Heart Disease Mortality in the Life Span Study, 1950–2008

In published epidemiological studies, the risk of cardiovascular disease (CVD) has been reported after exposure to ionizing radiation from radiotherapy (1–6 ), occupational and environmental settings (4, 6–14 ) and atomic bombs (15 ). While a published finding from the Life Span Study (LSS) of atomic bomb survivors indicated that mortality from CVD increased even six decades after whole-body irradiation (0.5–1.0 Gy) (15 ), several issues have yet to be elucidated. One issue of concern is that the elevated risks were observed in limited subtypes of CVD, such as rheumatic heart disease, hypertensive heart disease and heart failure. Another issue is that the causal inference of radiation exposure and increased CVD mortality is complicated because of a number of unavoidable factors, which may affect the evaluation of radiation risks, particularly temporal changes in background risks. For example, incidence and mortality of coronary heart disease increased due to the westernization of lifestyles, including a high-fat diet and a sedentary working pattern associated with socioeconomic development beginning in the 1960s (16–20 ). Trends in coronary heart disease mortality and its risk factors have not remained constant over the follow-up period and have varied for the different birth cohorts. Further complicating the analyses are revisions that were made in 1995 to the International Classification of Diseases (ICD) (21, 22 ) may have substantially modify the Japanese death statistics. Misclassification was particularly evident between 1994–1995 after the ICD revision, where a marked increase (>30%) in mortality rates due to ischemic heart disease and cerebral infarction were coupled with a decrease (<50%) of mortality due to heart failure in Japan (21, 22 ). In the ICD-10 version, it was recommended that physicians not use “heart failure” as an underlying cause of death (23 ). Any of these changes may have induced uncontrolled confounders. While it is generally accepted that radiation risks are associated with high doses, a clear mechanism to explain this association remains to be found. Therefore, we explored radiation-associated CVD risks at lower dose ranges through subtype- and period-specific analyses.

The study population consisted of 86,600 subjects with estimated individual radiation doses in the LSS cohort, which includes a large portion of the atomic bomb survivors who were located within 2.5 km from the hypocenter at the time of the bombings of Hiroshima and Nagasaki, together with a similar-sized age- and sex-matched sample of people located between 2.5–10 km from the hypocenters. The subjects included both men and women exposed at all ages (mean = 21.9 years, range = 0–89) at the time of the bombings. Individual radiation dose was calculated using the DS02 dosimetry system, which incorporates a person's location and shielding at the time of the bombings. We estimated the risk of exposure using weighted absorbed colon dose (Gy) by summing the gamma dose plus 10× the neutron dose, to allow for the greater biological effectiveness of neutrons. As Radiation Effects Research Foundation (RERF) risk estimates do not depend critically on the relative biological effectiveness (RBE) and neutron doses are small relative to gamma-ray doses, a constant neutron RBE value of 10 was applied in this study (24 ). To adjust for the impact of bias arising as a result of random errors in individual dose estimates, shielded kerma estimates above 4 Gy were truncated to 4 Gy; colon doses were truncated by multiplying by the kerma reduction factor (i.e., 4/nontruncated shielded kerma) and were then adjusted according to the method of Pierce et al. Details of the study population and dose estimation are described elsewhere (15, 25–27 ).

Mortality follow-up took place from October 1, 1950 to the end of 2008 and was based on the Japanese family registry system (koseki), which essentially covers all of Japan (more than 99% complete). The underlying cause of death was abstracted from the death certificate. Cause of death was classified into one of the four subtypes according to the pathophysiological basis: 1. ischemic heart disease (IHD) including myocardial infarction (MI); 2. valvular heart disease (VHD); 3. hypertensive organ damage (HOD); and 4. heart failure (HF) (Appendix Table A1). These disease categories were defined to be as similar as possible considering ICD revisions across periods. Valvular heart disease consists of rheumatic and nonrheumatic valvular disease. In addition, we conducted detailed analyses based on aortic and mitral valve manifestations in nonrheumatic disease. Hypertensive organ damage included hypertensive damage to the heart or kidney because the circulatory system is maintained by interactions between the two organs (28 ). Although the category of HOD includes kidney damage, the effect of radiation on hypertensive vascular changes can be assessed, heart disease overall is defined as disorders of the heart that were based on various pathogeneses. We, therefore, did not include kidney damage in the heart disease overall analysis.

