GENETIC GAIN I N POPULATIONS OF TRZBOLZUM CASTANEUM

UNDER UNI-STAGE TANDEM SELECTION AND UNDER

RESTRICTED SELECTION INDICES

ELIYAHU SCHEINBERGZ, A. E. BELL AND V. L. ANDERSON

Population Genetics Institute, Purdue University, Lafayette, Indiana 47907
Received August 15, 1966

HE behavior of quantitative traits under diversified artificial conditions is of

primary interest in population genetics. The population geneticist interested
in the behavior of a given quantitative trait or in a complementary set of such
traits under specified conditions will probably use several systems of matings
combined with various methods of selection to determine !whichof these combina-
tions will best maximize genetic gain. Even though immediate interest may be in
a particular trait, the organism as an integral unit must be considered. Evaluation
of the individual organism or candidate for selection would be desirable in terms
of its total breeding value, based on as many phenotypically expressed attributes
as possible, relative to other individuals present in the same population. One
such method is a genetic selection index, which may be of two general forms:
the nonrestricted selection index described by SMITH(1936) for plants and by
HAZEL(1943) f o r animals, and the restricted or optimum genotype selection
index described by KEMPTHORNE and NORDSKOG (1959), and by TALLIS (1962),
respectively.
Theoretical and experimental considerations bearing on the validity of HAZEL’S
nonrestricted index have been given by more than a score of investigators. The
pioneer paper of HAZEL and LUSH (1942) and its extensions by YOUNG(1961)
and FINNEY(1962) have theoretically compared three methods of selection, one
of which is the nonrestricted selection index. Maximum genetic gain in total
score should result when selection is based on a nonrestricted index. This is
particularly true when the number of selected traits is large. The method of
independent culling leuels, though less efficient than the nonrestricted index is
more so than tandem selection. However, these theoretical expectations have not
always been realized in experimental situations (RASMUSON 1964). Very little
appears to have been done experimentally with the KEMPTHORNE-NORDSKOG
model for the restricted selection index, and no test of the biological properties of
TALLIS’S selection index for optimum genotype has been published.
The KEMPTHORNE-NORDSKOG restricted selection index has been biologically
evaluated by ABPLANALP, OGASAWARA and ASMUNDSON ( 1963), who observed

‘ Journal Paper No. 2818 of the Purdue Agricultural Experiment Station. Based on a dissertation submitted by the
seiiior author to the Graduate School, Purdue University, in partial fulfillment of the requirements for the degree of
Doctor of Philosophy.
Present address. Animal Genetics Section, Central Experimental Farm, Ottawa, Canada.

Genetics 5 5 : 69-90 January 1967.

70 E. SCHEINBERG et al.
the effectiveness of the restricted index in selecting for two body weights in tur-
keys at different ages by comparing it to mass or uni-stage tandem selection when
selection was practiced for either of the W O traits independently. Their results
agree, within a specified range of deviations, with the theoretcially expected
genetic gains. This is acceptable providing their appendix has been corrected for
the equations dealing with the expected genetic gain, aGi, for a given trait.
The present study was designed ( 1 ) to check the biological validity of the
restricted selection index, and (2) to consider the question: Which of the avail-
able selection methods tested here should one use to improve a stock most effi-
ciently under specified conditions and with a fixed cost?

