- Email: [email protected]
Quantitative measurement of human physiological age by profiling of body fluids and pattern recognition
Mechani,s'ms of Ageing and Develnpnwnt. 59 (1991) 47--67
47
Elsevier Scientific Publishers Ireland Ltd.
QUANTITATIVE M E A S U R E M E N T OF HUMAN P H Y S I O L O G I C A L AGE BY P R O F I L I N G OF BODY FLUIDS AND PATTERN R E C O G N I T I O N
A R T H U R B. ROBINSON and L A U R E L E E R. ROBINSON* Oregon Institute of Science and Medicine. 2251 Dick George Road. Cave Juncthm. Oregon (U.S.A.) (Received November 9th, 1990)
SUMMARY
Quantitative correlations with human age are demonstrated for 60 substances from a group of 200 substances measured in the urine of 235 men. Simplified pattern recognition calculations are used to combine these correlations into patterns of human age and to demonstrate their utility for the quantitative measurement of human physiological age and aging rate. The empirical use of these techniques for the extension of human life-span and diminution of human suffering from degenerative diseases is discussed. Current experimental limitations of this method are demonstrated and evaluated. The application of these techniques can form the basis for a significant advance in the quality of human life.
Key words: Aging; Human; Analysis; Pattern; Urine; Quantitative
There are many hypotheses regarding the primary cause of the observed intrinsic life span of organisms [1--10]. It is to be expected that the eventual resolution of this unsolved problem in molecular biology will lead to methods of altering the primary causes of aging and thereby the life span of organisms including humans. These alterations may be expected to be of the two types illustrated in Figs. I--3 [11]. First, the death rate of humans during the years much earlier than their intrinsic life span will be reduced. Second, the intrinsic life span itself will be increased. Correspondence to: A.B. Robinson, Oregon Institute of Science and Medicine, 2251 Dick George Road, Cave Junction, OR 97523, U.S.A. *Laurelee Robinson died on November 12,1988. She saw and approved all of the figures and tables in this paper. She did not read the final manuscript. 0047-6374/91/$03.50 Printed and Published in Ireland
© 1991 Elsevier Scientific Publishers Ireland Ltd.
48
100%'
LLI
80%
>
\
60% LU
a.
0
40%
LLI n
2O% 20
40
60
80 100 YEARS OF AGE
120
140
Fig. I. Aging curve for a population comprised of American males as calculated from life expectancycompilations of the U.S. Public Health Service.This figure is reproduced from Robinson, A.B., Mech. AA,ein A, Dev., 9 (1979) 225.
It should, however, be possible to extend h u m a n life spans by means of empirical e x p e r i m e n t a t i o n even prior to the a d v e n t of a t h o r o u g h u n d e r s t a n d i n g of the molecular biology of h u m a n aging. O b s e r v a t i o n s o f a n i m a l s and h u m a n s suggest that substantial progress o f the type illustrated in Fig. 2 a n d modest progress of the type illustrated in Fig. 3 should be possible.
100%
80% > ,~ 60% LLJ
0 LU a.
40%
20%
20
40
60
80 100 YEARS OF AGE
120
140
Fig. 2. Example of a change in the human aging curve through improvements in diet and other livingconditions, but with no change in intrinsic life span. This figure is reproduced from Robinson, A.B.. Mech. Ageing Dev.. 9 (1979) 225.
49
100% 80% > -- 60% <~ LU
, t
UJ
40% 0
uJ
o- 20%
20
40
60
80 100 YEARS OF AGE
120
140
Fig. 3. Example of a change in the h u m a n aging curve through improvements in diet and other living conditions and through an increase in the intrinsic life span. This figure is reproduced from Robinson, A.B., Mech. Ageing Dev.. 9 (1979) 225.
The principal impediment to this empirical progress is the fact that the humans carrying out the experiments have life spans of comparable length to those of the human subjects which they are studying. If carried to completion, a single rigorous experiment on human life span as a function of, for example, diet or environment
100%
LIFE REMAINING
I
0%
.
0
I
PROBABILITY OF ILLNESS X
WELL
,,
I
1.0
SICK
SEVERITY OF ILLNESS
Fig. 4. Representative axes of physiological age, probability of a specific illness, and the severity of a specific illness. It is the quantitative determination of the positions on these axes of specific individual humans, by means that are low cost and generally applicable to all well individuals at frequent intervals, which is the goal of our work on the physiological profiling of human body fluids.
50 could require as much as 50 to 80 years. Even if this is reduced to 10 to 20 years in some cases by clever experimental design, a complete series of experiments will still proceed too slowly to be of benefit to humans who are now living. In the absence of real progress and as a result of the understandably high human interest in this subject, many sensible but unproved aging hypotheses have given rise to fads and cults especially as pertain to diet. The impediment to human aging research of long experiment times would be removed if an objective and quantitative means for determining human physiological age with reasonable accuracy could be developed. This would allow empirical experiments on aging rate to be carried out with experiment times in groups of humans of a few months per experiment or less. Moreover, quantitative analysis of human age would allow the monitoring of the rate of aging in single individuals, so that biochemical individuality and individuality of particular environmental circumstances could be taken into account in devising conditions that minimize the rate of aging in each individual. We have long been interested in the development of a procedure for the quantitative measurement of human physiological age [l 1--14]. This interest has been a part of our general interest in the quantitative measurement of human health especially with respect to degenerative disease. If by means of inexpensive quantitative analysis the state of each person could be determined on quantitative axes as illustrated in Fig. 4, then their productive and actual life spans could be extended; their illnesses could be more effectively cured; and their health could be guarded by preventive measures which combat the probability of disease rather than disease itself. It is an essential obligation of scientists to do all that they can to optimize the quality of all human life. This optimization requires, in part, the minimization of disease and the maximization of life span for the entire human population. Techniques of the sort illustrated in Fig. 5 and discussed below have substantial potential to meet these parts of that obligation. We have limited ourselves to those techniques which have potential, in mass application, to quantitatively measure the amounts of a large number of substances in human body fluids at a cost of about $ l0 or less for an entire analysis including its computerized interpretation. This limitation assures that the techniques developed will be available to most people rather than a wealthy few. Having measured within this restriction as many parameters as possible, we then have attempted to extract as much information from these measurements as possible by simple computerized pattern recognition techniques. This is a departure from the usual approach of developing observational techniques for those substances thought to be of importance to a particular hypothesis or theory. Fundamental science is more entertaining. We have, in our work on deamidation and molecular clocks [10,11], enjoyed this, too. In fact, a side benefit of the profiling work has been the discovery of a few correlations which happen by
51
Fig. 5. One aspect of the optimization of the quality of human life.
chance to be relevant to basic research. Nevertheless, choice of substances by economics and simplicity of analysis rather than by hypothesis is an essential part of rapid progress in the practical profiling of human health. Our work has comprised the use of ion-exchange chromatography, gas-liquid chromatography, and direct mass spectrometry, including chemical ionization and field ion mass spectrometry (without chromatographic separation), of samples of urine, blood, and breath from humans and mice. We also have carried out some work on the whole body compositions of fruit flies [ 12].
HUMAN URINARY AMINES - AGING
10
20
30
40
50
SUBSTANCE NUMBERS
Fig. 6. An example chromatogram of the amines, amides, and amino acids in human urine as used in these experiments. The base line is computer drawn. The darkened strips along the bottom axis show peaks which the computerized procedure has decided to combine for computational analysis.
52
For the most part, however, this work has been restricted to humans. Since this approach is empirical in nature and has an immediate practical objective, we prefer to avoid extrapolations between animals and humans in the interpretations of results. In a highly automated laboratory (see for example Ref. 15), approximately 20 000 analyses of human body fluids including the quantitative analysis of over 2 million amounts of substances have been carried out by us and under our supervision (see for examples Refs. 12, 13, 15 and 16). The primary focus of this work has been on states of disease. A small part was devoted to the measurement of organismic age (see Refs. 12 and 14) in mice and fruit flies. We report herein a few experiments which we carried out to test the measurement of chronological and physiological age in humans by urine analysis. Figures 6 and 7 illustrate the analytical techniques used in these urine analysis experiments. Graphs of this sort were not, however, used in the experiments. The detector voltages as a function of time from the analytical machines were sampled by multiplex, collected on magnetic tape, and transferred to a calculation computer. There, by means of automatic computer programs which we developed, a baseline was fitted and subtracted: the peaks were integrated: retention time variation was eliminated and the peaks were matched within the entire experimental set: the peak areas were normalized in order to remove systematic variations in kidncy function or water elimination: the experiments were evaluated by non-parametric statistics and by diagnostic power calculations; the diagrams shown in Figs. 8, 9 and 13 17 were automatically drawn; and the values given in Tables 1 and I1 were automatically calculated. 'Diagnostic powers', DA, were calculated according to the method of Robinson and Westall [17] and the "diagnostic coefficients', RA, were calculated as described in Ref. 14. Casual urine samples and chronological ages of 205 men were obtained from Dr. A. Cherkin of the Veterans Administration hospital in Sepulvida, California. In
H U M A N URINE VAPOR SUBSTANCES
10
20
30
40
50
60
70
AGING
80
90
100
110
120
130
SUBSTANCENUMe£RS
Fig. 7. An example chromatogram of Ihc volatile c o m p o u n d s in human urine as used in these experiments. This analytical technique is described in detail in Rel'. 15. The base line is computer drawn. Thc darkened strips along the bottom axis show peaks which the computerized procedure has decided to combine for computational analysis.
53
URINE AMINES YOUNG MEN (35) VS OLD MEN (35)
50 F o-" oo
,< g
3oF !
o 20 - .z C¢ MJ
!
z
0
0.2
0.4 0.6 0.8 MAXIMUM PAGE
1.0
Fig. 8. Cumulative distribution function of the probabilities of correlation with human age of 51 urinary amines, amides, and amino acids.Without correlation the dots would fall along the diagonal line. About 30 of these substances are seen here to be age correlated.
a d d i t i o n , casual urines for 30 men were o b t a i n e d from the same source wherein the ages o f the individuals were kept at S e p u l v i d a a n d not c o m m u n i c a t e d to us until after the c a l c u l a t i o n s shown in T a b l e 111 had been c o m p l e t e d . Casual urine without dietary o r o t h e r c o n t r o l was used, so that the results w o u l d reflect those to be
URINE VAPOR SUBSTANCES YOUNG MEN (33) VS OLD MEN (33)
120 ~ILl
I--
,ooi
~
80 ~
0
60-
iv" W
:D. Z
40~20 h V
0
I
0.2
t
____
l
0.4 0.6 MAXIMUM PAGE
I
0.8
1.0
Fig. 9. Cumulative distribution function of the probabilities of correlation with h u m a n age of 135 urine vapor substances. Without correlation the dots would fall along the diagonal line. About 30 of these substances are seen here to be age correlated.