The analysis was focused primarily on four time periods: 1950–1968, 1969–1980, 1981–1994 and 1995–2008. We also included a sub-analysis using the latest sub-period, 2001–2008, to represent the most recent trends of heart disease deaths. The four primary periods coincide with revisions of the classification of underlying cause of death used by RERF as well as in Japan. The primary periods also coincide with trends of CVD mortality. The period of 1950–1968 corresponds to ICD versions 6 and 7. After the introduction of ICD-8 in the RERF database in 1968, the number of MI deaths sharply increased. Such sudden changes in death counts were also observed as a temporal decline in deaths from heart disease and a concomitant increase in deaths from stroke in Japan just after the introduction of ICD-10 in 1995, at which time it was recommended that physicians not use the diagnosis of heart failure for the underlying cause of death. The cutoff point of 1980 was adopted because advanced techniques for diagnosis of heart disease (e.g., coronary artery angiography, echocardiography) became available in the 1980s and saw increasing use through the 1990s.

Statistical methods were the same as those used in previously reported work (15, 27, 29 ). The primary data for analyses consist of a table listing the numbers of deaths and person-years classified by city, gender, weighted absorbed colon dose (22 categories with cutoff points of 0.005, 0.02, 0.04, 0.06, 0.08, 0.1, 0.125, 0.15, 0.175, 0.2, 0.25, 0.3, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0, 2.5 and 3.0 Gy), observation period (14 categories with cutoff points of January 1 for 1956, 1961, 1966, 1969, 1971, 1976, 1981, 1986, 1991, 1995, 1996, 2001 and 2006), attained age (19 categories of five-year intervals from age 5–95 and an additional category of age ≥100) and age at exposure (14 categories of five-year intervals from age 0–65 and that of ≥70). Poisson regression models for grouped survival data were used to describe the mortality rates and radiation effects on them.

The primary statistical model is a Poisson rate regression model with radiation risk modeled as the excess relative risk (ERR):

where λ₀ (•) is the background mortality rate (expected number of deaths divided by person-years), which depends on city (c), sex (s), attained age (a), birth cohort (b) and time period (p). ERR (d, x) describes the relative change in the rate at dose d with effect modification by x. The background rate λ₀ (•) was modeled with stratification with c, s, a, b and p to avoid imposing modeling constraints on the effects of these factors. Potential effect modifiers (x) included c, s, a, age at exposure (e) and p. The ERR function was fit using parametric functions of the form ρ(d)·ɛ(x), in which ρ(d) describes the shape of the dose-response function and ɛ(x) describes risk variation with effect modifiers. Effects of effect modifiers were examined using a log-linear model, ɛ(x) = e^δx. In addition to the linear dose response [ρ(d) = βd], we also considered alternative dose-response models including quadratic [ρ(d) = γd²] and linear-quadratic [ρ(d) = βd + γd²] functions and a linear-threshold model [ρ(d) = β(d – d₀) for d > d₀ ; ρ(d) = 0 for d ≦ d₀], but effect modification was only examined in the linear dose-response model. A model of dose response including threshold analysis covers full dose range, but there is uncertainty in risk of radiation at low-dose levels. To ascertain the lowest dose range for which a statistically significant radiation effect could be detected, we fit a segmented linear model to the full data with different ERRs below and above a particular dose (30, 31 ) [ρ(d) = (β₁d_l + β₂d_h) exp(δx), where, for various values of the particular dose d_p, d_l = d if d < d_p and d_l = 0 otherwise; d_h = d if d ≧ d_p and d_h = 0 otherwise]. The lowest dose range with a statistically significant ERR dose response was estimated by testing the null hypothesis that the low-dose slope (β_l) was zero by stepping up the cut point by 0.1 Gy increments.

The significance of risk estimates depends on the number of excess cases reflecting both magnitudes of risk and baseline rates. Temporal trends in crude mortality rates among the LSS subjects with dose <5 mGy were evaluated for adequate interpretation of possible difference in radiation dose-response across the different periods, assuming those subjects (<5 mGy) have little radiation effect (Appendix Fig. A1). Standardized mortality ratios (SMRs) of subtypes were calculated for comparison with Japanese population by period, referring to sex- and age-specific mortality rates in Japan (Appendix Fig. A2). Also, trend of the age-specific mortality rate of heart disease was shown (Appendix Fig. A3).