SOURCE OF DATA A N D EXPERIMENTAL PROCEDURES

The foundation population of Tribolium castaneum used i n this investigation was derived in
1954 from the systematic crossing of eight random mating laboratory stocks of varying geographic
origins. This synthetic stock was propagated each generation with a minimum of 250 randomly
*
chosen parents. I t was maintained in a control chamber at 22.2 t 1°C and 61.5 2% relative
humidity, in half-pint milk bottles containing standard medium (whole wheat flour sifted through
#2 silk bolting cloth and 5% dried brewers’ yeast). Upon commencement of this investigation,
90 adult males and 90 adult females chosen at random were transferred to another control
chamber. All populations were maintained at 32.8 -t 0.5% and at 70 t 1% relative humidity
in standard medium.
Three replications were set consecutively at one-week intervals. During the selection phase
each replication consisted of seven populations. Every individual in each population was measured
for five traits: (1) 13-day larval weight (LW); (2) Developmental time (DT) from egg to
pupa, checked at 12-hour intervals starting at day 13.5 of the cycle; (3) pupal weight (PW)
within 24 hours of pupation; (4) daily gain (DG) from pupal to larval weight [DG =
(PW-LW)/(DT-13)]; and (5) Number of larvae (LN) produced by pair-mating, from an
18-hour period of egg collection.
Traits 1 to 3 were considered to be primary, while 4 and 5 were considered secondary.
Traits 1 to 4 were measured on each candidate for selection, while (5) was scored for each
mating. Random mating of the selected individuals was chosen, with avoidance of full-sib mat-
ings. Propagation was by 15 single-pair matings per population. Five randomly chosen offspring
were observed per mating. In generation -1, (where generation 0 is the first in which selection
was practiced), six offspring per pair mating were observed. The number of matings within
each population was kept constant throughout selection by using reserve matings. These ac-
counted for less than 3% of the total matings in the selection phase; thus, little, if any, bias
should be expected in the selection differentials.
The seven populations, identical in their physical structure in each of the three replications
are given in Table 1 with their symbolic designation and selection criterion.
The mode of propagation and the detailed scheme of partitioning the base population in
successive generations into replications and populations which determined the experimental struc-
ture during the selection phase of the experiment are given in Figure 1. Numbers in parentheses
indicate the number of viable matings or individuals expected to contribute equally to the gene
pool of the next generation. Numbers realized are given at the left.
Random partitioning of the base population into replications and populations constituted one
of the main objectives of generations -2 through 0. This step can be usually accomplished in
one generation. However, a faulty humidity control at the beginning of generation -1 caused
an extension i n the partitioning process for a n additional generation. The one week interval
between successive replications, and the distribution of the adults within each replication into
the seven populations were accomplished in generation 0. Thus, selection of parents by truncation
was initiated in generation 0. and the progeny of the first cycle of selection were measured in
 RESTRICTED INDICES I N TRIBOLIUM 71
TABLE 1
List of symbols used to designate the selection methods used for the seven populations

Population Selection method

SI Uni-stage tandem selection for high larval weight.

s 2
Uni-stage tandem selection for short developmental time.
s, Uni-stage tandem selection for high pupal weight.
I, ?? Index selection for high larval weight while restricting develop-
mental time and pupal weight.
1 9 7 ,. Index selection for short developmental time while restncting
larval weight and pupal weight.
I,,,, Index selection for high pupal weight while restricting larval
weight and developmental time.
C No selection, propagated and measured as a control population.

generation 1 . For purposes of statistical and genetic analysis, as well as of parameter estimation
for index construction i n generation 0, each replication prior to selection was considered as a
large random mating population despite its physical separation into populations. As soon as the
parents of generation 1 were selected using the designated criterion, the selection schemes were
initiated and each population was treated as a separate and distinct genetic group. The general
procedure and the exact time of measuring the five traits. together with the stock husbandry
for all the selection cycles reported in this study, are given in Table 2.
Recorded measurements for all five traits observed, and for viability at selection time, were
required for each individual entering the statistical and genetic analyses. Individuals not meeting

Base population 9 m x 9w9

Generation -2
\/
689 Offspring (700)

Mating the parents of Generation - 1 32453

I 32499

Partitioning into replications 103 Matings (108) 102 Ebtings (108) I01 Matings (108)

Generation - I 604 Offspring (618) 586 Offspring (612) 598 Offspring (606)

Mating the parents o f Generation 0 140% x 14099

(Shown here for one Replication)

Partltioning into populations

i
'I 2' '3 '1.23 '2.13 '3.12

Generation 0 75 Offspring
I
(Shown here for one Population)

Selection of parents 15% x I599 + (5 reserve matings)

FIGURE 1 .-The mode of propagation and partitioning into replications and populations start-
ing with the base population.
72 E. SCHEINBERG et al.
TABLE 2
Sequence of events for each population within a replication and generation

Day of cycle Experimental procedure

0, 2 P M Parents set in 2 g of standard medium.

