Gamma Levels Predict Type 2 Diabetes in the Pima Indian Population Robert S. Lindsay, Jonathan Krakoff, Robert L. Hanson, Peter H. Bennett, and William C. Knowler

It has been proposed that inflammation or infection may contribute to the development of type 2 diabetes. We examined whether serum gamma globulin, a nonspecific ype 2 diabetes and obesity are associated with a measure of the humoral immune system, predicted cytokine-associated acute phase reaction (1). changes in glucose tolerance in 2,530 members of the This has led to a recently proposed hypothesis Pima Indian population, a group with a marked predis- that elements of the innate immune system, such position to type 2 diabetes. Cross-sectionally, gamma T as cytokines or the acute phase reactants that they stim- > P ,0.08 ؍ globulin was positively related to age (r ulate, contribute to the development of type 2 diabetes and ؍ 0.0005), BMI (r 0.09; P < 0.0001), and female sex obesity (1). In this model, the innate immune system (P < 0.0001). Gamma globulin concentrations were ,mediates the effects of factors such as nutrition, age ؍ familial, being positively correlated among siblings (r 0.23; P < 0.0001) and between parents and their chil- genes, fetal programming, and ethnicity on the later devel- P < 0.0001; father/child: opment of metabolic disease (1). Inflammatory mediators ,0.17 ؍ dren (mother/child: r ,P < 0.0001). Gamma globulin concentrations have been extensively studied in coronary heart disease ,0.25 ؍ r were higher with greater degrees of American Indian with convincing evidence that increased concentrations heritage (P < 0.004, with adjustment for age, sex, and of fibrinogen, C-reactive protein, , and leukocyte BMI) and in the presence of a family history of type 2 count are associated with higher rates of subsequent cor- diabetes (P < 0.04). Higher gamma globulin levels onary events (2). Similar evidence for prediction of type 2 predicted risk of diabetes. In univariate analysis,a1SD diabetes is less extensive. Higher fibrinogen and white cell difference in gamma globulin was associated with a 20% count and lower were all found to predict higher incidence of diabetes in those who were normal glucose tolerant at baseline (hazard rate ratio 1.20 [CI later type 2 diabetes in a single study (3). More recently, 1.11–1.30]; P < 0.0001) and remained as a significant a preliminary report has suggested that plasminogen acti- predictor of diabetes, even when controlled for effects vator inhibitor-1 (PAI-1) predicts the development of dia- of sex, BMI, and 2-h glucose as additional predictors betes (4). (hazard rate ratio for 1 SD difference in gamma globu- The Pima Indian community of Arizona has a very high Gamma globulin was prevalence of type 2 diabetes and obesity (5). As part of .(0.002 ؍ lin, 1.14 [1.05–1.24]; P also associated in univariate analysis with later devel- the long-standing epidemiological study in this population, opment of impaired glucose tolerance (IGT) (hazard community members have been invited to attend exami- rate ratio 1.15 [1.07–1.23]; P < 0.0001), but not with the nations for the presence of diabetes since 1965. In the transition from IGT to diabetes (hazard rate ratio 1.04 early years of the study (1966–1973), serum concentra- -Thus, gamma globulin levels pre .(0.6 ؍ P ;[1.20–0.90] dict increased risk of diabetes in the Pima population. tions of gamma globulin were also measured. Because We suggest that immune function or activation may play polyclonal increases in gamma are known to a role in the development of type 2 diabetes. Diabetes accompany infection and inflammatory disease (6,7), we 50:1598–1603, 2001 used gamma globulin levels as a marker of activation of the adaptive immune system, examining cross-sectional and prospective relationships of gamma globulin levels to obesity and type 2 diabetes in the Pima population. Prospective relationships of type 2 diabetes to rheumatoid factor, another immune marker measured routinely, were examined for comparison.