Epicure software was used for parameter estimation and testing (32 ). Significance tests and confidence intervals (CIs) were based on likelihood statistics. All P values are two-sided and P < 0.05 was considered to be statistically significant.

Numbers of subjects, observed person-years and the number of deaths from subtype-specific heart disease by weighted absorbed colon dose are shown in Table 1. Of the 86,600 subjects, a total of 55,654 persons died during the follow-up, including 9,303 deaths from heart disease overall, among which IHD and HF were of significant proportions (38% and 36%, respectively). Based on a linear model over the full-dose range, the ERR/Gy for heart disease overall was [0.14 (95% CI: 0.06 to 0.22)] (Fig. 1, Table 2). Subtype-specific analyses showed a statistically significant increased risk of death from VHD [ERR/Gy = 0.45 (0.13, 0.85)], HOD [ERR/Gy = 0.36 (0.10, 0.68)] and HF [ERR/Gy = 0.21 (0.07, 0.37)], but not for IHD. When the VHD category was further divided, analysis showed a highly increased risk for rheumatic VHD [ERR/Gy = 0.96 (0.28, 1.92)], whereas the risk of nonrheumatic VHD did not increase during the entire period. Among 1,122 deaths from HOD, including heart/kidney damages, risk of hypertensive damage limited to heart increased significantly [879 deaths, ERR/Gy = 0.37 (0.07, 0.73)], while ERR was not significant for hypertensive damage limited to kidney [174 deaths, ERR/Gy = 0.39 (−0.15, 1.30)] or hypertensive damage to both [69 deaths, ERR/Gy = 0.26 (−0.79, 1.31)], although these ERRs were estimated relative to similar levels.

We evaluated the dose response for each subtype based on linear, quadratic and linear-quadratic (LQ) models. The best fitting model for those showing statistically significant risk (heart disease overall; VHD, HOD and HF) was the linear model by the Akaike information criteria (AIC). Although the quadratic model fit the best, no model was statistically significant for IHD. Figure 2 shows the dose-response curve based on the linear model as well as the quadratic model for IHD. The best estimates of threshold dose (with upper 95% CIs) was 0 Gy (0.5 Gy) for heart disease overall; 0 Gy (0.5 Gy) for VHD, 0 Gy (0.4 Gy) for HF and 0.2 Gy (0.6 Gy) for HOD, respectively. Although there was no evidence for a threshold in a model function, the lowest dose d_p for which there was a statistically significant dose response when analysis was restricted to the range (0, d_p) was 0.7 Gy [ERR (95% CI) = 0.25 (0.08, 0.43)] for heart disease overall; 0.7 Gy [ERR = 0.80 (0.14, 1.60)] for VHD, 1.5 Gy [ERR = 0.34 (0.02, 0.72)] for HOD and 0.4 Gy [ERR = 0.44 (0.02 to 0.91)] for HF, respectively.

Table 2 and Fig. 3 show the risk estimates for heart disease subtypes by a period. A significant increase in ERR for heart disease overall was observed only in the second period (1969–1980), and no trends in the ERRs of each divided period were evident. ERR/Gy of IHD was not significantly different from zero in any single period or over the entire period. Neither MI nor other IHD showed an association with radiation dose. Although there was no significant upward trend, ERR for MI showed an increase in the last several years of the study period, 2001–2008 [ERR/Gy = 0.5 (0.00, 1.24), P = 0.05]. The high ERR may reflect the highest risk of those who were exposed at age of 20–29 years: ERR/Gy = 0.87 for 20–29 years, 0.02 for <20 years and 0.09 for ≥30 years, respectively, without any significant differences in risks among them.

ERR/Gy of total VHD increased with calendar period (trend P = 0.08) and was significantly greater than zero in the third (1981–1994) and fourth (1995–2008) periods (Table 2). The increasing trend was attributed to the trend of nonrheumatic VHD (trend P = 0.03). On the other hand, ERR/Gy of rheumatic VHD was significantly and remarkably high (1.34) in the first period (1950–1968) but not significant in the subsequent three periods despite relatively high point estimates (0.65–0.82). However, over the study period, heterogeneity of the ERRs for rheumatic VHD was not statistically significant. From the first to fourth periods, involved valves (i.e., mitral and aortic) were increasingly elucidated (28, 43, 76 and 92%, respectively), possibly attributed to the development of echocardiography as a valuable tool for identifying patients with regurgitant/sclerotic valves. The highest ERR/Gy for nonrheumatic aortic valve disease (0.95, not significant) was implied in the applicable periods, third and fourth combined a period of 1981–2008.