1, 8 A M Remove egg laying parents.
13, AM Weigh 5 random chosen larvae resulting from eggs layed on
day 0 and place each one in a n individual 3/4 oz creamer
with 0.1 g of standard medium.
8 PM Start developmental time check and continue to check at 12-
h r intervals.
14, AM Pupal weights of all pupating individuals starts and continues
at 24-hr intervals. Return pupae to their individual cream-
ers and add 0.4 g of standard medium.
24, AM Termination day: All larvae not pupating by this time are
rejected from their respective selection scheme.*
32, AM Each of the surviving adults is checked for viability.
10A M Parents of the next generation are selected.
2 PM Parents of the next generation are mated.
35, 2PM Same as Day 0, cycle starts over.
~~ _ _ _ ~ _ _ _ _ _ ~ _ _ _ _ _ _ _ _ _ _ _ _ ~ _ _ _ _

* Less than 0.2% of the larvae had to be discarded (thus there IS very little or no bias in the selection hfferentials).

these requirements (less than 4% of the total data) were omitted from all analyses. About one
half of the deletions were due to natural death between measurement of pupal weight and the
viability check at 32 days.

MODELS A N D STATISTICAL PROCEDURES

The growth and development of males and females from egg to pupa were considered as
the same manifestation of their genotype. I t is not practical to identify the sex earlier than the
pupal stage. While sexes do differ for some of the traits measured, sex is a minor source of
variation. The construction of indices as well as the estimates of genetic parameters were more
easily handled by making “sex” part of the environmental deviation rather than genetic i n the
genetic and statistical models.
The genetic and statistical models assumed i n this investigation are as follows:
I. Genetic model. Under the assumption that genotype-environment interaction is zero and
that gene action is additive, the genetic model used is:
x =p g + +
e, E(g) = E(e) = 0
where x = the phenotype, p = the population mean, g = the genetical contribution to the pheno-
type, and e = the environmental contribution to the phenotype.
11. Statistical models. The data reported here were analyzed to estimate (1) population
parameters for each trait within each population and replication, and (2) the same parameters
when pooled over the three replications. The analyses of variance and covariance models for
the nonpooled case are given below:
Analysis of uariance model:
1.3 = p
2.. + +
eii,
i = 1 , 2 ,..., m and
E ( q ) = E ( e 2 3.) = E(ai,eij) = O
i=1,2 ,..., ni
where x , = the observation on the jth individual from the ith matings, p = the overall mean,
ai = t h e effect of the ith mating, and eii = the residual effect of the jth individual from the
ith mating.
Analysis of couariance (cross products between two attributes i n the same individual) : The
 RESTRICTED INDICES IN T R I B O L I U M 73
model of this analysis corresponds to that for the analysis of variance except the terms in the
model represent cross product effects between two attributes rather than the effects of one
attribute.

RESULTS

The overall experimental census for the number of viable candidates available
for selection is given in Table 3 by generation for each population summed over
the three replications and the two sexes. The overall mean number of candidates
per population available for selection, starting with generation 0 and through
generation 10, was 72.1 with an expected mean value of 75.
The preselection populations: Estimates of the means for the five traits ob-
served in the base populations are given in Table 4. As mentioned previously,
mean larval weight dropped unexpectedly in generation -1 owing to faulty
humidity control.
From Table 4 it is evident that the means for larval weight and pupal weight
in generation -1 were lower than those observed in generation -2 while the one
for developmental time was higher. The change in the mean for larval weight
was the largest of all. I n fact, this trait in generation -1 is not the same trait
physiologically as the one observed in the previous generation. This may hold true
also for developmental time and pupal weight, except that the change in humidity
was corrected at an early stage of development and compensatory growth prob-
ably took place prior to measuring these two primary traits. The observed changes
in magnitude and direction for daily gain are consequences of changes in the
three primary traits. Since egg collection had occurred prior to the humidity