21 48 3 11 4.6 17 23
3O 36 16 14 13 42 2 15 33 39
28 18 10 34 44
6 1 45 9 8 12 22 7 5 20 19 24
SUBSTANCE NUMOER OLD OLD YOUNG YOUNG YOUNG YOUNG OL D YOUNG OL O OLD OLD OLD OLD OLD OLD OLD YOUNG 0 LD OLD OLD YOUNG YO UNG YOUNG OLD YOUNG OLD YOUNG YOUNG YOUNG YOUNG OLD OLD OLD OLD
PHOSPHOSERINE HiSTIOiNE 3 ASPARAGINE + G LUTAMINE 3 SERINE GLYCINE -
THREONINE 0 L UTATHI 0 N E METHIONINE CYSTATRIONE -
-
CYSTINE
~-AMiNO BUTYRIC ACID ALANINE 3 LYSiNE
PHOSPHOETHANOLAMINE VALINE
~-AMINO BUTYRIC ACIO ETRANOLAMINE -
TAURIRE GLUTAMIC ACID 3 ORNITHINE LEUCINE
HIGHER GROUP
ASPARTIC ACID 1
STANDARDSUBSTANCE WITHIDENTICAL ELUTION TIME
10- 5 10-5 10-4 10-4 10-6
1.5 x 1.6 x 1.6 x 1.6 x 1.7 x
8.2 x 9,5 x 9.8 x 1.4 x 10-1 10-1 10-1 10-1 10-1
10-2 10- 2 10-2 10-1
0.5 x 10-2 7.8 x 10-2
6.2 x 10-2 6.3 x 10-2 6.3 x 10-2
6,1 x 10-2
2.8 x 10-3 7.1 x 10-3 0.9 x 10- 3 1.6 x 10-2 2.5 x 10-2 3.4 x 10-2 4.2 x 10-2
2~5 x 8.3 x 3.0 x 4.0 x 9.3 x
4.8 x 10-7 2.9 x 10-6 4.1 x 10-6 1.2 x 10-6 1.3 x 10-5 2.2 x 10-5
< 1 0 -8
PAGE
7.4 7.6 0.0 6.3 0,5
4.2 4.0 6.0 7.2
3.3 4.0
2.7 3*2 3,2
2.0
0,14 0.30 0.45 0.00 1.3 1.7 2.1
0.9912 0.0042 0.016 0.026 0,047
0.000025 0.00015 0.00021 0.00061 0.00060 0.0011
NUMBER OF SUBSTANCES EXPECTED WITH P ~PAGE
1.71 -+0.11 4.00 -+0.33 3.63 -+0.10 24.0 -+2.2 0.43 -+1.03 11.10 -+0.42 3.40 -+0.24 1.00-+0.06
20.2 -+1.7 10.49 + 1.19 0.79 -+0,42 3.91 -+0.11 0.93 -+0.04 40.3 -+1,'6
31 32 33 34
28 29 30
23 24 25 26 27
2.92 ± 0.27 3.42 ± 0.71 52.8 -+4.9 6,03 -+0.23 2,75 -+0,12 3.77 + 0.19 0.62 ± 0.16
35.9 +2.2 2.34 -+0,10 1.71 -+0.31 44.0 +-5.3 6,21 -+0.27 3,12 -+0.20 4.08 ± 0.24 6.90 -+0.19
2,10 -+0,11 1.74 -+0.08 5.63 -+6.32 1.59 + 0,10 1.16 -+0,12 1.40 -+0.09 1,74 -+0.69 1.43 ± 0.06 4.09 ± 0.21 1.U -+0.29 1.64+-0.15 1,16 -+0.06 1.47 -+0,07 3.76 + 0.21 4.09 -+0,11
13 14 15 16 17 18 19 20 21 22
2L4 -+2.1 97.0 ±7.4 3.16 ± 0.11 16.43_+1.21 20.1 -+0.9 6.05 -+0.30 2.82 -+0,10 1.50 -+0,08
0.68
0.10
0 .06 0.45 0,47 0.39 0.35 0.47 0,41 0.39 0.28 0.18 0.24
6.23 -+0,16 9.44 ± 0.27 43.0 "+2.3 27,3 -+1,6 10.5 -+0.9 59.2 +4,8 3.67 -+0.11 9.00 -+0.62 30.2 ± 0,8 7.00 -+0.21 3,48 -+0.17 1,91 -+0.10
4.70 -+0.19 7.46 "+0.27 50.2 -+1.0 40.0 -+2,0
I 2 3 4 5 6 7 0 9 10 11 12
AGE DIAGNOSTIC POWER
AVERAGE NORMALIZED AREA FOR 36 OLD MEN2
AVERAGE NORMALIZED AREA FOR 35 YOUNG MEN 2
NUMBER OF SUBSTANCES FOUND WITH P~PAGE
A G I N G P A T T E R N FOR MEN AS D E T E R M I N E D BY 51 N I N H Y D R I N - P O S I T I V E U R I N A R Y A M I N E S , AMIDES, A N D A M I N O ACIDS. T H E SUBSTANCES A R E LISTED 1N O R D E R O F D E C R E A S I N G P R O B A B I L I T Y OF C O R R E L A T I O N WITH A G E
TABLE I
ALLYL ISOTHIOCYANATE 2-HEPTANONE
42
109 134 07
3-PENTANONE 2-NONED.4-ONE 3-HEPTANONE 2-NONANONE CARVONE
16 01 41 83 117 38 124
44
3-METHYL-2-BUTANONE; BENZENE; 2-METHYLTETRAHYOROFURAN
2-PENTANONE
14 77
12
2-OUTANONE; 2-METHYLFURAN; 3.METHYLFURAN
STANDARD SUBSTANCE WITH IDENTICAL ELUTION TIME
8
87
SUBSTANCE NUMBER
OLD OLD YOUNG
YOUNG
YOUN6 YOUN6 YOUNG OLD YOUNG YOUNG YOUNG OLO
YOUNG
YOUNG OLD
YOUNG
YOUNG
HIGHER GROUP
1.0 x 10-2 1.7x 10-2 1.9 x 10-2
1.0 x 10-3
2.2 x 10-3 2.4 x 10-3 2,4 x 10-3 3.0 x 10- 3 4.1 x 10-3 5.5 x 10-3 7n x 10-3 0.0 x 10.3
1.3x 10-3
2,4 x 10-4 4,4x 10-4
1,4 x 10-4
2.9 x 10-6
PA6E
1,4 2.3 2.6
1.4
0.30 0.33 0.33 0,,47 O.U 0.75 0.04 1.2
0,18
0,032 0.0H
0.019
0.00427
NUMBER OF SUBSTANCESEXPECTEO WiTH P
16 16 17
14
E 7 0 9 10 I1 12 13
6
3 4
2
1
NUMiER OF SUBSTANCESFOUND WITH P
-+7.5
0.248 -+O.N4 0.037 + 0.906 0,431 + O.OS2
5.26 +O.H
1.14 +0.12 0.740 + 0 M S IJHI -+2.69 0.155+0.014 0,120 +0.013
0.223+ 0.018
1.59 +0.11
0.826+-0.073
1.77 ±0.11
10,4 +2.0 0.400 ± 0.027
41.9
0.323 ±0.038
AVERA6E NORMALIZEO AREA FOR 33 YOUNG MEN1
±0,46 0210 +0.017 0,0§7 ± 0,0BS 0266 -+0.037
3.04
0A16 + 0.036 I . N ±0JO 0.1U +0.010 2.00 2 + 0.74 0.96 ± O.0OS 2.16 ± 1.66 0.106 +0.017 0.161 +-0.012
124 -+0,11
0.5 ±2,0 0.975 + 0.123
±2.4
0.104 -+0.012 15.9
15. T H E
0.30
0.47 0,4a
0.48
0.02
AGE OlAGNOSTIC POWER
IN R E F E R E N C E
AVERAGE NORMALIZED AREA FOR 33 0LD MEN1
A G I N G P A T T E R N F O R M E N A S D E T E R M I N E D BY 135 U R I N E V A P O R S U B S T A N C E S D E T E R M I N E D A S D E S C R I B E D S U B S T A N C E S A R E L I S T E D IN O R D E R O F D E C R E A S I N G P R O B A B I L I T Y O F C O R R E L A T I O N WITH AGE
TABLE lI
56
TABLE 111 PREDICTED YEARS OF LIFE R E M A I N I N G FOR 30 MEN OF AGES U N K N O W N TO THE INVESTIGATORS AT THE TIME OF PREDICTION. THE AGES OF THE 7 Y O U N G E S T AND 8 OLDEST MEN WERE SUBSEQUENTLY REVEALED A N D ARE LISTED IN THE TABLE PERSON PREDICTED YEARS ACTUAL NUMBER RAGE LIFEREMAINING AGE 35.4 32.1 29.2 24.3 24.3 21.0 19.7 10.3 19.2 18.5 16~ 12.6 12,0 11.7 11.2 10.9
97 70 67 95 51 93 58 56 46 97 26 77 90
68 64 99 92 52 49 49 45 45 44 41 35 34 34 33 33
20 25 24 26 24 24
24
PERSON PREDICTED YEAR~ ACTUAL NUMBER RAGE LIFEREMAINING AGE 21 15 61 29 44 40 12 3 59 54 23 35 10 48
6.2 6.4 6.2 6.02 5,7 , 1.3 -2.7 -3.2 -3.7 -5.1 -13.0 -13.9 - I 5.6 -39.9
29 26 26 26 25 19 13 12 12 10 01 01 01 01
67
62 50 63 55 52 52 60
15 URINE SUBSTANCES OF 70 MEN LINEAR REGRESSION LINES
De /
r_
/
W 0
~ J 20 40 60 80 C H R O N O L O G I C A L AGE
,.
100
Fig. 10. Linear regression lines for the 15 substances with strongest diagnostic power for age in these experiments. The substance numbers on the lines correspond to those in Tables I and 11. Seven of the substances decrease with age and 8 of the substances increase with age.The regression lines were extrapolated to ages 0 and 100 years and then plotted to pass through those years at 1.0 solely for the purposes of this illustration. All data was accumulated for men in the age ranges shown on the figure.