Although ERR/Gy of HOD was significantly high for the entire period without evidence of heterogeneity or trend, the risk of divided periods was evident during 1950–1968 (0.59) and unobservable thereafter (Fig. 3D, Table 2). ERR/Gy of HF was also significantly elevated for the overall period, particularly for the middle periods (1969–1980, 1981–1994), but heterogeneity or trend over the calendar periods was not evident (Fig. 3E, Table 2).

Table 3 shows ERR/Gy of heart disease subtypes by levels of potential effect modifiers and the results of homogeneity tests. There were no statistically significant heterogeneities in ERR/Gy by the various modifiers (city, sex, age at exposure and attained age) although differences in some categories approached significance. Women tended to have higher risks than men for heart disease overall, HOD, HF and VHD (ERRs for women/men were, respectively: 0.21/0.06; 0.55/0.11; 0.30/0.11; and 0.64/0.06). For HOD and HF, ERR/Gy was significantly high for those exposed at age >40 years. While applying the different cutoff point as for age at exposure and attained age, there was no significant variation across them. Variation across the attained age (<60, 60–79, ≥80) was marginally significant for VHD and HOD. For HOD the difference in ERRs between <60 (1.92) and ≥60 (0.28) was statistically significant (P = 0.04).

This study demonstrates risks of exposure to atomic bomb radiation for death due to certain subtypes of heart disease and presents corresponding dose responses in the Life Span Study of atomic bomb survivors by sub-periods between 1950 and 2008. Radiation dose of atomic bomb survivors was generally much lower than those in other studies that demonstrated increased risks of vascular diseases due to ionizing radiation exposure. Consistent with a previously published analysis (15 ), the current analysis shows significantly increased risks over the entire period for heart disease overall, as well as for VHD, HOD and HF, but not for IHD. All significant dose-response relationships were best described by a linear model with ERR/Gy ranging from 0.21 for HF to 0.45 for VHD. As to uncertainty of radiation risk at a low dose level, the lowest dose range with a statistically significant dose response was 0–0.7 Gy for heart disease overall and VHD, 0–1.5 Gy for HOD and 0–0.4 Gy for HF.

Breaking down the entire period into four sub-periods in accordance with changes in the ICD versions, development of medical equipment and pathogenesis of CVD in Japan (33 ), radiation-related excess risks presented different patterns for different subtypes (Table 2). Notable findings were an upward trend in ERR for VHD over the study period (trend P, 0.08), in particular for nonrheumatic VHD (trend test P, 0.03). Although a significant trend was not evident, the risk of MI, a subcategory of IHD, also seemed to appear in the sub-analysis of the most recent period (2001–08). In contrast, increased risks of rheumatic VHD and HOD were limited to the early period (1950–1968), and increased risk of HF was limited to the middle period (1981–1994).

To adequately interpret potential variation in radiation dose response by period, time trend in baseline rates also need to be assessed. In Japan, age-specific mortality rates of heart disease overall for the elderly (>80) remarkably increased in the first period of this study, then plateaued during the second and third periods and rapidly decreased just before the fourth period (Appendix Fig. A3). The decrease was thought to be attributed to the revisions of ICD-9 to ICD-10 and the recommendation not to diagnose “heart failure” for the underlying cause of death in 1995. Crude mortality rates of heart disease overall increased with calendar period, except for the latest decline in men (Appendix Fig. A1). Although the patterns are different, upward changes in crude rates for each subtype over the study period as well as growth in baseline rates may particularly reflect the concentration of elderly in the LSS, a fixed cohort. The SMRs of heart disease were also examined to compare the LSS subjects (<5 mGy) with Japanese population as a reference population (Appendix Fig. A2). In the LSS groups, higher rates were estimated for IHD in 1981–1994 and HF in 1995–2008, and lower rates for IHD in 1995–2008, respectively, although the reasons for this are unclear. Except for these, few differences between the LSS group and Japanese population were found in subtype-specific SMRs. From these data, the baseline rates and their periodic trends in the LSS are equal to or greater than that of the Japanese population.