TABLE 3
The number of viable individuals wilhin each population, pooled over replications and sexes

Population
Gene1ation c $1 '2 '3 'I 21 '2 13 '3 12 Total

-2 689' 689
-1 1788t 1788
0 225$ 225 225 224 225 223 223 1570
1 220 223 217 212 217 216 217 1522
2 222 221 222 222 222 222 220 1551
3 207 214 219 211 216 215 212 14.94.
4 213 218 217 216 208 206 216 1494
5 220 224 216 223 219 218 217 1537
6 203 209 215 211 217 207 202 1464
7 222 223 219 217 213 214 214 1522
8 216 221 218 220 218 223 219 1535
9 205 208 209 213 207 207 216 1465
10 216 210 211 212 213 218 216 14.96

Total 4846 2396 2388 2381 2375 2369 2372 19127

* Expected 700.
f Expected 1836.
t Expected 225.
74 E. SCHEINBERG et al.
TABLE 4
Estimated means and standard errors for the fiue measured traits in the base populations

Developmental Number
Larval weight* tim+ Pupal weight* of larvae Daily gain$
Numbers -~
Generation observed Mean SE Mean SE Mean SE Mean SE Mean SE

-2 689 221.61 1.62 16.46 0.04 209.33 0.75 . . . . -4.74

, , 0.27
-1 1788 153.57 1.14 17.19 0.03 24l3.73 0.46 14.87 0.26 10.49 0.25
0 1570 213.71 1.05 16.19 0.02 216.56 0.50 14.22 0.21 -0.33 0.24

* 0.01 mg.
j-Days.
:0.01 mg/day.
fault, reproductive and hatching fitness in terms of the number of larvae per pair
mating would not be affected unless the environmental stress was severe enough
to affect viability. The observed number of larvae for generation -1 indicates
that viability was not affected.
This irregularity did not affect any of the traits in succeeding generations and
has been emphasized here to stress the importance of controlled environmental
conditions, especially humidity and temperature, in rearing this organism in
selection experiments.
Population parameters required for the construction of the restricted selection
indices were estimated in generation 0 by means of the analyses of variance and
covariance. These analyses pooled over the three replications included a total of
315 matings and 1570 offspring. Estimates of the genetic and phenotypic variances
and covariances, which are the primary ingredients for calculating the restricted
selection indices, are given in Table 5.
Equal parametric weighting coefficients were used in this study for larval
weight, pupal weight and developmental time. It should be emphasized that when

TABLE 5
Estimates of genetic and phenotypic variances and covariances estimated in generation 0 and
used for indices construction. Genetic variances and covariances are above the diagonal

LW DT PW
-8.6613 430.7603
LW

-2.2876
DT
-16.8752

PW
41.6518 -2.541 2

LW: larval weight. DT developmental time. PW: pupal weight.

