| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
* Department of Pathology and Geriatrics Center, University of Michigan School of Medicine, University of Michigan Institute of Gerontology, and Ann Arbor Department of Veterans Affairs Medical Center, and
Unit for Laboratory Animal Medicine, University of Michigan, Ann Arbor, MI 48109
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The incidence rate of many forms of lethal neoplasm increases exponentially with age. To some extent this relationship reflects the requirement for multiple rare events to occur in sequence before an initially normal cell is able to undergo transformation to a clinically significant malignancy (12). The idea that changes in antitumor defense mechanisms, potentially including alterations in immune function, might contribute to the dramatic increase in neoplasia in aged mice and humans is controversial (13, 14), although it is clear that severe immunodeficiency of the kind seen in athymic nude rodents and after irradiation or chemotherapy in humans does not predispose to the entire range of tumor types seen in normal aged individuals.
To investigate the relationship, if any, between immune decline in middle age and the timing of late life diseases including spontaneous neoplasia, we have conducted a prospective study in a genetically heterogeneous population of mice, in which each animal was tested at 8 mo and then again at 18 mo of age for peripheral blood levels of seven T cell subsets and then was allowed to live until natural death or until the animal was severely ill. A statistical method using principal components (PC)3 was then used to see whether differences among the mice in T cell subset patterns predicted all-cause mortality rates and to see whether correlations between subset pattern and life span were equivalently strong in mice dying of three specific neoplastic illnesses: lymphoma, fibrosarcoma, and mammary adenocarcinoma.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Animals used in this study were bred as the progeny of CB6F1 females and C3D2F1 males, a cross referred to in other publications as UM-HET3. Because the parents are heterozygotes at many loci, each mouse in the UM-HET3 population is genetically unique and can be considered a full sibling of every other mouse tested. Mice were weaned into same-sex cages at the age of 3–4 wk at three to four mice per cage and were provided free access to laboratory chow and fresh water. To document the specific pathogen-free status of the colony, sentinel animals (not part of the test population) were exposed to pooled spent bedding and then examined for pinworms and for serological evidence of infection with Sendai virus, mycoplasma, or mouse coronavirus. Such testing was conducted quarterly and proved negative throughout the course of the experiment.
In some cages of female mice, a stud male mouse was introduced at 
8
wk of age to create a group of "mated females." Litters were
removed from these cages within the first week after birth, and the
male was removed when the females reached 6 mo of age. Cages of virgin
males, virgin females, and mated females were all housed within the
same room. Husbandry practices were in accordance with American
Association of Laboratory Animal Care and institutional guidelines.
Exclusion criteria
Cages in which fighting among males had led to serious wounding
were culled from the experiment (25% of male cages, all at ages
before 12 mo). Among the 571 mice for which complete sets of T cell
subset measures were available at either 8 or 18 mo of age, 50 were
excluded because it was not possible to infer a single cause of death
from the necropsy. The remaining 521 animals are the subject of this
report.
T cell subsets
Blood samples were taken by tail venipuncture at ages 8 and 18
mo. Each sample was divided into aliquots for two-color assessment of T
cell subsets using flow cytometry. The protocol has been described in
detail elsewhere (15) and is summarized in Table I
.
|
Mice were inspected at least daily. Mice suspected to be ill (because of weight loss, poor grooming, or visible tumor) were observed twice daily except on weekends. Mice judged by an experienced technician to be so severely ill that survival for more than a few additional days was unlikely were taken to the necropsy suite and humanely euthanized; this group made up 59% of the total. Mice found dead (41%) were also submitted for necropsy. The necropsy protocol has been described in detail elsewhere (16) and involves both gross inspection and histological examination of sections from 37 organs.
Statistical analyses: introduction
The overall goal of the analysis was to see whether T cell subset patterns were predictors of life span in groups of mice dying of different causes. The approach used had three stages.