RESEARCH DESIGN AND METHODS The study subjects are participants in the National Institutes of Health survey From the National Institute of Diabetes and Digestive and Kidney Diseases, of diabetes in the Gila River Indian community. Informed consent was National Institutes of Health, Phoenix, Arizona. obtained from all participants, and ethical approval was received from both Address correspondence and reprint requests to Robert Lindsay, MB, PhD, the National Institutes of Health and the Gila River Indian community. All Visiting Associate, National Institute of Diabetes and Digestive and Kidney members of the community older than age 5 years are invited to a biennial Diseases, 1550 East Indian School Rd., Phoenix, AZ 85014. E-mail: rlindsay@ examination, with measurement of a 75-g oral glucose tolerance test (OGTT). mail.nih.gov. Diabetes was diagnosed at these examinations by World Health Organization Received for publication 6 September 2000 and accepted in revised form 29 (WHO) 1985 criteria (8) or, if made between research examinations, the March 2001. ANOVA, analysis of variance; ARIC Study, Atherosclerosis Risk in Commu- clinical diagnosis was reviewed using these criteria. Height and weight were nities Study; IGT, impaired glucose tolerance; NGT, normal glucose tolerance; measured, with subjects wearing light clothing and no shoes, for calculation of OGTT, oral glucose tolerance test; PAI-1, plasminogen activator inhibitor-1; BMI (kg/m2). WHO, World Health Organization. The fraction of American Indian heritage was calculated from self-report of

1598 DIABETES, VOL. 50, JULY 2001 R.S. LINDSAY AND ASSOCIATES

TABLE 1 Baseline characteristics of subjects divided by glucose tolerance at initial examination NGT IGT Type 2 diabetes n 1,861 227 442 Sex (M, F) 902, 959 102, 125 189, 253 Age (years) 21.5 Ϯ 0.4 38.3 Ϯ 1.4* 51.1 Ϯ 0.7*† BMI (kg/m2) 24.0 Ϯ 0.2 30.0 Ϯ 0.5* 30.9 Ϯ 0.3* 2-h glucose (mg/dl) 102 Ϯ 0.4 159 Ϯ 1.0* 352 Ϯ 6.6*† Gamma globulin (g/100 ml) 1.16 (1.159–1.168) 1.23 (1.223–1.233)* 1.21 (1.203–1.212)* Characteristics of 2,530 Pima subjects divided by glucose tolerance at baseline examination. Data are means Ϯ SE, with the exception of gamma globulin (geometric mean and range of Ϯ 1 SE). ANOVA was in keeping with significant differences among groups for all variables (age, BMI, 2-h glucose, gamma globulin; P Ͻ 0.0001) after adjustment for gender. *(P Ͻ 0.05) vs. NGT (Student-Newman-Keuls); †P Ͻ 0.05 for type 2 diabetic group vs. IGT (Student-Newman-Keuls). individuals at interview regarding the heritage of their parents, grandparents, individuals were censored at age 60 years for this analysis. Incident cases and great-grandparents. In this sample, it overwhelmingly derives from either were defined as individuals diagnosed as diabetic at ages within the range of Pima or Tohono O’odham heritage, these being two closely related groups the age group. Person years at risk were counted from baseline examination geographically, culturally, and genetically. Although self-reported, this mea- to either the last examination (for those who did not develop diabetes) or the sure is strongly related to estimation of admixture by testing of a range of date of diagnosis of diabetes, divided into age strata. Standard errors of age- genetic markers (9). In this population, nonϪAmerican Indian heritage and sex-adjusted incidence rates were calculated (12). represents a variety of influences, mainly a mixture of the population with To control for additional predictors of diabetes, the relationship of gamma Mexican-American and European-American groups (9). globulin at baseline to incidence of type 2 diabetes was assessed with Total gamma globulin levels were measured in serum between 1966 and proportional hazards regression in all subjects. In subjects with normal 1973 with the zinc sulfate turbidity method (10). Because concentrations of glucose tolerance (NGT) at baseline, gamma globulin was modeled as a total gamma globulin were skewed, a log-transformed value was used in continuous variable along with other baseline covariates (age, age2, BMI, sex, statistical analysis, as this approximates more closely a normal distribution. and 2-h glucose). The closing date for the proportional hazards model was Results in the main part of this study are limited to samples measured either the date of the last examination (for those who had not developed between September 1966 and September 1969 because of a change in diabetes) or the date of diabetes diagnosis. For comparison, a separate measurement in 1969. Across these 3 years, assay measurement appeared proportional hazards model was also applied to subjects with impaired stable (date of examination accounting for Ͻ1% of the variance in the glucose tolerance (IGT) at baseline (2-h glucose Ն140 mg/dl but Ͻ200 mg/dl) population mean across these years). Measurements taken from September and a model including both groups (IGT and NGT at baseline) was also fitted. 