 RESTRICTED INDICES I N TRIBOLIUM 75
one restricts (n-I) of n attributes, the parametric weighting coefficients of the
restricted traits, regardless of their sign and magnitude, do not change the
coefficients of the restricted index associated with these attributes. The parametric
weighting coefficients of the nonrestricted attribute has the function of a scalar
multiplier.
The restricted selection indices for populations I, L 3 , I, and I, ,,, used through-
out the ten generations of selection, were computed using the method presented
by KEMPTHORNE and NORDSKOG (1959) and are listed below:
I, +
,I = 0.2809 LW 10 1121 DT - 0.3190 PW
I, ,I = 0.004.9 LW f 0.3371 DT - 0.004.2 PW
I, 12 = 0 1884 LW - 5.3276 DT 0.3167 PW +
Heritabilities, genetic and phenotypic correlation coefficients together with
their sampling errors are given in Table 6. In light of the close genetic association
between larval weight, developmental time and pupal weight and their high
heritabilities, it can be expected that uni-stage tandem selection for any one trait
will result in appreciable correlated responses in the others. One point of interest
should be brought up here. Since parameter estimation and thus all the prediction
equations related to the genetic gain are based on the estimates of variances and
covariances among full-sib groups. the estimates for additive genetic variance
for each trait would be inflated by ?$, cdC) '/z vZAd + + +
u~~~ 1/8 d D D % uLA4d +
t etc., in addition to twice the variance due to maternal effects and a haction of
the variance due to sex-linked genes. In spite of these limitations, the variance-
covariance components approach was considered the most practical for the
exploratory investigation reported here.
The control population: Temporal variations in the environment during the
course of this study were measured in terms of the performance of a randomly
mated control population. The underlying theory justifying this method of meas-
uring temporal variations in the environmental effects rests on the assumption
that no directional genetic changes have occurred in the control population in
successive generations.
It is difficultto provide a control which will remain constant genetically unless

TABLE 6
Genetic and phenotypic correlation coefficients and heritability estimates from
uariance-couariance components in generation 0. Genetic correlations
aboue the diagonal and phenotypic correlations below.
The diagonal elements are heritability estimates
~~ ~~~~ ~~

__ 1,W DT PW
LW 0.59 f 0.06* -0.67 f 0.051 0.78 t 0.04
DT -0.66 +- 0.02$ 0.43 t 0.06 -0.31 f 0.08
PW 0.55 Zk 0.02 -0.20 k 0.03 0.80 t 0.06

1.W. larval weight. DT: developmental time. PW: pupal weight.

and PATERSON
* Sampling errors were computed according to OSRORNE (1952).
t Sampling errors were computed according to TALLIS
(1959).
Sampling errors were computed according to SCHEINREHC
(1966).
76 E. SCHEINBERG et al.
it is a completely homozygous inbred line or an F, of two such inbred lines.
Genotype by environment interactions can render this hypothetically ideal con-
trol inadequate in revealing the effects of environmental changes on experi-
mental populations (BRAY,BELL,and KING 1962). The control population used
in this study was of the same effective size as each of the selected populations,
was initiated from the same gene pool and was handled in the same way except
that random selection was practiced.
The selection phase: Since the main objective was to test theoretical expecta-
tions of the selection method on a biological system, it seems logical that a previ-
ously known procedure should be used as a yardstick for comparison after the
tested populations were looked at independently. Therefore, seven populations,
each replicated three times and with generally different goals, were established
as outlined in Table 1. Three of them, S,, S,, and S,, were formed to establish
evidence on direct and correlated response for the various traits when selection
was made on a single primary trait. The next three, I,.,,, I,.,,, and I, ,?, were
established to evaluate experimentally the restricted selection index. The seventh
population, C, was a control population.
The analyses of variance for all five traits with the realized genetic gain per
generation as the criterion analyzed are shown in Table 7. Since replications were
considered as a random effect and all the ratios of mean square of Replications x
Methods to Generations x Replications x Methods were less than one, the effect
of the interaction of Replications x Methods can be assumed to be negligible.
This implies that the genetic gain per generation for each method was similar in
the three replications, and provides a basis for a wider inference than that pro-
vided in selection studies which all too frequently are not replicated.
Generations were also found to have a nonsignificant effect on the genetic gain
per generation at the 0.05 level of probability for all traits except developmental

TABLE 7
The analyses of variance for the five obserued traits testing the equality of the obserued
genetic gain per generation in generations I to 10. Generations nnd methods
are fixed while replications are random

Mean squares
Larval Developmental Pupal Number of Dajly
Source df weight time weight larvae gain

Generations (G) 9 1675.46 0.8145' 200.47 15.63 88.11

Replications (R) 2 94.39 0.0288 9.74 7.47 0.20
GxR 18 1175.86** 0.2911** 132.05" 26.12" 74.28**
Methods (M) 5 828.53** 0.0944** 389.78" 2.14 33.54**
GxM 46 120.66 0.1348** 34.35 2.63 8.96
RxM 10 55.03 0.0021 19.91 0.79 1.99
GxRxM 90 130.80 0.0575 34.25 2.41 8.17

Total 179

Significant at the 0.05 level of probability.