The first stage was to combine the observed T cell subset values into a single number, called a first principal factor, that could serve as an overall indicator of the T cell subset pattern for each individual mouse. The starting data set contained measures of seven different T cell subsets for each mouse. The computer algorithm takes each subset value, multiplies it by a weight, and then adds all seven of these weighted values together to calculate a sum, called the first principal factor, for each individual mouse. (For technical reasons, the actual subset values are first normalized, i.e., expressed as the number of SDs above or below the mean value for the subset across the whole set of mice.) The weights used are selected, by the algorithm, so that the variation (SD) of the factor, within the whole group of mice, is as high as possible. This procedure assigns one new number, the principal factor, to each mouse as a quantitative measure of its T cell subset pattern as a whole. As explained in Results, the particular weights that emerged from this calculation suggest that the first principal factor in this mouse data set is a reflection of the "immunological age" of the mouse.
The second stage of the analysis is to test the hypothesis that the first principal factor, calculated as explained above, can predict life span, considering the entire mouse population. The method used for testing this hypothesis is similar to linear regression, using equations in which one or several independent variables are used to calculate the expected value of a dependent variable. In this case, the dependent variable is a measure of the mortality risk, and the independent variables are the principal factor (the measure of T cell subset pattern, calculated as described above) plus other variables such as age and gender. The null hypothesis here is that there is no relationship between T cell subset pattern and life expectancy. The regression calculation yields a p value, the probability that the null hypothesis is correct. A low p value provides support for the conclusion that T cell subset patterns, as condensed into a single principal factor, are indeed able to predict mortality risk and life span. The regression calculation also produces a regression coefficient that tells how strongly mortality risk is associated with the immune subset principal factor. For example, a coefficient of 1.37 would imply that a one-SD shift in the principal factor is associated with a 37% increase (or decrease) in mortality risk.
The third stage of the analysis is to repeat stage 2, but this time using only subpopulations of the mice, specifically groups of mice that differ by cause of death. The hypotheses to be tested here are that immune subset patterns are significant predictors of mortality risk in mice dying of lymphoma, in mice dying of fibrosarcoma, etc.
Three other points deserve note. First, the first principal factor can be calculated at any age. We use the term "F1_18" to refer to the first principal factor calculated using T cell subset data from mice 18 mo of age. "F1_8" refers to the same calculation, but using data from mice at 8 mo of age. Second, it is possible to calculate second and third principal factors to sum up variation among the mice that remains after adjustment for the first principal factor. These second and third factors, if evaluated in 18-mo-old mice, would be called F2_18 and F3_18. As shown in Results, these second and third factors were found not to be good predictors of life span, and so information about them is not included in the tables that present the regression results, i.e., stages 2 and 3 of the analysis.
Lastly, some of the tables also report "factor loadings." These are ordinary correlations between the principal factor and the T cell subsets used in the calculations. A high positive factor loading implies that mice with high values of the particular T cell subset tend to score high in the principal factor. A high negative factor loading implies that mice with high values of the particular T cell subset tend to have low scores for the principal factor. A factor loading that is close to zero implies that the T cell subset has little influence on the principal factor score.
Statistical analyses: technical methods
The initial data set consisted of measures of seven T cell subsets for each of 521 mice, for which each T cell subset had been measured at 8 mo, at 18 mo, or at both ages. In previous work (17) we have found that four of these subsets, measured at 18 mo of age, were each individually a significant predictor of longevity in these four-way-cross mice. A PC method was used to combine data from these seven T cell subsets into a smaller number of composite indices to see whether one or more of the uncorrelated composite factors could serve as a predictor of remaining longevity. The PC algorithm, implemented in NCSS statistical software (NCSS Statistical Software, Kaysville, UT), calculates PC factors directly from the data (i.e., not from a correlation or covariance matrix) using multivariate regression to estimate missing values. PC factors with an eigenvalue >1 capture more variance than any single subset measure alone, and for this reason they were used as predictor variables in the proportional hazard regression model. Factor loadings—the correlation between each of the PC factors with each of the original trait