57 expected from mass, simple sampling of the population at large. One sex was used, since our earlier experiments had shown strong sex-dependent patterns in humans which were not the subject of these experiments. A staggered alternation procedure for order of analysis of these samples was used to avoid possible systematic errors from machine drift. In addition, extensive experience with the analytical techniques assured their stability and reliability. Urinary amines, amides, and amino acids were determined by application of urine directly to the high pressure liquid chromatography column of an automated Durrum amino acid analyzer with a ninhydrin detector. Volatile urine compounds were analyzed as described in Ref. 15. After integration and peak matching a unique normalization procedure was applied to the peak areas. Each peak area was normalized by division by the sum of most of the matched peak areas in its individual chromatogram. Peaks with large areas were weighted less so as not to dominate the normalization. In this way systematic variations in kidney function and water excretion are far more effectively eliminated than with the usual procedures of timed samples or normalization to creatin~ne or other single substances. We have previously developed a graphical method for determination of the existence and extent of patterns which distinguish paired sets of samples. This method was used in these aging experiments. First, the non-parametric probability, P, of the null hypothesis that there is no quantitative difference between the young and old groups was calculated by means of a two-tailed Wilcoxon test [18] for each substance. The Wilcoxon test statistic was assumed to be asymptotically normal, corrected for ties and by a continuity correction of 0.5 according to Verdooren and Lehman, and corrected by the first term of the Fix and Hodges correction. All of the statistical and experimental procedures were selected so that any errors in the estimated values of P would always tend to make these values higher and too pessimistic rather than lower and too optimistic. Second the cumulative distribution functions of these calculated values of P were plotted as shown in Figs. 8 and 9. For the purposes of this calculation the men were divided into two age groups as shown in Fig. 10. If no aging pattern existed, the values of P would fall on the straight diagonal line shown in the figures. This has been the case in some other types of experiments which we have performed. In the case of age, however, these values show a very strong pattern. The asymptotes of the curved data lines indicate that about 30 urine amines, amides, and amino acids and about 30 urine volatile compounds show such strong age correlation that they can be experimentally demonstrated with fewer than 40 men in each group. These age correlations are shown in Tables I and II. These tables give the substance identity where known, the direction of the correlation, PAGE, expected and found numbers of substances at various values of P, normalized area, and age diagnostic powers, DPAcE, for each substance. The values of DPAGE were calculated as
58
LIFE EXPECTANCIES FOR THE MEN OF KNOWN AGE - -
T
---T-
r
--7
T
OLD GROUP
~3
T
YOUNG GROUP
ILl
>.
zILl o
0
10
20 30 40 50 60 YEARS OF REMAINING LIFE
70
Fig. I 1. Life expectancy distributions for the young and old groups of men used for the calculations illustrated in Figs. 13--17. The life expectancy of each individual was tabulated from his chronological age and U.S. Public Health Service compilations.The smooth curved lines were then drawn by interpolation.
described [17] and illustrated in Figs. 13--17. DPA6E is underestimated in these groups, because the overlap in life expectancies makes a perfect separation improbable. For this reason a corrected value, DP c, was also calculated as shown in Figs. 13--17.
URINE AMIDE VALUES OF 205 MEN GROUPED AND AVERAGED
u.
W I'L
O~
40
.a+
20
zv.
10
.d
°
50
30 20 10 YEARS OF REMAINING LIFE 40
10
Fig. 12. Normalized urine amide values for the 205 men grouped and averaged in six groups of similar age. The groups contained 38, 60, 53, 32, 13 and 9 men from young to old, respectively. The years of remaining life were calculated from actual ages of the men and U.S. Public Health Service compilations. The error bars are drawn for one standard deviation of the means.
59
Figure 10 shows linear regression lines for the 15 most strongly age diagnostic substances. Seven of these 15 substances decrease with age, while 8 of them increase with age. The diagnostic strength of these correlations is not only a function of age dependence, but also a function of contributions to the substance distribution functions from diet and the many other uncontrolled variables in this real-world sampling. A substance might be very strongly age-dependent and yet have a low diagnostic power, because its distribution function is broadened by other variables. Also, some of the chromatogram peaks contain two or more substances in comparable amounts which usually weakens their correlations. Life expectancy distribution functions for the two groups of men of known ages are shown in Fig. 11. These were calculated from U.S. Public Health Service statistics as applied to the age of each individual. The expectation values of the years of remaining life were then used to construct Fig. 11. Figure 12 shows, for example, a graph of the fourth most diagnostic peak as a function of years of remaining life for these men. The two curves in Fig. !1 overlap significantly; we are measuring physiological parameters which are expected to correlate best with physiological age rather than
AGE CLASSIFICATION WITH 5 URINE AMINES THAT DECREASE WITH AGE CORRECTLY CLASSIFIED YOUNG MEN Z
g,i :S
30
20
I
I
10
0
I
Z
klJ
DP = 0.63 DPc = 0.69
30 _.1
0 o
_1
o 10Q
ILl
UJ m
U,.
u.
2o ¢n ,<
.J
20 ~
-1
J
FW
~1o W
3o ~
0
o
z 0
10
20
30
INCORRECTLY CLASSIFIED YOUNG MEN
Fig. 13. Age classification with 5 ninhydrin positive urine compounds which decrease with age. Those with the highest diagnostic powers from Table i have been used. By means of their diagnostic coefficients, RAGE, the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
60
chronological age; and we have only chronological ages available for the human subjects of the experiment. Therefore, we cannot expect this technique to show perfect chronological age separation between the two groups of men corresponding to a DPAcE equal to 1.00. Figure 13 shows a diagnostic power graph utilizing the equally weighted 5 urine amines which decrease with age and show the highest DPAG E. The diagonal straight line is the line expected with uncorrelated substances. A point in the origin corresponds to DPAc E equal to 1.00. The smooth curved line represents a perfect result taking into account the overlap of the distribution functions of life expectancy as shown in Fig. 11. The stepped function is the experimental result. DP and DP~ are the fractions of separation achieved experimentally compared to a point in the origin and compared to the curved line respectively. The computerized calculation determines RAC E [14] for each sample and thereby places each individual on a linear axis between the young and old patterns. It then separates the individuals at all possible points along this axis and tabulates and
AGE CLASSIFICATION OF 7 URINE AMINES T H A T INCREASE WITH AGE CORRECTLY CLASSIFIED YOUNG MEN 30
20
10
0
z
Ll.I
DP = 0.78
30
/
;
.J
o
I.U
20 .-I ¢.) ._/ w
30
o
~"
z 0
10
20
30
INCORRECTLY CLASSIFIED YOUNG MEN Fig. 14. Age classification with 7 ninhydrin-positive urine compounds which increase with age. Those with the highest diagnostic powers from Table 1 have been used. By means of their diagnostic coefficients, RAGE, the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
61
graphs tile errors as shown. To avoid bias, the average patterns were recalculated for each individual excluding that individual from the averages. Where two experiments are available (as in the blind experiment described below and in Table !II), none of the test samples are included in the average patterns. The series of Figs. 13--17 shows the gradual increase of reliability of this technique as more substances are added to the calculation. The 5 decreasing amines have DPc of 0.69 (Fig. 13). This is increased to 0.91 when the 7 increasing amines are included (Fig. 15) and to 0.93 when the 3 most diagnostic urine vapor compounds are included (Fig. 17). Figures 14 and 16 show DPc for the increasing amines and for the urine vapor substances alone. A pattern for aging was calculated for these 15 substances and evaluated by means of the diagnostic coefficient, RACE, as described in Ref. 14. By this method the individuals can be positioned along a linear axis of the type illustrated in Fig. 4. RA~E values for the 66 men after separation into five groups with similar life expectancies are shown in Fig. 18. Finally each of the 30 men whose ages had not been given to us were placed on the linear aging axis by means of their individual RA~E as determined from urine
AGE CLASSIFICATION WITH 12 URINE AMINES CORRECTLY 30 z
\
CLASSIFIED YOUNG MEN 0 20 10
~
~
Z
ILl
kkl
30 •J
DP = 0 . 8 3
~e~Oo ~'~ "1
~
=
J
10 °
¢3 Ill
0.91
m g.
20 ~ .J
1-10
o
1-
=
30 ~
o o
0 o
zD
0 10 INCORRECTLY
20
30
CLASSIFIED YOUNG MEN
Fig. 15. Age classification with 12 ninhydrin-positive urine compounds which increase and decrease with age. Those with the highest diagnostic powers from Table I have been used. By means of their diagnostic coefficients, RAGE. the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation. a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
62
AGE CLASSI FICATION WITH 3 URINE VAPOR SUBSTANCES CORRECTLY CLASSIFIED YOUNG MEN 30 20 10 0 z ~: 30 ,.J 0 u.J
u. 20 .,.d
W re"
3O
Q Z
.... L
o
10
20
30
I N C O R R E C T L Y C L A S S I F I E D YOUNG MEN
Fig. 16. Age classification with 3 urine vapor substances.. Those with the highest diagnostic powers from Table 11 have been used. By means of their diagnostic coefficients, RA(;I, the men have been placed on a one-dimensionalaxis extending from the young to old aging panerns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DPc are explained in the text.
analysis, the fifteen substances, a n d calculations identical to those described above. The successful results of this blind experiment along with the s u b s e q u e n t l y revealed ages of 15 of the men are given in Table Ill. These 15 included, according to A. Cherkin, the seven youngest a n d eight oldest members of the g r o u p of 30. It must be emphasized that these are results from the analysis of substances picked only by ease of analysis, not by interest to aging researchers, and samples o b t a i n e d
casually from h u m a n s without special controls. Sample controls can n a r r o w the d i s t r i b u t i o n functions substantially as can sampling of genetically h o m o g e n e o u s p o p u l a t i o n s a n d other precautions. F o r example, in Fig. 19 (from Ref. 12) the RAc;E for six sets of 24 enzymatically digested Drosophila melanogaster, four each from each day of six consecutive 6-day periods, are given for only two correlations the increase in gtutamic acid and decrease in glutamine which occurs with age. This experimental result is remarkably exact, but it is also of little immediate applicability to the q u a n t i t a t i v e m e a s u r e m e n t of physiological age in large p o p u l a t i o n s of h u m a n s . A b o u t 60 substances or a b o u t one-third of the substances measured in these h u m a n experiments were f o u n d to be age correlated, but only 15 of these have been
63
AGE CLASSI FICATION WITH 12 URINE AMINES AND 3 URINE VAPOR SUBSTANCES CORRECTLY CLASSIFIED YOUNG MEN 30 20 10 0
Z I,U
a 30 .J
o o
j 10 O
DPc = O.93
W
g.