* * Significant at the 0.01 level of probability.
 RESTRICTED INDICES I N TRIBOLIUM 77
time. This indicates that a uniform genetic gain per generation was observed for
the traits: Larval weight, pupal weight, number of larvae and daily gain. Methods
have a significant effect at the 0.01 level of probability on all traits except the
number of larvae. The results obtained for methods were expected since each
of the six methods was established to accomplish different objectives.
The response to uni-stage tandem selection for each of the five traits in popu-
lations S,, S, and S3 is given in Figure 2 a, b, c. Uni-stage tandem selection f o r
larval weight, developmental time and pupal weight was effective. It is of interest
to notice that the high genetic correlation coefficients between the three primary
traits as presented in Table 6 play an important role in establishing the magni-
tude and the direction of the correlated response.
A clear distinction should be made when considering all five traits in the
populations where restriction was practiced. There are three classifications for
these traits: (1) The nonrestricted primary traits, i.e., larval weight, develop-
mental time and pupal weight in populations I, 2 1 , IL and I, 12, respectively,
(2) the restricted primary traits, and ( 3 ) the secondary or correlated traits, i.e.
the number of larvae and daily gain.
In order to evaluate the realized genetic gains, three criteria of expected genetic
gains were introduced: ( 1 ) Theoretically expected genetic gain arrived at by
using estimates of the parameters in generation 0 and a theoretical selection dif-

TABLE 8
The theoretical and obserued genetic gains per generation for the three primary traits
Theoretical genetic gains
Generation Cl Selection phase
With a theoretical With a theoretical With the observed
Population selection differential selection differential selection differential Observed gains

Larval weight
Sl 22.788 18.748 f 0.159 14.491 f 0.119 7.832 t 0.475
S, 12.886 9.070 f 0.223 6.660 t 0.152 4.680 f 0.575
s3 20.546 17.621 t 0.377 14.164 k 0.306 8.298 t 0.410
I,.,, 12.795 24.294 t 0.868 7.312 k 0.346 2.640 ir 0.397
12.1, 0.000 -23.439 i 0.632 -10.874 k 0.363 -6.851 t 0.480
LE 0.000 20.426 i 0.549 - 8.086 i 0.343 -2.494 t 0.401
Developmental time
Sl -0.201 - 0.184 ir 0.008 - 0.140 f 0.005 -0.064 i 0.010
s2 -0.255 - 0.161 t 0.004 - 0.115 f 0.003 -0.122 i 0.015
s3 -0.108 - 0.103 t 0.006 - 0.081 f 0.004 -0.056 f 0.012
11.23 0.000 0.231 f 0.018 0.074 f 0.007 0.067 t 0.011
12.13 -0.239 0.087 f 0.009 0.035 f 0.004 -0.015 f 0.013
13.12 0.000 0.439 k 0.080 0.066 t 0.005 0.011 k 0.010
Pupal weight
Sl 8.836 8.508 t 0.102 6.546 f 0.088 3.618 f 0.242
s2 3.586 3.122 f 0.123 2.327 i 0.085 0.815 f 0.241
s3 14.997 13.618 t 0.161 10.977 f 0.113 6.652 f 0.141
I1.2, 0.000 - 6.831 f 0'.983 0.327 f 0.194 -0.464 f 0.1 71
12.13 0.OOO - 9.468 f 0.287 - 6.152 I
.0.203 -4.652 I?I 0.149
13.12 10.622 4.794 f 0.453 1.427 I
.0.143 2.672 i 0.082
 a POPULATION SI