1969 to September 1973 were also stable, but systematically lower, than those The assumption of proportionality inherent in these models was assessed by taken in earlier years. Although unable to pool measurements of gamma inclusion of an interaction term (gamma globulin* log of follow-up time), globulin across all available years, we repeated the main analysis (modeling of which was not significant in any model. To allow comparison, predictor the relationship of baseline gamma globulin to later type 2 diabetes) using variables (apart from age and sex) were standardized to a mean of 0 and SD these later measures (September 1969 to September 1973) to assess results for of 1 using data from all subjects (IGT and NGT at baseline). To assess consistency. Quintiles and tertiles of gamma globulin were calculated within predictors of later IGT, subjects with NGT were assigned a closing date of sex and age in 10-year groups from age 5 years on. Rheumatoid factor was either the first examination diagnostic for IGT, or the date of a last NGT measured by the bentonite flocculation test (11). examination in those remaining NGT. When subjects had progressed to Relationships of gamma globulin to age and BMI were assessed by diabetes without an examination showing IGT, an assumption was made that calculation of partial correlation coefficients. Relationships to sex, glucose IGT had occurred at a point midway between their last NGT examination and tolerance, parental diabetes, and ethnicity were assessed with general linear the date of diagnosis with type 2 diabetes, as previously described (13). models after adjustment for other covariates (BMI and age in assessment of Relationships of gamma globulin to weight gain were also examined. The sex; BMI, age, and sex for other models) in baseline examinations (the first presence of diabetes is associated with diminution of weight gain or weight examination for each subject). Parents were considered 1) nondiabetic if all loss in this population (5). Analysis of weight change over time was therefore recorded OGTTs were nondiabetic (by 1985 WHO criteria), and at least one restricted to subjects who had nondiabetic examinations at baseline and test had been performed when they were over age 30 years, or 2) diabetic if follow-up. Gamma globulin levels, along with other potentially confounding a diagnosis of diabetes had been made at any time using WHO 1985 criteria. baseline covariates (age, BMI, and sex) and length of follow-up were assessed Familial relationships of gamma globulin levels were assessed by calcula- using a general linear model. tion of the intraclass correlation coefficient conditional upon family member- ship (for siblings) or Pearson interclass correlation coefficients (between pairs of parents and the mean of their children or between mothers and fathers). RESULTS The intraclass correlation coefficient was calculated by a maximum likelihood Baseline characteristics and cross-sectional relation- method (assessment of the influence of family membership on gamma globulin when entered as a random effect in a “mixed” model). Analysis was ships with diabetes and IGT. Characteristics of the restricted to families with at least two siblings in the study (2,215 siblings in 2,530 subjects are shown in Table 1. Concentrations of 636 nuclear families). The correlation of gamma globulin between parents and gamma globulin were positively related with age (Pearson their children was examined by Pearson interclass correlation coefficients partial correlation coefficient: r ϭ 0.08; P Ͻ 0.0005; ad- after pairing maternal or paternal values to the mean of their children in 690 ϭ Ͻ families with measurement of gamma globulin in mother and child (1,814 justed for sex and BMI), BMI (r 0.09; P 0.0001; children) and in 417 families with measurement of gamma globulin in father adjusted for sex and age), and female sex (geometric and child (1,260 children). Finally, the Pearson interclass correlation of mean Ϯ 1 SE; males, 1.159 [1.152–1.165] g/100 ml; females, gamma globulin between parents was assessed in 307 mother-father pairs 1.190 [1.187–1.199] g/100 ml; P Ͻ 0.0001; adjusted for age when gamma globulin was available for both parents. In all cases, gamma and BMI) across all subjects. globulin levels were adjusted for age, a quadratic term for age, sex, year of birth, and 2-h glucose during the OGTT before the correlation analysis. Both diabetes and IGT were associated at baseline with Diabetes incidence during follow-up was computed for subjects from the higher gamma globulin concentrations (P Ͻ 0.0001 ad- first visit when gamma globulin had been measured, glucose tolerance was justed for sex) (Table 1) as well as older age and higher normal (2-h glucose Ͻ140 mg/dl), and a measure of BMI was available. BMI. Differences in gamma globulin dependent on glucose Incidence rates (as the number of new cases divided by the person years at risk of diabetes) were calculated for ages 5Ϫ19, and in 10-year age groups for tolerance status appeared to be largely mediated by obe- ages 20Ϫ59, after stratification for sex and tertile of gamma globulin at sity (analysis of variance [ANOVA] for effect of glucose baseline. Because of the small numbers of person years above age 60 years, tolerance on gamma globulin, P ϭ 0.13 with adjustment for