.J
,-I
~1o ua n-.
tll
O
ii
z
~3O
O tj
0
10 20 30 INCORRECTLY CLASSIFIED YOUNG MEN
Fig. 17. Age classification with 15 urine compounds: 5 ninhydrin positive compounds that decrease with age; 7 ninhydrin positive compounds that increase with age, and 3 urine vapor compounds. Those with the highest diagnostic powers from Tables I and Ii have been used. By means of their diagnostic coefficients, RAGE, the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
LIFE REMAINING FOR 66 MEN VS. DIAGNOSTIC COEFFICIENT OF AGE FOR 12 URINE AMINE AND 3 URINE VAPOR SUBSTANCES I
I
l
I
I
I
I
I
¢3 < +30 IL
o I.- +20 z ILl +10 m ii
u. 0 uJ O ¢J -10 ¢J I'-
~-20 Z
~ < a
_30 0
I
10
20 3O 4O YEARS LIFE REMAINING
I
I
5O
Fig. 18. The diagnostic coefficients, RAGE, for the 15 substances of the 66 men used in the calculations shown in Figs. 13--17. The men have been grouped into five groups of similar age. The ages and diagnostic coefficients were then averaged to give the points shown on this figure. The error bars are drawn for one standard deviation o f the means.
64
AGING
1
-10
2
DROSOPHILA
3
1 -5
I
MELANOGASTER
4
i 0
5
I 5
6
1(]
DIAGNOSTIC COEFFICIENT, RAGE FOR GLUTAMICACID AND GLUTAMINE AFTER ENZYMATICDIGESTION Fig. 19. Values of the age diagnostic coefficient, RAG E, of glutamic acid and glutamine for six sets of 24 Drosophila melanogaster, each representing consecutive 6-day periods. The groups are consecutively numbered with 1 designating days 1--6, 2 designating days 7--12 and so on. This figure is reproduced from Robinson, A . B , Willoughby, R. and Robinson, L.R., /~\x-p. Gerontol., I I (1976) 113.
included in the calculations reported here. Inclusion of the next few most diagnostic substances has no significant effect on the results. Thereafter, the DPAGE decreases when additional substances are included in the calculation. These additional substances include some independent age information, but their distribution functions also contain so much noise from other effects such as diet that they do not make a positive contribution to these calculations. STANDARDIZED HUMAN SAMPLES.
The principal limitation of these experiments and of advances in clinical chemistry in general is the lack of standardized human samples with which to test and calibrate the techniques. Herein, for example, we have used samples characterized as to chronological age, although our interest is in physiological age. We do not know of the existence anywhere of a set of human urine samples which have been carefully and quantitatively characterized as to the physiological age of the subjects. We have previously proposed (see for example Ref. l l) that a national sample bank of human blood, urine, breath, and other tissues be established. This should be done by the sampling of approximately 50 000 healthy Americans at least at 6-month intervals with the samples stored at - 7 6 ° C or lower in a bank of small freezers with appropriate equipment duplication and sample redundancy. Over the course of a 5-year period a statistically useful number of these 50 000 people will fall ill of various diseases down to and including the incidence of multiple sclerosis. The bank will then contain prospective samples for the individuals and for the group which can be used to calibrate machines which may be able to detect diseases before the individuals fall ill. Medicine could then combat the probability of illness, rather than illness itself. This bank would also allow the use of individuals as their own controls thereby improving the resolution of clinical techniques. This advance in preventive medicine would itself have a positive effect on human longevity. However, direct effects on aging research would be even more pronounced.
65
As the years pass and the 50 000 individuals gradually die, the bank of samples would contain an increasingly valuable set of samples from individuals whose years of life remaining would then be known. Subdivided with respect to cause of death, these samples, characterized as to true individual physiological age, would be very valuable. The samples could be used to calibrate techniques for the quantitative analysis of human physiological age. Since they would be rigorously characterized and available to various experimenters from a single source, the samples could also be used in comparative experiments to determine which techniques are most useful for the measurement of age. This sample bank should probably be started uniquely for this purpose and should be of the highest possible quality. Efforts to salvage old sample sets from other projects would probably lead to expensive, yet suboptimum results. It is important to emphasize that the bank must contain samples, not the results of analyses. It would be tragic to undertake this project with limitation to one or more particular analytical techniques which happen to now be available and popular. Surely those techniques would be obsolete by the time results were obtained. If, however, the samples had not been retained, no further advance would be possible. If we could make only one contribution to the advancement of the use of modern technology for the improvement of human health, the creation of this sample bank of body fluids from well individuals would be our choice. If this bank had been created when we first formally proposed it 14 years ago, substantial improvements in clinical medicine and in empirical aging research would already be in evidence. CONCLUSION
We conclude that there exists a very strong systematic metabolic pattern for human age, and we estimate that these techniques should be able to be refined with relative ease to measure the physiological age of individual human subjects to an accuracy of 3 years or less. This would permit the measurement of physiological age in a group of 300 individuals to within an accuracy of 2 months or 1300 individuals to within an accuracy of 1 month. Rate of aging experiments could then be performed to within an accuracy of about 10"/,,over a period of I year. As experience is gained, this should be improvable by about one additional order of magnitude which would permit smaller, shorter, or more accurate experiments. Improvements can also be expected from the separation of individuals into types. We observed for example, in an experiment in which we measured 60 urinary amines, amides, and amino acids in the urine of 1000 newborn infants (Robinson, A.B. et al., unpublished data), that the human distribution functions of amounts of many of these substances were not unimodal. Instead they were bimodal and multimodal. We expect this to be generally true, so that discrete typing of individuals will eventually markedly increase the resolution of these techniques. Improvements can also be made by particular constraints such as dietary control.
66
We have already demonstrated that with strict dietary control the urinary patterns become so unique that computerized fingerprinting of single individuals with a high degree of accuracy can be carried out. Strict dietary control, however, sharply limits the utility and breadth of applicability of these techniques. Refinement need not make use of these particular analytical techniques. Any method which measures about 200 metabolic substances quantitatively should be sufficient. (Some methods such as field ionization mass spectrometry have the potential to measure 2000 or more substances within a few minutes or less.) Refinement, however, absolutely requires that a sample set be available in order to calibrate the technique. Until this sample set is carefully gathered, stored and made generally available, this technique and many other potential analytical advances in human health will probably remain research curiosities rather than proper contributors to the quality of human life. We emphasize the following: (1) About one-third or more of the metabolic constituents of human body fluids correlate systematically with age, as do the constituents of flies and mice. For practical, empirical purposes, therefore, analytical techniques for profiling should emphasize those substances which are easy and inexpensive to measure quantitatively. Any sufficiently large set of casually selected metabolic substances contain within their amounts the necessary information as to individual physiological age. (2) Simple computerized techniques exist for the test and utilization of data from these profiling experiments. We have also used more sophisticated pattern recognition techniques, but have generally found them cumbersome and without concomitant additional utility. 'Eyeballing' chromatograms or mass spectra is, however, essentially without value. Proper non-parametric tests of reliability on statistically useful sets of analyses must be performed. A simple procedure for doing this is illustrated herein. (3) The principal limitation to this work is in sampling - - not in analysis or computation. Although other analytical techniques can be utilized and many interesting reports written regarding the analytical profiling of age, further real advances in this effort will probably not be possible until a proper prospective sample bank is established. REFERENCES 1 2 3 4 5 6
C.S. Minot. The problem of age, growth, and death. Pop. Sci. Month(v. 71 (19071 481. D. Harman, Aging: a theory based on free radical and radiation chemistry. J. Geromol.. II (1956! 298. J. Bjorksten, A c o m m o n molecular basis for the aging syndrome. J. Am. Geriatr. Sot., 6 ~1958) 740. L. Orgel, The maintenance of the accuracy of protein synthesis and its relevance to ageing. Pro~, Nat. Acad. Sci. USA.. 49 (19631 517. Z,A, Medvedev, The nucleic acids in development and aging. Adv. Gerontol, Res., 1 (1964) 181. B.L. Strehler, Code degeneracy and the ageing process: a molecular genetic theory of aging. Proc. 7th Congr. Gerontol. Vienna. (1966) 177.
67 7 8 9 10
11 12 13 14 15 16 17 18
L. Walford, The bnmunoh~gic'al Theo O" q/Aghlg, Williams and Wilkins, Baltimore, 1969, pp. 70. R.W. Hart and R.V. Setlow, Correlation between DNA excision repair and life span in a number of mammalian species. Proc. Nutl. Acad. Sci. USA, 71 (1974) 2169. J.L. Bada, Aspartic acid racemization in tooth enamel l¥om living humans. Proz'. NatL Ac~ul. Sc'i. USA, 72 (1975) 2891. A.B. Robinson, J.H. McKerrow and P. Cary, Controlled deamidation of peptides and protcins: an experimental hazard and a possible biological timer. Pro~'. Natl. Acad. Sc'i USA. 66 (1970) 753. See also A.B. Robinson, Proc. Natl. Acad Sci.. 71 (1974) 885 and A.B. Robinson and C. Rudd, Deamidation of glutaminyl and asparaginyl residues in peptides and proteins. Curr. Topic's Celluhn" Reg., 8 (1974) 247. A.B. Robinson, Molecular clocks, molecular profiles, and optimum diets: three approaches to the problem of aging. Mech. Ageing Dev.. 9 (1979) 225. A.B. Robinson, R. Willoughby and L.R. Robinson, Age dependent amines, amidcs and amino acids residues in Drosophila mekmogtlster. Exp. Gerontol.. II (1976) 113. A.B. Robinson, M. Weiss, W.E. Reynolds and L.R. Robinson, Use ¢~l'Mass Spectrometrv.[br Orthomolecular Diagnosis, 23rd Ann. Conf., Mass Spect., Houston, Texas (1975) 182. A.B. Robinson, H. Dirren, A. Sheets and R. Lundgren, Quantitative aging pattern in mouse urine vapor as measured by gas-liquid chromatography. Exp. Gerontol. II (1976) II. A.B. Robinson, D. Partridge, M. Turner, R. Teranishi and L. Pauling, An apparatus for thc quantitative analysis of volatilc compounds in urinc. J. Chroouttograph.r, 85 (1973) 19. A.B. Robinson, Antioch Review. 1981, p. 383. A.B. Robinson and F.C. Westall, The use of urinary aminc measurement for orthomolccular diagnosis of multiple sclerosis. J. Orth. Psych., 3 (1974) 1. F. Wilcoxon, Biometrics, I (1945) 80: Lehman, S.Y., J. Am. Statist. Assoc.. 56 (1961) 293; L.R, Vcrdooren, Biometrika, 50 (1963)177: and E. Fix and J.L. Hodges, Ann. Math. Stcttist., 26 (1955) 301.
47
Elsevier Scientific Publishers Ireland Ltd.