GENERATION QENERATION

b. POPULATION S,

QENERATION GENERATION

c. POPULATION S,
LW

-4 5 0 I P 04 IE 8i 10-

GENERATION GENERATION

FIGURE
2 (a-f).-Selection responses as deviation from the control for the three primary and
two secondary traits. Ten units on the daily gain scale equal to one larva.
RESTRICTED INDICES I N T R I B O L I U M 79
d . POPULATION 11,23

1.5 .

or . -
1.0 ' /

I-
I- -0.5 . W

'40
-10 -
-760

-1.5 .
1
0 0 4 6 0 K )

OENERATION GENERATION

e. POPULATION I e l 3

15.
'
60
1.0
. 40
0.5 . . e

-1.5 .
0 L 4 6 0 la
OEWERATION 0E N W T I O N

E. POPULATION I , , ,

I5

60
IO.

PW ' 4 0
05

-I0 .
'-60
-I 5 .
b t . 4 , ; 0 . 2

OCNERATION OENERATION
80 E. SCHEINBERG et al.
ferential (0.95 standard units), (2) A second computed from the genetic and
phenotypic parameter estimates in each generation during the selection phase
using the same theoretical selection differential in each generation, and (3) A
third based on the same parameter estimates as (2) except that the observed
selection differentials were used rather than a theoretical value. The last two
criteria, based on the selection phase, take into account the changes in parameters
as a consequence of selection and are an average value covering generations 0
through 10. The values of these theoretically expected genetic gains and the ones
for the observed gains are given in Table 8. The realized portion of the expected
genetic gain for each of the three primary traits is given in Table 9. The results
given in these two tables together with the results given in Table 10 (willbe dis-
cussed further when comparing the theoretically expected genetic gain to the
realized gain for each trait in each of the selected populations.
(1) Population Zl.23. The postulated objective in this population was to maxi-
mize the genetic gain for high larval weight while restricting the two other
primary traits. From Figure 2d, it is evident that selection was effective for high
larval weight while the restriction imposed on developmental time and pupal
weight was not as expected. Developmental time responded in the positive direc-
tion while the response of pupal weight was in the opposite direction. The number
of larvae responded slightly in the positive direction though not significantly at
the 0.05 level of probability. On the other hand, daily gain responded signifi-
cantly in the negative direction. The negative response in daily gain was due to
the fact that larval weight was higher than pupal weight in this population. This
is not solely due to selection for high larval weight since the results were similar
in populations S, and S, where larval weight was not selected for, but was ob-
served as a correlated trait.
These findings are confirmed by statistical test of the regression coefficients
(,6) of the genetic gain for each trait on generation number. These regression
coefficients are given in Table 10. The hypothesis p = 0 cannot be accepted at
the 0.05 level of probability for larval weight, developmental time and pupal
weight. This is also true for daily gain while for the number of larvae the hy-
pothesis cannot be rejected at the same level of probability. This population is

TABLE 9
The realized portion of the expected genetic gain in the three primary traits
Population Larval weight Developmental time Pupal weight

SI 0.34* 0.4.91 0.54% 0.32* 0.35+ 0.462 0.41' 0.42-t 0.55%

S, 0.36 0.52 0.70 0.48 0.76 1.06 0.23 0.26 0.35
s, 0.41) 0.47 0.59 0.52 0.54 0.69 0.44 0.49 0.61
11.23 0.21 0.11 0.36 .. . 0.29 0.90 ... 0.07 -1.41
12.1, ... 0.29 0.63 0.06 -0.17 0.43 ... 0.W 0.76
L I Z ... -0.12s 0.31 .. . 0.02 0.17 0.25 0.56 1.87

* Observed gainhheoretical expected gain from generation 0.

i Observed gaidtheoretical expected gain from the selection phase with a theoretical selection differential.
d Observed gainhheoretical expected gain,from the selection phase with the observed selection differentials.
S A negative sign indicates response opposite in direction to the one expected.
 RESTRICTED INDICES I N TRIBOLIUM 81