DIABETES, VOL. 50, JULY 2001 1599 GAMMA GLOBULIN LEVELS PREDICT DIABETES

subjects who also had measurement of gamma globulin. Parental diabetes was associated with a significant in- crease in gamma globulin in their offspring (geometric mean [Ϯ1 SE]): neither parent diabetic, 1.126 [1.103–1.149] g/100 ml; one diabetic parent, 1.136 [1.125–1.148] g/100 ml; both parents diabetic, 1.172 [1.162–1.182] g/100 ml; P ϭ 0.03 for between-groups effect) with adjustment for other predictors of gamma globulin (age, sex, and BMI). An estimate of American Indian heritage was available in 2,524 subjects (99.8%, in 1,304 families) of those with baseline gamma globulin measurement. As a continuous variable, fraction of American Indian heritage was posi- tively associated with gamma globulin (P Ͻ 0.004) after adjustment for age, year of birth, BMI, glucose tolerance, and sex. It should be noted that the great majority of subjects were of full American Indian heritage (2,303 FIG. 1. Gamma globulin levels by American Indian heritage. Levels of gamma globulin (geometric mean adjusted for age, BMI, sex, and birth [91%]) (Fig. 1). year ؎ 1 SE on the log scale) are significantly different among groups Prospective relationship of total gamma globulin (P < 0.0001) of subjects divided into those of full American Indian with diabetes. Of the total 2,088 subjects who were not ؍ or 3/4 heritage or less (n ,(55 ؍ heritage (n 8/7 ,(2,303 ؍ heritage (n 166). diabetic at baseline, follow-up information on later glu- cose tolerance was available in 91% of those with NGT and age, sex, and BMI). In a similar fashion, although gamma 85% of those with IGT at baseline (Table 2). Progression to globulin levels had a positive correlation to 2-h glucose diabetes was higher in those with IGT at baseline (66% (r ϭ 0.06; P Ͻ 0.003; n ϭ 2,530), this was not statistically progressed to type 2 diabetes vs. 34% of those with NGT). significant after allowing for other covariates (r ϭ 0.02; In those who had NGT at baseline, later development of P ϭ 0.4; adjusted for sex, BMI, and age). IGT or diabetes was associated with higher initial age, Familial and ethnic relationships of total gamma BMI, and 2-h glucose (Table 2). There was a graded in- globulin. Gamma globulin levels (adjusted for age, sex, crease in gamma globulin levels, with the highest levels in and BMI) were positively correlated among siblings (1,752 those who later developed type 2 diabetes and intermedi- siblings in 528 nuclear families: intraclass coefficient of ate levels in those who later developed IGT (Table 2). The correlation r ϭ 0.23; P Ͻ 0.0001), mothers and their sex-adjusted incidence of type 2 diabetes, stratified by age children (1,391 children in 555 families: r ϭ 0.18; P Ͻ and tertile of gamma globulin, is shown in Fig. 2 for 1,694 0.0001), and fathers and their children (926 children in subjects who were normal glucose tolerant at baseline. 334 families: r ϭ 0.25; P Ͻ 0.0001). Correlation between The association of diabetes incidence with gamma globu- parents was not statistically significant (238 father-mother lin was most marked in ages 40Ϫ49 and 50Ϫ59. Although pairs: r ϭ 0.08; P ϭ 0.2). stratified by age group and sex, the incidence rates do not Information on parental diabetes was available for 870 take account of other potentially confounding variables,

TABLE 2 Baseline characteristics of subjects divided by outcome NGT Developed/remained IGT Developed type 2 diabetes NGT at baseline N 792 334 568 Sex (M, F) 450, 342 136, 198 207, 361 Follow-up (years) 15.5 Ϯ 0.4 14.9 Ϯ 0.5* 16.8 Ϯ 0.4*† Age (years) 18.3 Ϯ 0.6 24.3 Ϯ 1.1* 22.4 Ϯ 0.6* BMI (kg/m2) 21.8 Ϯ 0.2 23.9 Ϯ 0.4* 26.9 Ϯ 0.3*† 2-h glucose (mg/dl) 99 Ϯ 0.6 104 Ϯ 1.1* 106 Ϯ 0.8* Gamma globulin (g/100 ml) 1.14 (1.132–1.146) 1.16 (1.153–1.177)* 1.19 (1.183–1.202)*† IGT at baseline N 39 27 127 Sex (M, F) 23, 16 10, 17 43, 84 Follow-up (years) 11.7 Ϯ 1.6 10.5 Ϯ 1.6 8.7 Ϯ 0.7 Age (years) 38.5 Ϯ 4.3 44.6 Ϯ 4.5† 33.8 Ϯ 1.4 BMI (kg/m2) 27.3 Ϯ 1.0 29.4 Ϯ 1.4* 32.0 Ϯ 0.6* 2-h glucose (mg/dl) 157 Ϯ 1.9 159 Ϯ 3.6 158 Ϯ 1.4 Gamma globulin (g/100 ml) 1.18 (1.149–1.218) 1.15 (1.085–1.214) 1.25 (1.232–1.274) Characteristics of 1,887 Pima subjects who had either NGT (n ϭ 1,694) or IGT (n ϭ 193) at baseline and in whom at least one follow-up examination was available. Data are means Ϯ SE with the exception of gamma globulin (geometric mean [range of Ϯ 1 SE]). In those with IGT at baseline, ANOVA was in keeping with significant differences among groups for age (P Ͻ 0.05) and BMI (P Ͻ 0.0005) after adjustment for sex. In those with NGT at baseline, ANOVA was in keeping with significant differences among groups for age, BMI, 2-h glucose and gamma globulin (P Ͻ 0.0005) and follow-up time (P Ͻ 0.05) after adjustment for sex. *P Ͻ 0.05 vs. NGT (Student-Newman-Keuls); †P Ͻ 0.05 for type 2 diabetic group vs. IGT (Student-Newman-Keuls).