QUANTITATIVE M E A S U R E M E N T OF HUMAN P H Y S I O L O G I C A L AGE BY P R O F I L I N G OF BODY FLUIDS AND PATTERN R E C O G N I T I O N
A R T H U R B. ROBINSON and L A U R E L E E R. ROBINSON* Oregon Institute of Science and Medicine. 2251 Dick George Road. Cave Juncthm. Oregon (U.S.A.) (Received November 9th, 1990)
SUMMARY
Quantitative correlations with human age are demonstrated for 60 substances from a group of 200 substances measured in the urine of 235 men. Simplified pattern recognition calculations are used to combine these correlations into patterns of human age and to demonstrate their utility for the quantitative measurement of human physiological age and aging rate. The empirical use of these techniques for the extension of human life-span and diminution of human suffering from degenerative diseases is discussed. Current experimental limitations of this method are demonstrated and evaluated. The application of these techniques can form the basis for a significant advance in the quality of human life.
Key words: Aging; Human; Analysis; Pattern; Urine; Quantitative
There are many hypotheses regarding the primary cause of the observed intrinsic life span of organisms [1--10]. It is to be expected that the eventual resolution of this unsolved problem in molecular biology will lead to methods of altering the primary causes of aging and thereby the life span of organisms including humans. These alterations may be expected to be of the two types illustrated in Figs. I--3 [11]. First, the death rate of humans during the years much earlier than their intrinsic life span will be reduced. Second, the intrinsic life span itself will be increased. Correspondence to: A.B. Robinson, Oregon Institute of Science and Medicine, 2251 Dick George Road, Cave Junction, OR 97523, U.S.A. *Laurelee Robinson died on November 12,1988. She saw and approved all of the figures and tables in this paper. She did not read the final manuscript. 0047-6374/91/$03.50 Printed and Published in Ireland
© 1991 Elsevier Scientific Publishers Ireland Ltd.
48
100%'
LLI
80%
>
\
60% LU
a.
0
40%
LLI n
2O% 20
40
60
80 100 YEARS OF AGE
120
140
Fig. I. Aging curve for a population comprised of American males as calculated from life expectancycompilations of the U.S. Public Health Service.This figure is reproduced from Robinson, A.B., Mech. AA,ein A, Dev., 9 (1979) 225.
It should, however, be possible to extend h u m a n life spans by means of empirical e x p e r i m e n t a t i o n even prior to the a d v e n t of a t h o r o u g h u n d e r s t a n d i n g of the molecular biology of h u m a n aging. O b s e r v a t i o n s o f a n i m a l s and h u m a n s suggest that substantial progress o f the type illustrated in Fig. 2 a n d modest progress of the type illustrated in Fig. 3 should be possible.
100%
80% > ,~ 60% LLJ
0 LU a.
40%
20%
20
40
60
80 100 YEARS OF AGE
120
140
Fig. 2. Example of a change in the human aging curve through improvements in diet and other livingconditions, but with no change in intrinsic life span. This figure is reproduced from Robinson, A.B.. Mech. Ageing Dev.. 9 (1979) 225.
49
100% 80% > -- 60% <~ LU
, t
UJ
40% 0
uJ
o- 20%
20
40
60
80 100 YEARS OF AGE
120
140
Fig. 3. Example of a change in the h u m a n aging curve through improvements in diet and other living conditions and through an increase in the intrinsic life span. This figure is reproduced from Robinson, A.B., Mech. Ageing Dev.. 9 (1979) 225.
The principal impediment to this empirical progress is the fact that the humans carrying out the experiments have life spans of comparable length to those of the human subjects which they are studying. If carried to completion, a single rigorous experiment on human life span as a function of, for example, diet or environment
100%
LIFE REMAINING
I
0%
.
0
I
PROBABILITY OF ILLNESS X
WELL
,,
I
1.0
SICK
SEVERITY OF ILLNESS
Fig. 4. Representative axes of physiological age, probability of a specific illness, and the severity of a specific illness. It is the quantitative determination of the positions on these axes of specific individual humans, by means that are low cost and generally applicable to all well individuals at frequent intervals, which is the goal of our work on the physiological profiling of human body fluids.
50 could require as much as 50 to 80 years. Even if this is reduced to 10 to 20 years in some cases by clever experimental design, a complete series of experiments will still proceed too slowly to be of benefit to humans who are now living. In the absence of real progress and as a result of the understandably high human interest in this subject, many sensible but unproved aging hypotheses have given rise to fads and cults especially as pertain to diet. The impediment to human aging research of long experiment times would be removed if an objective and quantitative means for determining human physiological age with reasonable accuracy could be developed. This would allow empirical experiments on aging rate to be carried out with experiment times in groups of humans of a few months per experiment or less. Moreover, quantitative analysis of human age would allow the monitoring of the rate of aging in single individuals, so that biochemical individuality and individuality of particular environmental circumstances could be taken into account in devising conditions that minimize the rate of aging in each individual. We have long been interested in the development of a procedure for the quantitative measurement of human physiological age [l 1--14]. This interest has been a part of our general interest in the quantitative measurement of human health especially with respect to degenerative disease. If by means of inexpensive quantitative analysis the state of each person could be determined on quantitative axes as illustrated in Fig. 4, then their productive and actual life spans could be extended; their illnesses could be more effectively cured; and their health could be guarded by preventive measures which combat the probability of disease rather than disease itself. It is an essential obligation of scientists to do all that they can to optimize the quality of all human life. This optimization requires, in part, the minimization of disease and the maximization of life span for the entire human population. Techniques of the sort illustrated in Fig. 5 and discussed below have substantial potential to meet these parts of that obligation. We have limited ourselves to those techniques which have potential, in mass application, to quantitatively measure the amounts of a large number of substances in human body fluids at a cost of about $ l0 or less for an entire analysis including its computerized interpretation. This limitation assures that the techniques developed will be available to most people rather than a wealthy few. Having measured within this restriction as many parameters as possible, we then have attempted to extract as much information from these measurements as possible by simple computerized pattern recognition techniques. This is a departure from the usual approach of developing observational techniques for those substances thought to be of importance to a particular hypothesis or theory. Fundamental science is more entertaining. We have, in our work on deamidation and molecular clocks [10,11], enjoyed this, too. In fact, a side benefit of the profiling work has been the discovery of a few correlations which happen by
51
Fig. 5. One aspect of the optimization of the quality of human life.
chance to be relevant to basic research. Nevertheless, choice of substances by economics and simplicity of analysis rather than by hypothesis is an essential part of rapid progress in the practical profiling of human health. Our work has comprised the use of ion-exchange chromatography, gas-liquid chromatography, and direct mass spectrometry, including chemical ionization and field ion mass spectrometry (without chromatographic separation), of samples of urine, blood, and breath from humans and mice. We also have carried out some work on the whole body compositions of fruit flies [ 12].
HUMAN URINARY AMINES - AGING
10
20
30
40
50
SUBSTANCE NUMBERS
Fig. 6. An example chromatogram of the amines, amides, and amino acids in human urine as used in these experiments. The base line is computer drawn. The darkened strips along the bottom axis show peaks which the computerized procedure has decided to combine for computational analysis.
52
For the most part, however, this work has been restricted to humans. Since this approach is empirical in nature and has an immediate practical objective, we prefer to avoid extrapolations between animals and humans in the interpretations of results. In a highly automated laboratory (see for example Ref. 15), approximately 20 000 analyses of human body fluids including the quantitative analysis of over 2 million amounts of substances have been carried out by us and under our supervision (see for examples Refs. 12, 13, 15 and 16). The primary focus of this work has been on states of disease. A small part was devoted to the measurement of organismic age (see Refs. 12 and 14) in mice and fruit flies. We report herein a few experiments which we carried out to test the measurement of chronological and physiological age in humans by urine analysis. Figures 6 and 7 illustrate the analytical techniques used in these urine analysis experiments. Graphs of this sort were not, however, used in the experiments. The detector voltages as a function of time from the analytical machines were sampled by multiplex, collected on magnetic tape, and transferred to a calculation computer. There, by means of automatic computer programs which we developed, a baseline was fitted and subtracted: the peaks were integrated: retention time variation was eliminated and the peaks were matched within the entire experimental set: the peak areas were normalized in order to remove systematic variations in kidncy function or water elimination: the experiments were evaluated by non-parametric statistics and by diagnostic power calculations; the diagrams shown in Figs. 8, 9 and 13 17 were automatically drawn; and the values given in Tables 1 and I1 were automatically calculated. 'Diagnostic powers', DA, were calculated according to the method of Robinson and Westall [17] and the "diagnostic coefficients', RA, were calculated as described in Ref. 14. Casual urine samples and chronological ages of 205 men were obtained from Dr. A. Cherkin of the Veterans Administration hospital in Sepulvida, California. In
H U M A N URINE VAPOR SUBSTANCES
10
20
30
40
50
60
70
AGING
80
90
100
110
120
130
SUBSTANCENUMe£RS
Fig. 7. An example chromatogram of Ihc volatile c o m p o u n d s in human urine as used in these experiments. This analytical technique is described in detail in Rel'. 15. The base line is computer drawn. Thc darkened strips along the bottom axis show peaks which the computerized procedure has decided to combine for computational analysis.
53
URINE AMINES YOUNG MEN (35) VS OLD MEN (35)
50 F o-" oo
,< g
3oF !
o 20 - .z C¢ MJ
!
z
0
0.2
0.4 0.6 0.8 MAXIMUM PAGE
1.0
Fig. 8. Cumulative distribution function of the probabilities of correlation with human age of 51 urinary amines, amides, and amino acids.Without correlation the dots would fall along the diagonal line. About 30 of these substances are seen here to be age correlated.
a d d i t i o n , casual urines for 30 men were o b t a i n e d from the same source wherein the ages o f the individuals were kept at S e p u l v i d a a n d not c o m m u n i c a t e d to us until after the c a l c u l a t i o n s shown in T a b l e 111 had been c o m p l e t e d . Casual urine without dietary o r o t h e r c o n t r o l was used, so that the results w o u l d reflect those to be
URINE VAPOR SUBSTANCES YOUNG MEN (33) VS OLD MEN (33)
120 ~ILl
I--
,ooi
~
80 ~
0
60-
iv" W
:D. Z
40~20 h V
0
I
0.2
t
____
l
0.4 0.6 MAXIMUM PAGE
I
0.8
1.0
Fig. 9. Cumulative distribution function of the probabilities of correlation with h u m a n age of 135 urine vapor substances. Without correlation the dots would fall along the diagonal line. About 30 of these substances are seen here to be age correlated.