Ischemic heart disease, especially MI, is a major long-term risk associated with radiotherapy (10–40 Gy) (34 ). High-dose radiation therapy is well known to evoke pathological changes in the circulatory system, such as damage to arterial walls (6, 36 ) or fibrotic changes and small/coronary vascular damage (35 ). In contrast, for radiation exposure at low- to middle-dose level (<1 Gy), previously reported epidemiological studies from occupational and environmental settings and atomic bombs have shown inconsistent results (4, 6–13, 15 ). Experimental findings at low-dose levels (<0.5 Gy) are still controversial: a few studies have indicated that radiation exposure is associated with suppressive, rather than accelerating, effects on atherosclerosis (37, 38 ).

The current investigation did not reveal an increased risk of IHD, even in a subcategory of MI or other IHD over 1950–2008 except for an increase in our preliminary analyses for 2001–2008. Increase in MI risk since 2001 may reflect high ERRs among the subjects who were exposed at relatively young age (20–29 years). A previous report from the Adult Health Study (AHS), a clinical subcohort of the LSS may be partially in agreement with our current finding in the LSS: that AHS report did not indicate any increased risk of incidence of CVD subtype in a linear dose-response model, but showed significant quadratic dose-response for nonfatal MI in members exposed at age <40 years (relative risk at 1 Gy = 1.25) (44 ). Other considerations need to be addressed for adequate interpretation. In the LSS and AHS, different outcomes were observed, which are not easily compared. Additionally, some published studies showed the improved reliability of MI diagnosis at death with improved coding in the ICD-10 after 1995 (39 ) and the enhanced diagnostic tools, such as cardiac troponin tests (40–43 ). In contrast, other studies revealed low validity of the MI diagnosis; sensitivity of diagnosis of IHD on death certificates was 86.5% (95% CI: 77.6, 92.3), specificity was 64.9% (58.4, 71.0), positive predictive value was 50.6% (42.7, 58.5) and negative predictive value was 92.0% (86.5, 95.5) when all cases were confirmed according to the criteria of the World Health Organization MONICA Project (Multinational MONItoring of trends and determinants in CArdiovascular disease) during the period of 1997–1998 (45, 46 ). Cause-unspecified acute deaths tended to be listed as MI instead of HF on death certificates (21, 22 ), which may have resulted in that low validity of MI diagnosis. A final consideration is that the development of interventional therapy since the end of the 1990s may also have some impact on the improvement of the case fatality of MI. From these considerations, careful follow-up of MI deaths is essential for adequate interpretation.

A previously published LSS study about cardiovascular death showed a significant dose response in rheumatic heart disease (RHD), a cardiac consequence of rheumatic fever due to the group A streptococcal infection (47 ). Rheumatic heart disease had once been the most prevalent etiology of valvular cardiac diseases worldwide, but over the past few decades, the etiology of most valvular diseases has shifted towards nonrheumatic degenerative etiologies in industrialized countries (48 ). It was often difficult to distinguish valvular etiologies accurately in the clinical setting before the broad application of echocardiography, beginning in the 1980s (49, 50 ). Therefore, in the current analysis, the focus was placed on cardiac valvular deaths in combination with 223 rheumatic and 521 nonrheumatic disorders.

Valvular thickening and fibrosis have been recognized as adverse effects of mediastinal radiotherapy (51 ). Our data from the atomic bomb survivors indicated a significant association between atomic bomb radiation exposure and mortality from nonrheumatic VHD with an upward trend over the period. Although the biological mechanisms to explain our findings remain unclear, it is possible that they mainly derive from aortic valve manifestation. Of the 95 deaths from aortic valve involvement during 1981–2008, 21 subjects participated in the ABCC-RERF examination. Two-thirds of them (14 subjects) had confirmed treatment for aortic valve disorders or had potential symptoms from aortic valve manifestations. This finding thus suggests that deaths diagnosed with the aortic valvular disease were relatively reliable during the period of 1981–2008. Degenerative calcification is implicated in the pathogenesis of aortic valvular disease through pathways, similar to aortic sclerotic lesions (52 ), so it is possible that tissue reactions after radiation exposure accelerated the development of calcification in aortic valves among the atomic bomb survivors (53 ).

