) t.31) t .29)** (.?I) (.26)** (.22)** the signifi-

c .Ol.

events and depression on postchallenge TSN-a,, controlling for baseline TSN-a, . In the OLS regression, these four variables accounted for a large (86%) proportion of the variance in postchallenge TSN-(Y,. Surprisingly, baseline TSN-a, levels accounted for only a small portion of the variance. 21%, while life events, depressive symptoms, and their interaction accounted for increments of 45. 9, and 12%. respectively. (The F’s for the AR”s were significant beyond the .OOl level.) Thus, the psychosocial variables accounted for three times the amount of variance in postchallenge TSN-a, as did baseline TSN-(-w,. In the full model with all variables entered (see Table 4). it was clear that the interaction between life events and depression was more important than the re-

Summary

of Hierarchial

TABLE 4 Regression Equations Examining Interaction Effects Depression on Postchallenge TSN-CY, Levels tN = IX) OLS

Predictors

b

Baseline TSN-a, Life events Depression Interaction Intercept

.62 107.13 442.88 341.80 906.37

* For the bootstrap.

the regression .Ol;

t p < .lO; *p < .05: **p
g. Btrll. 95, 576-83. Efron. B.. & Gong. G. (1983). A leisurely look at the bootstrap, the jackknife, and cross-validation. Am.

Stut.

37.

3648.

Goldstein. A. L.. Low. T. L.. Zatz. M. M.. Hall. N. R.. & Naylor, P. H. (1983). Thymosins. In J. F. Bach (Ed.). Clinks in immwdo,qy and tr//rrg.v. pp. 119-132. Saunders: London. Hausman, P. B.. & Weksler. M. E. (1985). Changes in the immune response with age. In C. E. Finch & E. L. Schneider (Eds.), Hundhook ofthe hidog! ofogin~c, 7 2nd ed.. pp. 4lti32. Van Nostrand Rheinhold: New York. Hinkley. D. V. (1978). Improving the jackknife with special reference to correlation estimation. Biomerriku 65, 13-2 I. Hoon, E. F.. Hoon, P. W.. Rand, K. H.. Johnson, J.. Hall, N. R.. & Edwards, N. B. (1991). A psycho-behavioral model of genital herpes recurrence. J. Ps~c~hoso~n. Res. 35, 25-36. Irwin. M.. Daniels. M.. Bloom, E., Smith. T., & Weiner. H. (1987). Life events, depressive symptoms, and immune function. Am. J. Pswhiutry 144, 437-441. Kiecolt-Glaser. .I. K.. & Glaser. R. (1989). Psychoneuroimmunology: Past. present. and future. Health

Psyhol.

8, 677~X2.

Kiecolt-Glaser, J. K., Glaser. R., Willinger. D.. Stout. J., Messick, G.. Sheppard. S.. Ricker. D.. Romisher, S. C.. Briner, E., Bonnell, G.. & Donnerberg. R. (1985). Psychosocial enhancement of immunocompetence in a geriatric population. Heulrh t’.s~~~hol. 4, 25411. Koninkx, J. F.. Schreurs, A. D., Penninks. A. H.. & Seinen, W. (1984). Induction of postthymic T-cell maturation by thymic humoral factor(s) derived from tumor cells of epithelial origin. Tltvmu.s 6. 39G-109. Kowalski, C. J. (1972). On the effects of non-normality on the distribution of the sample productmoment cor;relation. Appl. Slur. 21, I-15. Lewis, V. M., Twomey. J. J.. Bealmeer. P., Goldstein, S.. & Good. R. A. I 1978). Age. thymic involution, and circulating thymic hormone activity. J. C/in. Endocrinol. Merub. 47, 145-150.

218

ALDWIN

ET AL.

Lunneborg. C. (1985). Estimating the correlation coefftcient: The bootstrap approach. Psycho/. Bull. 98, 209-215. Makinodan. T. (1980). Immunity and aging. In C. Finch & L. Hayflick (Eds.). Hundbook of the biology ofaging. 1st ed.. pp. 379-408. Van Nostrand Reinhold: New York. McClure, J. E.. Lameris, D.. Wara. &Goldstein. A. L. (1981). Immunochemical studies on thymosin: Radioimmunoassay of thymosin alpha-l. /. Immunol. 128. 368. McGillis. J. P.. Hall. N., & Goldstein. A. L. (1983). Circadian rhythm of TSN-a, in normal and thymectomized mice. J. Immwzo/. 131, 148-151. Melnechuk, T. (1988). Emotions, brain, immunity, and health: A review. In M. Clynes &J. Panksepp t Eds.). Ehorions and ps~ci~opafholo~~~~ Plenum: New York. Mishra, S. 1.. Aldwin, C. M., Colby, B. N., & Oseas, R. S. (1991). Adaptive potential. stress, and natural killer cell activity in older adults. J. Healfh Aging. in press. Mishra, S. I., Colby, B. N., Milanesi, L. C., Cesario. T. C.. & Youseli. S. 11990). Re.sistancr resources. health, und immune firnctioning umong older MWW~I: A cross-cultural perspective. Paper presented at the I Ith Annual Meeting of the Society for Behavioral Medicine. Chicago. SAS Institute (1987). SAS procedures guide,for personal computers. rsersion 6. SAS Institute: Cary. NC. Silver, N. C.. & Dunlap, W. P. (1987). Averaging correlation coefficients: Should Fisher’s : transformation be used’? J. Appl. Psychol. 72, 146148. Spangelo, B. L., Hall. N. R.. & Goldstein, A. L. (1987). Biology and chemistry of thymosin peptides: Modulators of immunity and neuroendocrine circuits. Ann. N. Y. Acud. Sci. 496, 196-201. Sparrow. D., Borkan. G. A., Gerzof. S. G.. Wisniewski, C., & Silbert, J. K. (1986). Relationship of fat distribution to glucose tolerance: Results of computed tomography in male participants of the Normative Aging Study. Diabetes 35, 41 l-415. Stein, M.. Keller, S. E.. & Schleifer, S. J. (1985). Stress and immunomodulation: The role of depression and neuroendocrine function. J. Immunol. 135, 827~833s. Steinman, G. G. (1986). Changes in the thymus during aging. C~vr. Top. Pufhol. 75, 43-88. Thomas. P. D., Goodwin, J. M., & Goodwin, J. S. (1985). Effect of social support on changes in cholesterol level, uric acid level, and immune function in an elderly sample. Am. J. Psychiutr~ 143, 735-737. Zautra, A. J.. Okun, M. A., Robinson, S. E., Lee, D.. Roth, S. H.. & Emmanual, J. (1989). Life stress and lymphocyte alterations among rheumatoid arthritis patients. Healrh Psycho/. 8, l-14. Received

January

16, 1989