21 48 3 11 4.6 17 23
3O 36 16 14 13 42 2 15 33 39
28 18 10 34 44
6 1 45 9 8 12 22 7 5 20 19 24
SUBSTANCE NUMOER OLD OLD YOUNG YOUNG YOUNG YOUNG OL D YOUNG OL O OLD OLD OLD OLD OLD OLD OLD YOUNG 0 LD OLD OLD YOUNG YO UNG YOUNG OLD YOUNG OLD YOUNG YOUNG YOUNG YOUNG OLD OLD OLD OLD
PHOSPHOSERINE HiSTIOiNE 3 ASPARAGINE + G LUTAMINE 3 SERINE GLYCINE -
THREONINE 0 L UTATHI 0 N E METHIONINE CYSTATRIONE -
-
CYSTINE
~-AMiNO BUTYRIC ACID ALANINE 3 LYSiNE
PHOSPHOETHANOLAMINE VALINE
~-AMINO BUTYRIC ACIO ETRANOLAMINE -
TAURIRE GLUTAMIC ACID 3 ORNITHINE LEUCINE
HIGHER GROUP
ASPARTIC ACID 1
STANDARDSUBSTANCE WITHIDENTICAL ELUTION TIME
10- 5 10-5 10-4 10-4 10-6
1.5 x 1.6 x 1.6 x 1.6 x 1.7 x
8.2 x 9,5 x 9.8 x 1.4 x 10-1 10-1 10-1 10-1 10-1
10-2 10- 2 10-2 10-1
0.5 x 10-2 7.8 x 10-2
6.2 x 10-2 6.3 x 10-2 6.3 x 10-2
6,1 x 10-2
2.8 x 10-3 7.1 x 10-3 0.9 x 10- 3 1.6 x 10-2 2.5 x 10-2 3.4 x 10-2 4.2 x 10-2
2~5 x 8.3 x 3.0 x 4.0 x 9.3 x
4.8 x 10-7 2.9 x 10-6 4.1 x 10-6 1.2 x 10-6 1.3 x 10-5 2.2 x 10-5
< 1 0 -8
PAGE
7.4 7.6 0.0 6.3 0,5
4.2 4.0 6.0 7.2
3.3 4.0
2.7 3*2 3,2
2.0
0,14 0.30 0.45 0.00 1.3 1.7 2.1
0.9912 0.0042 0.016 0.026 0,047
0.000025 0.00015 0.00021 0.00061 0.00060 0.0011
NUMBER OF SUBSTANCES EXPECTED WITH P ~PAGE
1.71 -+0.11 4.00 -+0.33 3.63 -+0.10 24.0 -+2.2 0.43 -+1.03 11.10 -+0.42 3.40 -+0.24 1.00-+0.06
20.2 -+1.7 10.49 + 1.19 0.79 -+0,42 3.91 -+0.11 0.93 -+0.04 40.3 -+1,'6
31 32 33 34
28 29 30
23 24 25 26 27
2.92 ± 0.27 3.42 ± 0.71 52.8 -+4.9 6,03 -+0.23 2,75 -+0,12 3.77 + 0.19 0.62 ± 0.16
35.9 +2.2 2.34 -+0,10 1.71 -+0.31 44.0 +-5.3 6,21 -+0.27 3,12 -+0.20 4.08 ± 0.24 6.90 -+0.19
2,10 -+0,11 1.74 -+0.08 5.63 -+6.32 1.59 + 0,10 1.16 -+0,12 1.40 -+0.09 1,74 -+0.69 1.43 ± 0.06 4.09 ± 0.21 1.U -+0.29 1.64+-0.15 1,16 -+0.06 1.47 -+0,07 3.76 + 0.21 4.09 -+0,11
13 14 15 16 17 18 19 20 21 22
2L4 -+2.1 97.0 ±7.4 3.16 ± 0.11 16.43_+1.21 20.1 -+0.9 6.05 -+0.30 2.82 -+0,10 1.50 -+0,08
0.68
0.10
0 .06 0.45 0,47 0.39 0.35 0.47 0,41 0.39 0.28 0.18 0.24
6.23 -+0,16 9.44 ± 0.27 43.0 "+2.3 27,3 -+1,6 10.5 -+0.9 59.2 +4,8 3.67 -+0.11 9.00 -+0.62 30.2 ± 0,8 7.00 -+0.21 3,48 -+0.17 1,91 -+0.10
4.70 -+0.19 7.46 "+0.27 50.2 -+1.0 40.0 -+2,0
I 2 3 4 5 6 7 0 9 10 11 12
AGE DIAGNOSTIC POWER
AVERAGE NORMALIZED AREA FOR 36 OLD MEN2
AVERAGE NORMALIZED AREA FOR 35 YOUNG MEN 2
NUMBER OF SUBSTANCES FOUND WITH P~PAGE
A G I N G P A T T E R N FOR MEN AS D E T E R M I N E D BY 51 N I N H Y D R I N - P O S I T I V E U R I N A R Y A M I N E S , AMIDES, A N D A M I N O ACIDS. T H E SUBSTANCES A R E LISTED 1N O R D E R O F D E C R E A S I N G P R O B A B I L I T Y OF C O R R E L A T I O N WITH A G E
TABLE I
ALLYL ISOTHIOCYANATE 2-HEPTANONE
42
109 134 07
3-PENTANONE 2-NONED.4-ONE 3-HEPTANONE 2-NONANONE CARVONE
16 01 41 83 117 38 124
44
3-METHYL-2-BUTANONE; BENZENE; 2-METHYLTETRAHYOROFURAN
2-PENTANONE
14 77
12
2-OUTANONE; 2-METHYLFURAN; 3.METHYLFURAN
STANDARD SUBSTANCE WITH IDENTICAL ELUTION TIME
8
87
SUBSTANCE NUMBER
OLD OLD YOUNG
YOUNG
YOUN6 YOUN6 YOUNG OLD YOUNG YOUNG YOUNG OLO
YOUNG
YOUNG OLD
YOUNG
YOUNG
HIGHER GROUP
1.0 x 10-2 1.7x 10-2 1.9 x 10-2
1.0 x 10-3
2.2 x 10-3 2.4 x 10-3 2,4 x 10-3 3.0 x 10- 3 4.1 x 10-3 5.5 x 10-3 7n x 10-3 0.0 x 10.3
1.3x 10-3
2,4 x 10-4 4,4x 10-4
1,4 x 10-4
2.9 x 10-6
PA6E
1,4 2.3 2.6
1.4
0.30 0.33 0.33 0,,47 O.U 0.75 0.04 1.2
0,18
0,032 0.0H
0.019
0.00427
NUMBER OF SUBSTANCESEXPECTEO WiTH P
16 16 17
14
E 7 0 9 10 I1 12 13
6
3 4
2
1
NUMiER OF SUBSTANCESFOUND WITH P
-+7.5
0.248 -+O.N4 0.037 + 0.906 0,431 + O.OS2
5.26 +O.H
1.14 +0.12 0.740 + 0 M S IJHI -+2.69 0.155+0.014 0,120 +0.013
0.223+ 0.018
1.59 +0.11
0.826+-0.073
1.77 ±0.11
10,4 +2.0 0.400 ± 0.027
41.9
0.323 ±0.038
AVERA6E NORMALIZEO AREA FOR 33 YOUNG MEN1
±0,46 0210 +0.017 0,0§7 ± 0,0BS 0266 -+0.037
3.04
0A16 + 0.036 I . N ±0JO 0.1U +0.010 2.00 2 + 0.74 0.96 ± O.0OS 2.16 ± 1.66 0.106 +0.017 0.161 +-0.012
124 -+0,11
0.5 ±2,0 0.975 + 0.123
±2.4
0.104 -+0.012 15.9
15. T H E
0.30
0.47 0,4a
0.48
0.02
AGE OlAGNOSTIC POWER
IN R E F E R E N C E
AVERAGE NORMALIZED AREA FOR 33 0LD MEN1
A G I N G P A T T E R N F O R M E N A S D E T E R M I N E D BY 135 U R I N E V A P O R S U B S T A N C E S D E T E R M I N E D A S D E S C R I B E D S U B S T A N C E S A R E L I S T E D IN O R D E R O F D E C R E A S I N G P R O B A B I L I T Y O F C O R R E L A T I O N WITH AGE
TABLE lI
56
TABLE 111 PREDICTED YEARS OF LIFE R E M A I N I N G FOR 30 MEN OF AGES U N K N O W N TO THE INVESTIGATORS AT THE TIME OF PREDICTION. THE AGES OF THE 7 Y O U N G E S T AND 8 OLDEST MEN WERE SUBSEQUENTLY REVEALED A N D ARE LISTED IN THE TABLE PERSON PREDICTED YEARS ACTUAL NUMBER RAGE LIFEREMAINING AGE 35.4 32.1 29.2 24.3 24.3 21.0 19.7 10.3 19.2 18.5 16~ 12.6 12,0 11.7 11.2 10.9
97 70 67 95 51 93 58 56 46 97 26 77 90
68 64 99 92 52 49 49 45 45 44 41 35 34 34 33 33
20 25 24 26 24 24
24
PERSON PREDICTED YEAR~ ACTUAL NUMBER RAGE LIFEREMAINING AGE 21 15 61 29 44 40 12 3 59 54 23 35 10 48
6.2 6.4 6.2 6.02 5,7 , 1.3 -2.7 -3.2 -3.7 -5.1 -13.0 -13.9 - I 5.6 -39.9
29 26 26 26 25 19 13 12 12 10 01 01 01 01
67
62 50 63 55 52 52 60
15 URINE SUBSTANCES OF 70 MEN LINEAR REGRESSION LINES
De /
r_
/
W 0
~ J 20 40 60 80 C H R O N O L O G I C A L AGE
,.
100
Fig. 10. Linear regression lines for the 15 substances with strongest diagnostic power for age in these experiments. The substance numbers on the lines correspond to those in Tables I and 11. Seven of the substances decrease with age and 8 of the substances increase with age.The regression lines were extrapolated to ages 0 and 100 years and then plotted to pass through those years at 1.0 solely for the purposes of this illustration. All data was accumulated for men in the age ranges shown on the figure.