In contrast, the increased risk of rheumatic VHD only in 1950–1968 is thought to be caused by a high incidence of rheumatic fever after streptococcal infection, that occurred under the poor economic and hygienic conditions in the periods immediately after the war (54 ), among proximal survivors who suffered under worse conditions than distal survivors, due to the stronger physical destruction and catastrophic damage by the atomic bombs (55 ). Thus, the increased risk of rheumatic VHD in the early period may be a secondary association with poor health conditions of the survivors. Indeed, in a previously published study, no apparent association between radiation exposure and the risk of fatal infectious disease was found (56 ).

Published data from childhood cancer survivors suggest that radiotherapy is linked with HOD by elevated blood pressure through long-lasting endothelial dysfunction (57 ). Atomic bomb radiation has also been shown to increase the risk of HOD (15 ), which is consistent with the results of the current analysis. These findings may be attributable to high blood pressure (58 ) or hypertension (44 ). In addition, increased risk of mortality from chronic kidney dysfunction with radiation exposure (59 ) may be potentially associated with hypertension and HOD among the atomic bomb survivors because kidney dysfunction is a major condition intermediate to hypertension (60 ).

Myocardial infarction, as well as hypertension, have been reported to contribute to a large proportion of heart failure cases in general (60, 61 ), but current results did not suggest that radiation-related heart failure can be attributable to MI. Increased risks for high blood pressure (58 ) or hypertension (44 ) and mortality from HOD may partially explain the high ERR of radiation exposure for HF in the current analysis, but it should be noted that the temporal patterns were different between HOD and HF. Increased risk of HF death was observed in the third period (1981–1994) while increased risk of HOD was observed in the earliest time (1950–1968). Other than hypertensive damage to the myocardium, radiation-induced microvascular damage (6 ) may be linked to myocardial dysfunction resulting in HF. Diagnosis of HF on death certificates is occasionally made for the end stage of various cardiac and noncardiac disease, which is known as “garbage coding” (63 ). Such misclassification leads to a diagnosis of HF that is far from the exact underlying cause of death. While garbage codes have been used less frequently in death diagnosis in developed countries, including Japan, the code of HF for ill-defined or unspecified conditions at death continues to be used (64 ). These issues make it difficult for causal inference from observational data (54, 62 ). Therefore, additional investigations are required to further our understanding of the associations of HOD and HF with radiation exposure using clinical information.

Our study has several strengths, which include a large population, 58-year follow-up with virtually complete mortality ascertainment, a broad but relatively low-to-moderate range of radiation doses (0–<4 Gy) and well-characterized individual dose estimates. Death from valvular disease was validated using prior clinical information when available. Among the study's limitations, a primary concern is that a breakdown of population and outcomes by disease subtype and calendar period could reduce statistical power. Our period definitions may not precisely capture underlying changes in diagnostic practices, since such changes do not generally occur instantaneously. Another concern is that with health examinations, preclinical and subclinical disease can be detected, and early treatment may influence the natural course of the disease, affecting the cause of death. The members of the AHS subcohort of the LSS have been invited to undergo biennial health examinations at ABCC-RERF since 1958. Therefore, we calculated background mortality rates adjusted for participation in the AHS health examination, but results did not change substantially (not shown). However, this adjustment may be insufficient because atomic bomb survivors have received social services in the form of periodic health examinations that include cancer screening. Whether our findings apply to other ethnicities is unknown because both the incidence and pathogenesis of CVD differ between Japan and Western countries. Mortality from heart diseases in Western countries is thought to stem from atherosclerotic progression, but the most dominant risk in Japan is thought to be hypertension. (65 )

In conclusion, estimated radiation risks of heart disease mortality in the LSS tended to vary by disease subtypes and observation period, in accordance with the different pathogenesis of radiation effects and trends in CVD incidence in Japan over the protracted observation period. In the past, disorders related to infection and hypertension tended to be associated with radiation exposure, while more recently risks of atherosclerotic disorders appear to be emerging. The risks at low-dose levels, however, remain uncertain.

The Radiation Effects Research Foundation (RERF), Hiroshima and Nagasaki, Japan is a public interest foundation funded by the Japanese Ministry of Health, Labour and Welfare (MHLW) and the U.S. Department of Energy (DOE). Funding for this research was also provided in part by DOE award no. DE-HS0000031 to the National Academy of Sciences. This work was supported by RERF Research Protocol No. 1-75. The views of the authors do not necessarily reflect those of the two governments. Special thanks are expressed to Ms. Mikiko Hayashi for her assistance.