1600 DIABETES, VOL. 50, JULY 2001 R.S. LINDSAY AND ASSOCIATES

gamma globulin ϭ 1.19 [1.1–1.29]; P Ͻ 0.0001) and multi- variate models (hazard rate ratio for 1 SD difference in gamma globulin ϭ 1.09 [1.01–1.18]; P ϭ 0.04; model including sex, age, BMI, and 2-h glucose). Gamma globulin was again associated with later development of IGT in univariate (768 incident cases, hazard rate ratio for 1 SD difference in gamma globulin ϭ 1.15 [1.1–1.29]; P Ͻ 0.0001), but not multivariate analysis. Prospective relationship of total gamma globulin with later weight gain. A total of 1,616 subjects were nondiabetic at baseline and one follow-up examination (mean Ϯ SD follow-up: 16.1 Ϯ 9.6 years). Baseline gamma globulin was not significantly associated with BMI at later examination (P ϭ 0.3) after adjustment for other predic- tors (age, a quadratic term for age, BMI at baseline, sex, length of follow-up), all of which were highly significant Ͻ FIG. 2. Sex-adjusted incidence of diabetes per 1,000 person years. Data predictors (P 0.0001). Similar analysis was also carried are presented as rate ؎ SE stratified by age group (on the x axis) and out in groups stratified by age (in 10-year age groups from tertile of gamma globulin. Total person years of follow-up: ages 5؊19 age 5 years) at baseline, with no effect of gamma globulin ,years, 11,729 years; ages 20؊29 years, 10,720 years; ages 30؊39 years .years; ages 40؊49 years, 3,303 years; and ages 50؊59 years, 1,636 being observed in any age group 7,308 years. DISCUSSION such as glucose concentration at baseline, age within age There has been recent interest in interactions between groups, and BMI. The prospective relationship of gamma factors in the innate immune system and metabolic and globulin to later development of type 2 diabetes was vascular disease (1). In this study, we showed that total therefore further analyzed by proportional hazards regres- gamma globulin concentration, a nonspecific measure of sion. the adaptive immune system, is influenced in the Pima In univariate analysis, age, BMI, 2-h glucose, and gamma population by both familial and ethnic factors and is globulin were significant predictors of diabetes in those associated cross-sectionally with obesity. Furthermore, with NGT at baseline (Table 3). After adjustment for other higher levels of gamma globulin are associated with higher covariates, gamma globulin remained a significant predic- incidence of type 2 diabetes prospectively. tor of later type 2 diabetes, witha1SDdifference in log Gamma globulin was measured by the zinc sulfate gamma globulin being associated with a 14% difference in turbidometric technique as reported by Kunkel in 1947 hazard rate of diabetes (Table 3). (10) and acts as a nonspecific measure of activity of the Gamma globulin also predicted progression of NGT to humoral immune system. Concentrations of gamma glob- IGT in univariate analysis (902 incident cases; hazard rate ulin by this method agree with quantification by electro- ratio 1.15; 95% CI 1.07–1.23; P Ͻ 0.0001), but in contrast to phoresis (14) but have been almost entirely replaced in results for prediction of type 2 diabetes, this was not clinical practice with newer methods allowing quantifica- significant in the multivariate model (hazard rate ratio tion of immunoglobulin subfractions. Kunkel’s method 1.05; 95% CI 0.98–1.13; P ϭ 0.14). was developed mainly for use in the assessment of hepa- In those who had IGT at baseline, although the highest tocellular disease (10), but is also sensitive to nonspecific gamma globulin concentrations were again found in those who progressed to diabetes, differences among groups dependent on progression were not statistically significant TABLE 3 Predictors of type 2 diabetes (Table 2). Gamma globulin did not appear to predict progression of IGT to diabetes. In the 193 subjects with Variable Hazard rate ratio P IGT at baseline, gamma globulin was not a significant pre- dictor in either univariate (hazard rate ratio 1.04; 95% CI Univariate models Female gender 1.18 (1.00–1.40) 0.055 0.90–1.20; P ϭ 0.6) or multivariate models (0.98; 0.85–1.14; Ͻ ϭ Age 1.35 (1.28–1.42) 0.0001 P 0.8; baseline age, sex, BMI, and 2-h glucose included BMI 1.96 (1.81–2.11) Ͻ0.0001 as covariates). 2-h glucose 1.58 (1.40–1.78) Ͻ0.0001 Measurement of rheumatoid factor was available in Gamma globulin 1.20 (1.11–1.30) Ͻ0.0001 5,967 nondiabetic individuals along with follow-up exam- Multivariate model ination for diabetes in the population study (1,325 incident Female gender 0.95 (0.80–1.13) 0.6 cases). Rheumatoid factor did not predict development of BMI 1.73 (1.50–1.87) Ͻ0.0001 type 2 diabetes (data not shown). 2-h glucose 1.37 (1.22–1.54) Ͻ0.0001 The consistency of gamma globulin in predicting later Gamma globulin 1.14 (1.05–1.24) 0.0022 IGT or type 2 diabetes was also assessed by separate Data are means (95% CI). Predictors of diabetes in 1,694, subjects analysis of the relationship of measurements of gamma who were normal glucose tolerant at baseline (568 incident cases of globulin made between 1969 and 1973 to later type 2 diabetes). To allow comparison, all predictor variables were stan- dardized to a mean of 0 and SD of 1, apart from gender (male ϭ 0 and diabetes. In subjects who had NGT at baseline, gamma female ϭ 1) and age (1 unit increase ϭ 10 years). The multivariate globulin again predicted type 2 diabetes in both univariate model also contained significant (P Ͻ 0.01) linear and quadratic (475 incident cases; hazard rate ratio for 1 SD difference in terms for age.