measurements—were used as guides to the biological interpretation of each of the extracted factors. To determine whether one or more of the extracted factors was a predictor of longevity, we calculated a proportional hazards regression analysis in which longevity was modeled as a function of five predictor variables: each of the three PC factors with eigenvalues >1, plus sex (male or female), plus mating status (mated or not mated). This procedure—calculation of PC followed by proportional hazards regression—was conducted twice: once using T cell subsets measured at 8 mo of age and then using T cell subsets measured at 18 mo of age. In each case the regression considered only those mice for which a complete set of seven subset measurements was available at the age in question, because only for these mice is it possible to calculate the PC factor value from the eigenvectors that relate the observed subset values to the calculated composite indices. When the proportional hazard regression indicated a significant association between predictor variables and longevity (as indeed it did for both ages), the regression was then repeated for subpopulations of the mice, i.e., those dying of specific common lethal illnesses.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
summarizes the
results of this set of necropsies. Three groups of mice are shown that
differ in gender and, for the females, in reproductive history. There
are substantial differences among these three groups in the
distribution of several varieties of neoplasia. Pituitary adenoma and
lymphoma, for example, were more frequent as a cause of death for
females than for males, whereas in contrast males tended to die more
frequently of hepatocarcinoma or of pulmonary adenocarcinoma. The
incidence of lethal mammary adenocarcinoma was more common in the mated
than in the virgin females. Among males, 32% died of a mouse urinary
syndrome (18, 19) that is thought to reflect psychological
stress secondary to adjustments in dominance hierarchy among
group-housed males. The miscellaneous group of neoplastic illnesses
included hemangiosarcomas, a granulocytic leukemia, harderian gland
adenocarcinomas, rhabdomyosarcomas, and squamous cell carcinomas, among
other lesions. The miscellaneous group of nonneoplastic lesions
included cases of endometritis, enamel organ dysplasia, glomerular
amyloidosis, myocarditis, intussusception of the jejunum, and abdominal
hematoma secondary to disseminated intravascular coagulation, among
other lesions. Some form of neoplasia was deemed responsible for death
in 82% of the diagnosable females and in 47% of the diagnosable
males.
|
includes
the factor loadings for each of the three factors, referred to
respectively as F1_8, F2_8, and F3_8. The factor loadings, which are
the correlations between the factors and the individual T cell subset
measures, show that high values of F1_8 are found in mice that, on
average, tend to have relatively high levels of CD4M and CD8M cells and
relatively low levels of CD4 and CD4V cells. Previous work in isogenic
as well as in genetically heterogeneous mice (3) has shown
that aging increases mean levels of CD4M and CD8M and decreases mean
levels of CD4 and CD4V subsets, and these age effects were confirmed in
the current set of mice (data not shown). Thus, high values of F1_8
identify mice whose immune systems resemble those of chronologically
older mice, and in this sense F1_8 may be interpreted as a composite
index of immunological aging. Being able to calculate for each mouse a
single value that combines information from several T cell subsets into
a numerical measure of T cell aging is a first step in testing the
hypothesis that T cell changes may be associated with differences in
late life susceptibility to diseases (see below).
|
shows that F2_8 reflects
the outcome of a process that influences expression of P-glycoprotein
on both CD4 and CD8 T cells, and F3_8 is influenced largely by CD8
levels. Because CD4P and CD8P cells both increase with age, these
correlations between principal factors and individual subsets suggest
the hypothesis that F2_8 might also be associated with differences in
the mice in mortality risk.
We conducted similar calculations using T cell subset measures obtained
at 18 mo of age instead of the data from the 8-mo-old mice. Factor
loadings are shown on the right side of Table III for the three factors
(F1_18, F2_18, and F3_18) with eigenvalues >1. F1_18, like F1_8, shows
high loadings for the age-sensitive subsets CD4 M, CD8 M, CD4V, and
CD4. Unlike F1_8, F1_18 also has a relatively high loading for CD4P
which, like CD4 M and CD8 M, increases significantly in middle age.
Thus, F1_18 is also plausibly interpreted as an index of immunological
age, with high values seen in mice whose T cell subset patterns are
similar to that expected for older mice. It is worth noting that the
procedure used for calculating the first principal factors (F1_8 and
F1_18, respectively, at the two ages) does not incorporate any
information about the life span of the mice or the effects of aging on
the T cell subsets. The observation that these first PC are largely
dependent on T cell subsets that are known from previous work to be
age-sensitive suggests that much of the T cell subset variation in
adult and middle-aged mice may reflect differences among mice in the
pace of aging or the effects of aging on T cell homeostasis.