57 expected from mass, simple sampling of the population at large. One sex was used, since our earlier experiments had shown strong sex-dependent patterns in humans which were not the subject of these experiments. A staggered alternation procedure for order of analysis of these samples was used to avoid possible systematic errors from machine drift. In addition, extensive experience with the analytical techniques assured their stability and reliability. Urinary amines, amides, and amino acids were determined by application of urine directly to the high pressure liquid chromatography column of an automated Durrum amino acid analyzer with a ninhydrin detector. Volatile urine compounds were analyzed as described in Ref. 15. After integration and peak matching a unique normalization procedure was applied to the peak areas. Each peak area was normalized by division by the sum of most of the matched peak areas in its individual chromatogram. Peaks with large areas were weighted less so as not to dominate the normalization. In this way systematic variations in kidney function and water excretion are far more effectively eliminated than with the usual procedures of timed samples or normalization to creatin~ne or other single substances. We have previously developed a graphical method for determination of the existence and extent of patterns which distinguish paired sets of samples. This method was used in these aging experiments. First, the non-parametric probability, P, of the null hypothesis that there is no quantitative difference between the young and old groups was calculated by means of a two-tailed Wilcoxon test [18] for each substance. The Wilcoxon test statistic was assumed to be asymptotically normal, corrected for ties and by a continuity correction of 0.5 according to Verdooren and Lehman, and corrected by the first term of the Fix and Hodges correction. All of the statistical and experimental procedures were selected so that any errors in the estimated values of P would always tend to make these values higher and too pessimistic rather than lower and too optimistic. Second the cumulative distribution functions of these calculated values of P were plotted as shown in Figs. 8 and 9. For the purposes of this calculation the men were divided into two age groups as shown in Fig. 10. If no aging pattern existed, the values of P would fall on the straight diagonal line shown in the figures. This has been the case in some other types of experiments which we have performed. In the case of age, however, these values show a very strong pattern. The asymptotes of the curved data lines indicate that about 30 urine amines, amides, and amino acids and about 30 urine volatile compounds show such strong age correlation that they can be experimentally demonstrated with fewer than 40 men in each group. These age correlations are shown in Tables I and II. These tables give the substance identity where known, the direction of the correlation, PAGE, expected and found numbers of substances at various values of P, normalized area, and age diagnostic powers, DPAcE, for each substance. The values of DPAGE were calculated as
58
LIFE EXPECTANCIES FOR THE MEN OF KNOWN AGE - -
T
---T-
r
--7
T
OLD GROUP
~3
T
YOUNG GROUP
ILl
>.
zILl o
0
10
20 30 40 50 60 YEARS OF REMAINING LIFE
70
Fig. I 1. Life expectancy distributions for the young and old groups of men used for the calculations illustrated in Figs. 13--17. The life expectancy of each individual was tabulated from his chronological age and U.S. Public Health Service compilations.The smooth curved lines were then drawn by interpolation.
described [17] and illustrated in Figs. 13--17. DPA6E is underestimated in these groups, because the overlap in life expectancies makes a perfect separation improbable. For this reason a corrected value, DP c, was also calculated as shown in Figs. 13--17.
URINE AMIDE VALUES OF 205 MEN GROUPED AND AVERAGED
u.
W I'L
O~
40
.a+
20
zv.
10
.d
°
50
30 20 10 YEARS OF REMAINING LIFE 40
10
Fig. 12. Normalized urine amide values for the 205 men grouped and averaged in six groups of similar age. The groups contained 38, 60, 53, 32, 13 and 9 men from young to old, respectively. The years of remaining life were calculated from actual ages of the men and U.S. Public Health Service compilations. The error bars are drawn for one standard deviation of the means.
59
Figure 10 shows linear regression lines for the 15 most strongly age diagnostic substances. Seven of these 15 substances decrease with age, while 8 of them increase with age. The diagnostic strength of these correlations is not only a function of age dependence, but also a function of contributions to the substance distribution functions from diet and the many other uncontrolled variables in this real-world sampling. A substance might be very strongly age-dependent and yet have a low diagnostic power, because its distribution function is broadened by other variables. Also, some of the chromatogram peaks contain two or more substances in comparable amounts which usually weakens their correlations. Life expectancy distribution functions for the two groups of men of known ages are shown in Fig. 11. These were calculated from U.S. Public Health Service statistics as applied to the age of each individual. The expectation values of the years of remaining life were then used to construct Fig. 11. Figure 12 shows, for example, a graph of the fourth most diagnostic peak as a function of years of remaining life for these men. The two curves in Fig. !1 overlap significantly; we are measuring physiological parameters which are expected to correlate best with physiological age rather than
AGE CLASSIFICATION WITH 5 URINE AMINES THAT DECREASE WITH AGE CORRECTLY CLASSIFIED YOUNG MEN Z
g,i :S
30
20
I
I
10
0
I
Z
klJ
DP = 0.63 DPc = 0.69
30 _.1
0 o
_1
o 10Q
ILl
UJ m
U,.
u.
2o ¢n ,<
.J
20 ~
-1
J
FW
~1o W
3o ~
0
o
z 0
10
20
30
INCORRECTLY CLASSIFIED YOUNG MEN
Fig. 13. Age classification with 5 ninhydrin positive urine compounds which decrease with age. Those with the highest diagnostic powers from Table i have been used. By means of their diagnostic coefficients, RAGE, the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
60
chronological age; and we have only chronological ages available for the human subjects of the experiment. Therefore, we cannot expect this technique to show perfect chronological age separation between the two groups of men corresponding to a DPAcE equal to 1.00. Figure 13 shows a diagnostic power graph utilizing the equally weighted 5 urine amines which decrease with age and show the highest DPAG E. The diagonal straight line is the line expected with uncorrelated substances. A point in the origin corresponds to DPAc E equal to 1.00. The smooth curved line represents a perfect result taking into account the overlap of the distribution functions of life expectancy as shown in Fig. 11. The stepped function is the experimental result. DP and DP~ are the fractions of separation achieved experimentally compared to a point in the origin and compared to the curved line respectively. The computerized calculation determines RAC E [14] for each sample and thereby places each individual on a linear axis between the young and old patterns. It then separates the individuals at all possible points along this axis and tabulates and
AGE CLASSIFICATION OF 7 URINE AMINES T H A T INCREASE WITH AGE CORRECTLY CLASSIFIED YOUNG MEN 30
20
10
0
z
Ll.I
DP = 0.78
30
/
;
.J
o
I.U
20 .-I ¢.) ._/ w
30
o
~"
z 0
10
20
30
INCORRECTLY CLASSIFIED YOUNG MEN Fig. 14. Age classification with 7 ninhydrin-positive urine compounds which increase with age. Those with the highest diagnostic powers from Table 1 have been used. By means of their diagnostic coefficients, RAGE, the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
61
graphs tile errors as shown. To avoid bias, the average patterns were recalculated for each individual excluding that individual from the averages. Where two experiments are available (as in the blind experiment described below and in Table !II), none of the test samples are included in the average patterns. The series of Figs. 13--17 shows the gradual increase of reliability of this technique as more substances are added to the calculation. The 5 decreasing amines have DPc of 0.69 (Fig. 13). This is increased to 0.91 when the 7 increasing amines are included (Fig. 15) and to 0.93 when the 3 most diagnostic urine vapor compounds are included (Fig. 17). Figures 14 and 16 show DPc for the increasing amines and for the urine vapor substances alone. A pattern for aging was calculated for these 15 substances and evaluated by means of the diagnostic coefficient, RACE, as described in Ref. 14. By this method the individuals can be positioned along a linear axis of the type illustrated in Fig. 4. RA~E values for the 66 men after separation into five groups with similar life expectancies are shown in Fig. 18. Finally each of the 30 men whose ages had not been given to us were placed on the linear aging axis by means of their individual RA~E as determined from urine
AGE CLASSIFICATION WITH 12 URINE AMINES CORRECTLY 30 z
\
CLASSIFIED YOUNG MEN 0 20 10
~
~
Z
ILl
kkl
30 •J
DP = 0 . 8 3
~e~Oo ~'~ "1
~
=
J
10 °
¢3 Ill
0.91
m g.
20 ~ .J
1-10
o
1-
=
30 ~
o o
0 o
zD
0 10 INCORRECTLY
20
30
CLASSIFIED YOUNG MEN
Fig. 15. Age classification with 12 ninhydrin-positive urine compounds which increase and decrease with age. Those with the highest diagnostic powers from Table I have been used. By means of their diagnostic coefficients, RAGE. the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation. a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
62
AGE CLASSI FICATION WITH 3 URINE VAPOR SUBSTANCES CORRECTLY CLASSIFIED YOUNG MEN 30 20 10 0 z ~: 30 ,.J 0 u.J
u. 20 .,.d
W re"
3O
Q Z
.... L
o
10
20
30
I N C O R R E C T L Y C L A S S I F I E D YOUNG MEN
Fig. 16. Age classification with 3 urine vapor substances.. Those with the highest diagnostic powers from Table 11 have been used. By means of their diagnostic coefficients, RA(;I, the men have been placed on a one-dimensionalaxis extending from the young to old aging panerns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DPc are explained in the text.
analysis, the fifteen substances, a n d calculations identical to those described above. The successful results of this blind experiment along with the s u b s e q u e n t l y revealed ages of 15 of the men are given in Table Ill. These 15 included, according to A. Cherkin, the seven youngest a n d eight oldest members of the g r o u p of 30. It must be emphasized that these are results from the analysis of substances picked only by ease of analysis, not by interest to aging researchers, and samples o b t a i n e d
casually from h u m a n s without special controls. Sample controls can n a r r o w the d i s t r i b u t i o n functions substantially as can sampling of genetically h o m o g e n e o u s p o p u l a t i o n s a n d other precautions. F o r example, in Fig. 19 (from Ref. 12) the RAc;E for six sets of 24 enzymatically digested Drosophila melanogaster, four each from each day of six consecutive 6-day periods, are given for only two correlations the increase in gtutamic acid and decrease in glutamine which occurs with age. This experimental result is remarkably exact, but it is also of little immediate applicability to the q u a n t i t a t i v e m e a s u r e m e n t of physiological age in large p o p u l a t i o n s of h u m a n s . A b o u t 60 substances or a b o u t one-third of the substances measured in these h u m a n experiments were f o u n d to be age correlated, but only 15 of these have been
63
AGE CLASSI FICATION WITH 12 URINE AMINES AND 3 URINE VAPOR SUBSTANCES CORRECTLY CLASSIFIED YOUNG MEN 30 20 10 0
Z I,U
a 30 .J
o o
j 10 O
DPc = O.93
W
g.
.J
,-I
~1o ua n-.
tll
O
ii
z
~3O
O tj
0
10 20 30 INCORRECTLY CLASSIFIED YOUNG MEN
Fig. 17. Age classification with 15 urine compounds: 5 ninhydrin positive compounds that decrease with age; 7 ninhydrin positive compounds that increase with age, and 3 urine vapor compounds. Those with the highest diagnostic powers from Tables I and Ii have been used. By means of their diagnostic coefficients, RAGE, the men have been placed on a one-dimensional axis extending from the young to old aging patterns. They have then been divided into two groups at all possible division points along that axis and the errors of division between young and old tabulated. No individual has been compared to a pattern which included his own values in its determination. The diagonal line is the random expectation, a point in the origin would be perfect separation, the curved line represents the theoretically perfect result given the overlap in life expectancies, and the stepped function is the experimental result. DP and DP c are explained in the text.