DIABETES, VOL. 50, JULY 2001 1601 GAMMA GLOBULIN LEVELS PREDICT DIABETES rises in gamma globulin, as observed in a variety of tory mediators independent of relationships with adiposity infectious diseases (6,7). (21). Insulin resistance also predicts later type 2 diabetes In our analysis, higher gamma globulin levels were in the Pima population (22). If gamma globulin concentra- related to higher BMI and were also higher in those who tions were associated with insulin resistance, as appears had IGT or diabetes, although this relationship was largely to be the case for C-reactive protein (21), this might ex- explained by effects of covariates such as age and BMI. plain the prospective relationship of gamma globulin to Immunoglobulin concentrations (of IgA, IgG, and IgM type 2 diabetes that we observed. To our knowledge, such classes) have previously been reported to be higher in a relationship of gamma globulin to insulin resistance has those with diabetes (although these data were not ad- not been assessed. Against this interpretation, the prospec- justed for other covariates) (15). Interpretation of these tive relationship of some inflammatory mediators appears cross-sectional observations is difficult, as they may be to remain significant after adjustment for baseline insulin confounded by secondary effects; for example, diabetes (3), and white cell count shows only weak association with may both increase the likelihood of infection and alter the measures of insulin resistance in the Pima population (19). immune response. There were too few subjects in the Why should gamma globulin concentration predict type present study with repeated measures of gamma globulin 2 diabetes? Apart from the influences of ethnicity, famil- before and after changes in glucose tolerance to explore iality, age, sex, and BMI that we have described, we do not this issue. know what underpins variations in total gamma globulin in Gamma globulin concentrations were familial, being our population and, therefore, potentially the association significantly correlated among siblings and between par- with later type 2 diabetes. If this variation reflects the ents and their children but not between parents. Relatives presence of infectious or inflammatory disease, or indeed share genetic determinants of the immune response but activation of the innate immune system by these or other also environmental factors—importantly, exposure to in- factors (23), then this would support a possible role for fection. Clearly it is not possible to determine with this these agents in influencing the development of type 2 analysis whether the familial associations and associations diabetes. The place of infectious agents in the etiology of with American Indian ethnicity we observed reflect the type 2 diabetes is currently speculative, but evidence does results of shared genes or shared environment. The lack of exist that exposure to increases risk of type 2 correlation between parents, however, suggests that the diabetes (24). By contrast with our results for total gamma correlations among relatives are attributable to either globulin, rheumatoid factor does not appear to predict genetic factors or environmental influences acting early in later type 2 diabetes in this population, making it unlikely life. Evidence regarding differences in the immune system that the relationship we observed is explained by associ- between American Indian populations and other ethnic ations with diseases that increase rheumatoid factor, such groups are limited. One previous study has suggested as rheumatoid arthritis. It is also possible that gamma ethnic differences in gamma globulin levels, with higher globulin concentration is a marker of underlying gene- concentrations of IgG, IgA, and IgM in American Indian tic differences in the reactivity of the immune system. than Caucasian, U.S. military veterans (16). By contrast, Fernandez-Real and Ricart (25) suggested that genetic levels of PAI-1 were not higher in Pimas than in either variation in cytokine responses may underpin the predic- Caucasians or South Asians (17). tive relationship of inflammatory mediators and metabolic Our results add to the observations that factors in the disease. The ability of gamma globulin to predict later immune system may be associated with later metabolic diabetes in this report may therefore reflect underlying disease. A variety of measures of inflammation predicted genetic differences in the population. Importantly, the later type 2 diabetes in the Atherosclerosis Risk in Com- observations that gamma globulin levels are familial and munities (ARIC) Study (3), including raised fibrinogen, related to both American Indian heritage and parental white cell count, sialic acid, and , as well as diabetes is consistent with this interpretation, although lower serum albumin. Such inflammatory markers are the possibility that these relationships might be under- known to have positive cross-sectional associations with pinned by environmental rather than genetic influences BMI (3), and recently have been shown also to predict cannot be excluded. weight gain in the ARIC Study (18). It is important, then, to One weakness of this report is that we cannot determine recognize that the prospective effects observed may sim- which of several potential underlying mechanisms might ply reflect associations with baseline adiposity. However, underpin the association of gamma globulin with type 2 in the ARIC study, at least some of the inflammatory diabetes. It is also important to note that the predictive measures (white cell count, sialic acid, and orosomucoid) value of gamma globulin was confined to those with NGT. remained significant after adjustment for baseline BMI (3). In subjects with IGT at baseline, although gamma globulin We have shown that BMI is positively related to gamma levels were higher on average, and highest in the group globulin in our study. In previous studies in the Pima who progressed to later type 2 diabetes, gamma globulin population, white cell count was also positively associated was not significantly predictive of progression from IGT to with obesity (19), with the relationship proposed, at least diabetes. It is not possible to discern whether this re- in part, to be mediated by leptin (20). Nevertheless, the flected a difference in underlying biology or a failure to prospective relationship of gamma globulin with later type detect an effect of gamma globulin because of the smaller 2 diabetes that we observed appears to be over and above number of subjects with IGT at baseline. effects of adiposity. In a similar fashion, insulin resistance Our data add to the body of evidence of relationships might be a potential explanation and confounder of our among immune mediators, obesity, and type 2 diabetes. findings. Insulin resistance is associated with inflamma- This is of particular interest in the context of the high

1602 DIABETES, VOL. 50, JULY 2001 R.S. LINDSAY AND ASSOCIATES incidence of type 2 diabetes in American Indian popula- 7. Zitrin CM, Lincoln EM, Saifer A, Lewkowicz S: Serum gamma globulin in tions. Fernandez-Real and Ricart proposed that the rela- childhood tuberculosis. American Review of Tuberculosis and Pulmo- nary Diseases 74:15–28, 1956 tionship of inflammatory mediators and metabolic risk 8. World Health Organization: Diabetes Mellitus: Report of a WHO Study may be important in understanding the evolutionary con- Group. Geneva, World Health Organization, 1985 (Tech. Rep. Ser., no. 727) text of why certain individuals are at risk for type 2 9. Williams RC, Knowler WC, Pettitt DJ, Long JC, Rokala DA, Polesky HF, diabetes (25). In particular, they hypothesized that an Hackenberg RA, Steinberg AG, Bennett PH: The magnitude and origin of insulin-resistant genotype associated with a high cytokine European-American admixture in the Gila River Indian Community of response might have been advantageous in historical Arizona: a union of genetics and demography. Am J Hum Genet 51:101– 110, 1992 conditions of short life span, injury, and infectious disease, 10. Kunkel HG: Estimation of alteration of serum gamma globulin by a but disadvantageous today (25). They did not directly turbidimetric technique. Proc Soc Exp Biol Med 32:483–493, 1947 address why ethnic groups such as American Indians are 11. Bozevich J, Bunim J, Freund J, Ward SB: Bentonite flocculation test for so prone to obesity and type 2 diabetes compared with rheumatoid arthritis. Proc Soc Exp Biol Med 97:180–183, 1958 European populations. It is notable that, as well as being 12. Knowler WC, Bennett PH, Hamman RF, Miller M: Diabetes incidence and prevalence in Pima Indians: a 19-fold greater incidence than in Rochester, highly susceptibility to type 2 diabetes, American Indians Minnesota. Am J Epidemiol 108:497–505, 1978 have a dramatic history of recent epidemics of infectious 13. Saad MF, Knowler WC, Pettitt DJ, Nelson RG, Charles MA, Bennett PH: A disease. After first contact with Europeans in the late 15th two-step model for development of non-insulin-dependent diabetes. Am J century, American Indians were exposed to a range of Med 90:229–235, 1991 novel infectious diseases to which they had no prior 14. Wilson TE, Brown CH, Hainline A: The zinc