In the second stage of the analysis, proportional hazard regression was
used to determine whether one or more of these factors, measured at
either age, was a significant predictor of life span. Using data
obtained at 8 mo, the regression model incorporated five factors as
potential predictors of mortality risk: the three factors generated by
the PCA calculation and two others reflecting gender and reproductive
history. Of these, only F1_8 had a significant association with
longevity. The first line of Table IV
shows the results of this regression analysis for all 428 mice for
which F1_8 could be computed. The first two columns show the strength
of the association between F1_8 and life span (B) and the SE
for this estimate. Exp(B) represents the relative increase
in risk associated with a change of one SD in F1_8, i.e., 17% in this
instance. This means that a mouse with a value of F1_8 that is one SD
above the average would have a 17% increase in mortality risk compared
with the average mouse. Similarly, a mouse with an F1_8 score one SD
below average would show a 17% lower mortality risk than average. The
Z statistic and its associated probability,
p(Z), estimate the likelihood that
B = 0. In this case, F1_8 is seen to be significantly
(p = 0.001) associated with differences in
longevity among the mice, with high values of F1_8 predicting reduced
longevity. This indicates that the relationship between high scores of
F1_8 and high mortality risk is unlikely to have resulted by chance
alone. Two R2 values are included in
Table IV. The first of these shows R2
for F1_8 by itself; for all-cause mortality,
R2 = 2%, indicating that interanimal
variation in F1_8 accounts for only 2% of the differences in
mortality risk among these mice. The second
R2 shows the strength of the
correlation using the entire model, i.e., including variance accounted
for by the other two immunological factors, gender, and reproductive
history. Thus, differences among mice in F1_8 account for a small, but
significant, proportion of variation in life expectancy at this
age.
|
, except that this time
the calculations were performed only on groups of mice that had died of
each specific illness. Table IV shows that F1_8 is indeed significantly
associated with life span in each of these three classes of animals.
Indeed, this factor accounts for 12% of the variance in life
expectancy in mice dying of mammary adenocarcinoma or of fibrosarcoma
(as indicated by the values in the R2
column) and for 6% of the variance in the lymphoma deaths. For the
former two diagnoses, a change of one SD in F1_8 is associated with a
41% increase in mortality risk, as indicated by the value of 1.41 in
the Exp(B) column. Fig. 1
shows scatterplots illustrating the relationship between F1_8 and life
span in each of these three groups of mice. Neither F2_8 nor F3_8 had a
significant association with mortality risk in any of these diagnostic
subsets, although there was a marginal association (Z =
1.89, p = 0.06) for F3_8 in mice destined to die of
fibrosarcoma (data not shown).
|
shows the results of a similar
proportional hazards regression using subset data from 18-mo-old
animals. The model included all three factors (F1_18, F2_18, and F3_18)
as well as gender and mating status, but only F1_18 was found to have a
significant association with life span for all-cause mortality or for
any of the three diagnostic groups. The results in Table V show that
F1_18 is a highly significant predictor of subsequent life span, with
p(Z) < 0.0001 for all-cause mortality.
F1_18 also has significant ability to predict life span in four
independent subpopulations of mice, i.e., those dying of lymphoma, of
mammary tumors, or of fibrosarcoma, as well as those mice dying of any
other cause. These differences in T cell subset patterns account for
12–13% of the variance in life expectancy in the lymphoma and
fibrosarcoma groups. Fig. 2 presents the
scatterplots for the three leading causes of death.