LIFE REMAINING FOR 66 MEN VS. DIAGNOSTIC COEFFICIENT OF AGE FOR 12 URINE AMINE AND 3 URINE VAPOR SUBSTANCES I
I
l
I
I
I
I
I
¢3 < +30 IL
o I.- +20 z ILl +10 m ii
u. 0 uJ O ¢J -10 ¢J I'-
~-20 Z
~ < a
_30 0
I
10
20 3O 4O YEARS LIFE REMAINING
I
I
5O
Fig. 18. The diagnostic coefficients, RAGE, for the 15 substances of the 66 men used in the calculations shown in Figs. 13--17. The men have been grouped into five groups of similar age. The ages and diagnostic coefficients were then averaged to give the points shown on this figure. The error bars are drawn for one standard deviation o f the means.
64
AGING
1
-10
2
DROSOPHILA
3
1 -5
I
MELANOGASTER
4
i 0
5
I 5
6
1(]
DIAGNOSTIC COEFFICIENT, RAGE FOR GLUTAMICACID AND GLUTAMINE AFTER ENZYMATICDIGESTION Fig. 19. Values of the age diagnostic coefficient, RAG E, of glutamic acid and glutamine for six sets of 24 Drosophila melanogaster, each representing consecutive 6-day periods. The groups are consecutively numbered with 1 designating days 1--6, 2 designating days 7--12 and so on. This figure is reproduced from Robinson, A . B , Willoughby, R. and Robinson, L.R., /~\x-p. Gerontol., I I (1976) 113.
included in the calculations reported here. Inclusion of the next few most diagnostic substances has no significant effect on the results. Thereafter, the DPAGE decreases when additional substances are included in the calculation. These additional substances include some independent age information, but their distribution functions also contain so much noise from other effects such as diet that they do not make a positive contribution to these calculations. STANDARDIZED HUMAN SAMPLES.
The principal limitation of these experiments and of advances in clinical chemistry in general is the lack of standardized human samples with which to test and calibrate the techniques. Herein, for example, we have used samples characterized as to chronological age, although our interest is in physiological age. We do not know of the existence anywhere of a set of human urine samples which have been carefully and quantitatively characterized as to the physiological age of the subjects. We have previously proposed (see for example Ref. l l) that a national sample bank of human blood, urine, breath, and other tissues be established. This should be done by the sampling of approximately 50 000 healthy Americans at least at 6-month intervals with the samples stored at - 7 6 ° C or lower in a bank of small freezers with appropriate equipment duplication and sample redundancy. Over the course of a 5-year period a statistically useful number of these 50 000 people will fall ill of various diseases down to and including the incidence of multiple sclerosis. The bank will then contain prospective samples for the individuals and for the group which can be used to calibrate machines which may be able to detect diseases before the individuals fall ill. Medicine could then combat the probability of illness, rather than illness itself. This bank would also allow the use of individuals as their own controls thereby improving the resolution of clinical techniques. This advance in preventive medicine would itself have a positive effect on human longevity. However, direct effects on aging research would be even more pronounced.
65
As the years pass and the 50 000 individuals gradually die, the bank of samples would contain an increasingly valuable set of samples from individuals whose years of life remaining would then be known. Subdivided with respect to cause of death, these samples, characterized as to true individual physiological age, would be very valuable. The samples could be used to calibrate techniques for the quantitative analysis of human physiological age. Since they would be rigorously characterized and available to various experimenters from a single source, the samples could also be used in comparative experiments to determine which techniques are most useful for the measurement of age. This sample bank should probably be started uniquely for this purpose and should be of the highest possible quality. Efforts to salvage old sample sets from other projects would probably lead to expensive, yet suboptimum results. It is important to emphasize that the bank must contain samples, not the results of analyses. It would be tragic to undertake this project with limitation to one or more particular analytical techniques which happen to now be available and popular. Surely those techniques would be obsolete by the time results were obtained. If, however, the samples had not been retained, no further advance would be possible. If we could make only one contribution to the advancement of the use of modern technology for the improvement of human health, the creation of this sample bank of body fluids from well individuals would be our choice. If this bank had been created when we first formally proposed it 14 years ago, substantial improvements in clinical medicine and in empirical aging research would already be in evidence. CONCLUSION
We conclude that there exists a very strong systematic metabolic pattern for human age, and we estimate that these techniques should be able to be refined with relative ease to measure the physiological age of individual human subjects to an accuracy of 3 years or less. This would permit the measurement of physiological age in a group of 300 individuals to within an accuracy of 2 months or 1300 individuals to within an accuracy of 1 month. Rate of aging experiments could then be performed to within an accuracy of about 10"/,,over a period of I year. As experience is gained, this should be improvable by about one additional order of magnitude which would permit smaller, shorter, or more accurate experiments. Improvements can also be expected from the separation of individuals into types. We observed for example, in an experiment in which we measured 60 urinary amines, amides, and amino acids in the urine of 1000 newborn infants (Robinson, A.B. et al., unpublished data), that the human distribution functions of amounts of many of these substances were not unimodal. Instead they were bimodal and multimodal. We expect this to be generally true, so that discrete typing of individuals will eventually markedly increase the resolution of these techniques. Improvements can also be made by particular constraints such as dietary control.
66
We have already demonstrated that with strict dietary control the urinary patterns become so unique that computerized fingerprinting of single individuals with a high degree of accuracy can be carried out. Strict dietary control, however, sharply limits the utility and breadth of applicability of these techniques. Refinement need not make use of these particular analytical techniques. Any method which measures about 200 metabolic substances quantitatively should be sufficient. (Some methods such as field ionization mass spectrometry have the potential to measure 2000 or more substances within a few minutes or less.) Refinement, however, absolutely requires that a sample set be available in order to calibrate the technique. Until this sample set is carefully gathered, stored and made generally available, this technique and many other potential analytical advances in human health will probably remain research curiosities rather than proper contributors to the quality of human life. We emphasize the following: (1) About one-third or more of the metabolic constituents of human body fluids correlate systematically with age, as do the constituents of flies and mice. For practical, empirical purposes, therefore, analytical techniques for profiling should emphasize those substances which are easy and inexpensive to measure quantitatively. Any sufficiently large set of casually selected metabolic substances contain within their amounts the necessary information as to individual physiological age. (2) Simple computerized techniques exist for the test and utilization of data from these profiling experiments. We have also used more sophisticated pattern recognition techniques, but have generally found them cumbersome and without concomitant additional utility. 'Eyeballing' chromatograms or mass spectra is, however, essentially without value. Proper non-parametric tests of reliability on statistically useful sets of analyses must be performed. A simple procedure for doing this is illustrated herein. (3) The principal limitation to this work is in sampling - - not in analysis or computation. Although other analytical techniques can be utilized and many interesting reports written regarding the analytical profiling of age, further real advances in this effort will probably not be possible until a proper prospective sample bank is established. REFERENCES 1 2 3 4 5 6
C.S. Minot. The problem of age, growth, and death. Pop. Sci. Month(v. 71 (19071 481. D. Harman, Aging: a theory based on free radical and radiation chemistry. J. Geromol.. II (1956! 298. J. Bjorksten, A c o m m o n molecular basis for the aging syndrome. J. Am. Geriatr. Sot., 6 ~1958) 740. L. Orgel, The maintenance of the accuracy of protein synthesis and its relevance to ageing. Pro~, Nat. Acad. Sci. USA.. 49 (19631 517. Z,A, Medvedev, The nucleic acids in development and aging. Adv. Gerontol, Res., 1 (1964) 181. B.L. Strehler, Code degeneracy and the ageing process: a molecular genetic theory of aging. Proc. 7th Congr. Gerontol. Vienna. (1966) 177.
67 7 8 9 10
11 12 13 14 15 16 17 18
L. Walford, The bnmunoh~gic'al Theo O" q/Aghlg, Williams and Wilkins, Baltimore, 1969, pp. 70. R.W. Hart and R.V. Setlow, Correlation between DNA excision repair and life span in a number of mammalian species. Proc. Nutl. Acad. Sci. USA, 71 (1974) 2169. J.L. Bada, Aspartic acid racemization in tooth enamel l¥om living humans. Proz'. NatL Ac~ul. Sc'i. USA, 72 (1975) 2891. A.B. Robinson, J.H. McKerrow and P. Cary, Controlled deamidation of peptides and protcins: an experimental hazard and a possible biological timer. Pro~'. Natl. Acad. Sc'i USA. 66 (1970) 753. See also A.B. Robinson, Proc. Natl. Acad Sci.. 71 (1974) 885 and A.B. Robinson and C. Rudd, Deamidation of glutaminyl and asparaginyl residues in peptides and proteins. Curr. Topic's Celluhn" Reg., 8 (1974) 247. A.B. Robinson, Molecular clocks, molecular profiles, and optimum diets: three approaches to the problem of aging. Mech. Ageing Dev.. 9 (1979) 225. A.B. Robinson, R. Willoughby and L.R. Robinson, Age dependent amines, amidcs and amino acids residues in Drosophila mekmogtlster. Exp. Gerontol.. II (1976) 113. A.B. Robinson, M. Weiss, W.E. Reynolds and L.R. Robinson, Use ¢~l'Mass Spectrometrv.[br Orthomolecular Diagnosis, 23rd Ann. Conf., Mass Spect., Houston, Texas (1975) 182. A.B. Robinson, H. Dirren, A. Sheets and R. Lundgren, Quantitative aging pattern in mouse urine vapor as measured by gas-liquid chromatography. Exp. Gerontol. II (1976) II. A.B. Robinson, D. Partridge, M. Turner, R. Teranishi and L. Pauling, An apparatus for thc quantitative analysis of volatilc compounds in urinc. J. Chroouttograph.r, 85 (1973) 19. A.B. Robinson, Antioch Review. 1981, p. 383. A.B. Robinson and F.C. Westall, The use of urinary aminc measurement for orthomolccular diagnosis of multiple sclerosis. J. Orth. Psych., 3 (1974) 1. F. Wilcoxon, Biometrics, I (1945) 80: Lehman, S.Y., J. Am. Statist. Assoc.. 56 (1961) 293; L.R, Vcrdooren, Biometrika, 50 (1963)177: and E. Fix and J.L. Hodges, Ann. Math. Stcttist., 26 (1955) 301.