sulphate turbidity test. Gastroenterology 483–493, 1956 , leading to repeated epidemics (26,27) and 15. Ardawi MS, Nasrat HA, Bahnassy AA: Serum immunoglobulin concentra- marked declines in population (27,28). It is tempting to tions in diabetic patients. Diabet Med 11:384–387, 1994 speculate that these repeated epidemics may have led to 16. Mili F, Flanders WD, Boring JR, Annest JL, Destefano F: The associations selection of individuals both resistant to infectious disease of race, cigarette smoking, and smoking cessation to measures of the but also highly prone to diabetes. immune system in middle-aged men. Clin Immunol Immunopathol 59: 187–200, 1991 In conclusion, total gamma globulin concentrations pre- 17. Nagi DK, Knowler WC, Hanson RL, Mohamed-Ali VM, Yudkin JS: - dict development of type 2 diabetes in the Pima popula- ogen activator inhibitor (PAI-1) and non-insulin-dependent diabetes in tion. These observations support a number of recent Pima Indians, South Asians and Europeans: populations at varying risk of observations suggesting a role of inflammation or infection NIDDM and coronary artery disease. Thromb Haemost 75:921–927, 1996 in the pathogenesis of type 2 diabetes. 18. Duncan BB, Schmidt MI, Chambless L, Folsom AR, Carpenter M, Heiss G: Fibrinogen, other putative markers of inflammation, and weight gain in middle aged adults: the ARIC Study. Obes Res 8:279–286, 2000 ACKNOWLEDGMENTS 19. Pratley RE, Wilson C, Bogardus C: Relation of the white blood cell count We thank the members of the Gila River Indian Commu- to obesity and insulin resistance: effect of race and sex. Obes Res nity for their continued support and participation and the 3:563–571, 1995 staff of the Diabetes and Arthritis Epidemiology Section 20. Wilson CA, Bekele G, Nicolson M, Ravussin E, Pratley RE: Relationship of the white blood cell count to body fat: role of leptin. Br J Haematol for conducting this study. 99:447–451, 1997 21. Festa A, D’Agostino R Jr, Howard G, Mykkanen L, Tracy RP, Haffner SM: REFERENCES Chronic subclinical inflammation as part of the insulin resistance syn- 1. Pickup JC, Crook MA: Is type II diabetes mellitus a disease of the innate drome: the insulin resistance atherosclerosis study (IRAS). Circulation immune system? Diabetologia 41:1241–1248, 1998 102:42–47, 2000 2. Danesh J, Whincup P, Walker M, Lennon L, Thomson A, Appleby P, 22. Lillioja S, Mott DM, Spraul M, Ferraro R, Foley JE, Ravussin E, Knowler Gallimore JR, Pepys MB: Low grade inflammation and coronary heart WC, Bennett PH, Bogardus C: Insulin resistance and insulin secretory disease: prospective study and updated meta-analyses. BMJ 321:199–204, dysfunction as precursors of non-insulin-dependent diabetes mellitus: 2000 prospective studies of Pima Indians. N Engl J Med 329:1988–1992, 1993 3. Schmidt MI, Duncan BB, Sharrett AR, Lindberg G, Savage PJ, Offenbacher 23. Du Clos TW: Function of C-reactive protein. Ann Med 32:274–278, 2000 S, Azambuja MI, Tracy RP, Heiss G: Markers of inflammation and predic- 24. Mason AL, Lau JY, Hoang N, Qian K, Alexander GJ, Xu L, Guo L, Jacob S, tion of diabetes mellitus in adults (Atherosclerosis Risk in Communities Regenstein FG, Zimmerman R, Everhart JE, Wasserfall C, Maclaren NK, study): a cohort study. Lancet 353:1649–1652, 1999 Perrillo RP: Association of diabetes mellitus and chronic hepatitis C virus 4. Festa A, D’Agostino RB, Tracy RP, Haffner SM: Elevated levels of acute infection. Hepatology 29:328–333, 1999 phase proteins and plasminogen activator inhibitor-1 (PAI-1) predict the 25. Fernandez-Real JM, Ricart W: Insulin resistance and inflammation in an development of type 2 diabetes mellitus: the Insulin Resistance Athero- evolutionary perspective: the contribution of cytokine genotype/pheno- sclerosis Study (IRAS). (Abstract) Diabetes 49 (Suppl. 1):A24, 2000 type to thriftiness. Diabetologia 42:1367–1374, 1999 5. Knowler WC, Pettitt DJ, Saad MF, Charles MA, Nelson RG, Howard BV, 26. McNeill W: Plagues and Peoples. New York, Doubleday, 1976 Bogardus C, Bennett PH: Obesity in the Pima Indians: its magnitude and 27. Dobyns H: Their Numbers Became Thinned. Knoxville, Tennessee, Uni- relationship with diabetes. Am J Clin Nutr 53:1543S–1551S, 1991 versity of Tennessee Press, 1983 6. Maher FT, Snell AM, Mann FD: Turbidimetric estimation of the serum 28. Shoemaker N: An overview of American Indian demographic history. In colloids in the differential diagnosis of hepatobiliary disease. Gastroenter- American Indian Population Recovery in the Twentieth Century. Albu- ology 1949 querque, New Mexico, University of New Mexico Press, 1999, p. 1–13

DIABETES, VOL. 50, JULY 2001 1603