|
|
and V, for three groups of animals:
virgin males, virgin females, and mated females. Table VI shows the results. High values of
Z are associated with low p values and correspond
to a strong association between subset pattern and mortality risk for
the indicated group of mice. For mice tested at 18 mo the association
between F1_18 scores and life span was seen in all three groups of
mice, indicating that at this age T cell subsets are good predictors of
life span in virgin males, virgin females, and mated females. In
contrast, only the mated female group showed a clear association
between T cell status and life span when the T cell tests were made at
8 mo of age.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Developing useful biomarkers of aging has proven to be remarkably difficult, in part because many age-sensitive variables tested as candidate biomarkers are sensitive to genetic and nongenetic influences other than aging per se. The PC approach, by combining information from multiple, mutually correlated age-sensitive traits, may generate a more robust and reliable index of interindividual differences in aging rate. Any individual assay, for example a test of a specific T cell subset in a single blood sample, is likely to have a good deal of uncertainty, but the combination of results from related tests may increase the signal-to-noise ratio and thus provide stronger predictive power than any single assay by itself. The development of such age-sensitive indices in a genetically heterogeneous stock should help avoid the complications of strain-specific idiosyncrasies that have bedeviled previous attempts to seek a useful surrogate for biological age (20).
Although previous data from this laboratory have already shown that
CD4, CD4 M, CD8 M, and CD4V subsets, measured in 18-mo-old mice, are
significant predictors of all cause mortality (17), the
newly available data on cause of death pathology, together with the PC
method of analysis, lead to several new conclusions of interest. 1) The
results in Tables IV and V show that the association between subset
pattern and longevity is significant regardless of cause of death, at
least for those illnesses that are common enough to provide reasonable
statistical power. The data thus imply, for the first time, that three
dissimilar forms of cancer (lymph, breast, and fibroblast) are
influenced by the same underlying factors and furthermore that these
common factors affect T cell subset levels by slowing or speeding the
effects of aging on T cell balance. 2) By testing the subset/longevity
association in four nonoverlapping groups of animals (those dying of
lymphoma, adenocarcinoma, fibrosarcoma, and, for the 18-mo-old mice,
all other causes combined), the current study shows that the ability of
T cell subset patterns to predict life span is not merely due to
effects of a specific disease on T cells, or vice versa. 3) The
individual T cell subsets, analyzed one at a time in the earlier paper
(17), did not predict life span when measured at 8 mo of
age. In contrast, the use of the PC method to "pool" data from
different subset tests on the same mouse shows that the first PC is a
significant predictor of life span (at least in mated females) even at
this early age. This observation suggests that whatever developmental
or pathological pathways link subsets to death, these are detectable in
relatively young mice and thus do not reflect changes that occur only
in old age, again at least in mated females. The specific hormonal
factors that produce the association between T cell subsets and
longevity earlier in the life of mated female mice than in the other
groups clearly deserve further experimental study.
We have not attempted in this paper to test the hypothesis that T cell subset patterns can predict the cause of death, to predict, for example, whether an individual will die of lymphoma, mammary cancer, or some form of nonneoplastic illness. The analytical strategy instead addresses a separate issue: can our composite index of immune function predict longevity, and can it do so for subgroups of mice destined to die of different specific illnesses? Our findings suggest that indeed a statistical combination of data from different T cell subset tests can predict longevity and thus reflects biological processes that influence the incidence or progression of late-life diseases, including important forms of cancer. The demonstration that the composite factors predict longevity in separate subpopulations of mice (i.e., those dying of different diseases) provides multiple independent replications of the association between factor value and longevity, thus greatly strengthening the inference that the association is not merely a statistical fluke.
These correlations we have documented are consistent with two theoretical models. The "immune protection" model would suggest that the effect of aging on the immune system creates a characteristic pattern of T cell subset changes that can speed up or slow down multiple forms of neoplasia. It is difficult to disprove this idea, but there are at present no compelling reasons to think that tumor incidence or progression would be influenced by the relative proportions of CD4 M, CD8 M, CD4P, CD4, or CD4V cells, and there is a substantial body of evidence suggesting that alterations in immune function are not themselves a major risk factor for the forms of neoplasm most common in older individuals. It also seems unlikely that alterations in immune status detectable in 8-mo-old mice would influence the pace of diseases that are typically not detectable until 18–24 mo of age and that typically do not cause death until 27 mo of age. The varieties of cancer seen most commonly in our mice—lymphoma, mammary adenocarcinoma, and fibrosarcoma—overlap only partially with those common among older humans, but do reflect substantial diversity in germ layer and tissue origin, rate of progression, and viral or nonviral etiology. This diversity also argues against the "immune protection" model, because it seems unlikely that alterations in protective immunity would have similar effects on the timing of such biologically disparate forms of neoplasia, representing in this case tumors of hematopoietic, connective tissue, and secretory cell origin.
A second model suggests that both increased mortality risk and altered T cell subset patterns are dependent on interindividual differences in biological aging rate. This "gerontological" model implies that just as members of different species age at different rates, so might individual mice differ in the rates at which they exhibit changes in age-dependent traits, including changes both in age-sensitive T cell subset levels and in those unknown factors that increase the risk of late-life illness. It is particularly noteworthy that these composite immune indices are significant predictors of life span in subsets of mice dying of different causes, i.e., lymphoma, fibrosarcoma, and mammary adenocarcinoma, because such a finding suggests that the incidence or progression of all three forms of illness may be timed by a common pacemaker. The evidence that F1_18 is also a significant predictor of longevity in the group of mice dying of causes other than the three listed tumors suggests that one or more other late-life illnesses may also be regulated by this pacemaker. The gerontological model makes the prediction that middle-aged mice with relatively extreme levels of an age-sensitive immune factor (such as F1_18) will also show relatively extreme values in tests of other age-sensitive traits, such as changes in bone structure, muscle function, and patterns of gene expression. We have presented elsewhere (15) preliminary evidence that middle-aged mice with relatively advanced T cell subset aging also show relatively weak muscle strength, and a more comprehensive set of similar studies is now under way. Some of the age-sensitive T cell subsets used in the PCA calculation are known to be under the control of polymorphic alleles in this stock of genetically heterogeneous mice (21), and it will be interesting to determine to what extent the three immune composite variables may be regulated by segregating alleles.
The stratification by gender and mating history (Table VI) suggests
that the association between immune status and life expectancy may
develop more rapidly in mated female mice than in virgin males or
virgin females, although it is apparent in all three subpopulations by
18 mo of age. A previous analysis of individual T cell subsets
(17) has shown that the relative ability of specific T
cell subsets to predict longevity differs between mated and virgin
females and also between virgin males and virgin females. The PC
approach has allowed us for the first time to calculate an index of
immunological age that predicts longevity in all three of these
populations, does so for multiple causes of death, and has prognostic
power as early as 8 mo of age.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
2 Address correspondence and reprint requests to Dr. Richard A. Miller, Room 5316 CCGCB, Box 0940, University of Michigan, 1500 East Medical Center Drive, Ann Arbor, MI 48109-0940. E-mail address: millerr{at}umich.edu
3 Abbreviations used in this paper: PC, principal components; PCA, PC analysis.
Received for publication February 19, 2002. Accepted for publication May 21, 2002.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  W. R. Swindell, J. M. Harper, and R. A. Miller How Long Will My Mouse Live? Machine Learning Approaches for Prediction of Mouse Life Span J. Gerontol. A Biol. Sci. Med. Sci., September 1, 2008; 63(9): 895 - 906. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Acuna-Castillo, E. Leiva-Salcedo, C. R. Gomez, V. Perez, M. Li, C. Torres, R. Walter, D. M. Murasko, and F. Sierra T-kininogen: a biomarker of aging in fisher 344 rats with possible implications for the immune response. J. Gerontol. A Biol. Sci. Med. Sci., July 1, 2006; 61(7): 641 - 649. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. R. Warner The Future of Aging Interventions: Current Status of Efforts to Measure and Modulate the Biological Rate of Aging J. Gerontol. A Biol. Sci. Med. Sci., July 1, 2004; 59(7): B692 - B696. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. A. Rikke Early Life Predictors of Old-Age Life Expectancy Sci. Aging Knowl. Environ., May 19, 2004; 2004(20): pe21 - pe21. [Abstract] [Full Text]
|
|
|
|
 |
|
 M.-E. Blais, G. Gerard, M. M. Martinic, G. Roy-Proulx, R. M. Zinkernagel, and C. Perreault Do thymically and strictly extrathymically developing T cells generate similar immune responses? Blood, April 15, 2004; 103(8): 3102 - 3110. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