+I $1 t l $1 tI +I

tl +I tI $1 $1 $1
Z6 0g 0g 08 08 80
t $1 $1 tl tI +I

$*rqtrq"
6 0 0 0 0 0
tl tl tI +I $1 tl

ti00000
tl tl +I $1 +I +I

Ou)011MO $1 $1 $1 tI tI ti

tI $1 tI tl $1 $1
s 5 g 9a 9a 9z
800000
I l l I
$1 tl tl $1 tl tl
m m o h o - --gwcnrD
h h - m w
*"a?*$
0 0 0 0 0 0
30 80 00 30 S0 S0
$1 $1 tl $1 $1 ti I /

+I tl t l $1 $1 $1
82 E. SCHEINBERG et al.
the only one in this study where no significant correlated response was observed
for the number of larvae.
The realized genetic gains for one of the restricted traits, developmental time,
was close in magnitude to the theoretical genetic gains.estimated from the selec-
tion phase. The unexpected negative response of pupal weight, though statistically
significant, is of a rather small magnitude. The ratio of the observed genetic gain
for the nonrestricted primary trait, larval weight, to either one of the thearetically
expected genetic gains is of low magnitude (0.21, 0.1 1 and 0.36, respectively) as
shown in Table 9. Such a situation may provide information on the reliability
of applying an index based on parameters estimated from a large random mating
population to a smaller population (1570 versus 72.1 ) .
An analysis of trends in the genetic variances and covariances among traits
over the course of the experiment yields no immediate answer (Table 11) . While
the genetic variances for all three primary traits in Population I,,,, were dimin-
ishing, the only statistically significant change was that for developmental time,
the restricted trait which responded most closely to theoretical expectations. The
genetic covariances also show a downward trend with the one between larval
weight and pupal weight being statistically significant.
(2) Population Z2,13.The postulated objective in this population was to maxi-
mize the genetic gain for short developmental time while restricting the two
other primary traits. From Figure 2e, it is apparent that the realized genetic
gain for developmental time was far from its expectation and that for each of
the restricted traits obviously differed from the a priori formulated objective.
Developmental time showed a slight, though not significant, response in the
negative direction. The two restricted traits responded significantly in the nega-
tive direction which is contrary to the expectation of no response. Daily gain
responded significantly in the positive direction for the first time in this study.
This resulted from the response of pupal weight being less negative than that one
for larval weight. The number of larvae also responded significantly in the nega-
tive direction. This is the only case in this study where the number of larvae
responded in the negative direction.
The fact that the genetic gains obtained for developmental time and the re-
stricted traits, larval weight and pupal weight, are different from the ones ex-
pected from the restricted selection index formulated for this population deserves
further attention. From Table 8 it can be seen that the theoretically expected
genetic gain for developmental time based on parameter estimates from genera-
tion 0 is -0.239 days per generation while the one estimated from the selection
phase with the realized selection differentials is 0.035. The two theoretical expec-
tations differ not only in magnitude but also in direction. The observed gain is
-0.015 days per generation. Though not significantly different from zero, it is
in the same direction as the theoretical expectations from generation 0. As shown
in Table 8, the observed negative genetic gain of the two restricted traits is in fair
agreement with the theoretical ones predicted from the selection phase of this
study though this type of response was not anticipated. The genetic parameters
estimated from the selection phase again fail to explain the results. It can be seen
 RESTRICTED I N D I C E S I N T R I B O L I U M 83
in Table I1 that the genetic variances for all three traits decreased over genera-
tions, but only that for pupal weight was statistically significant. The changes
in the genetic covariances are not consistent and are of no appreciable magnitude.
( 3 ) Population I , The postulated objective in this population was to maxi-
mize the genetic gain for high pupal weight while restricting the other two pri-
mary traits. It seems that it has been realized for this population in a more
efficient way than in the other two populations where restrictions were imposed.
This is shown in Figure 2f. The realized response of pupal weight to selection,
as shown in Table 8, is about twice as large as the theoretical expectations esti-
mated from the selection phase. However. when comparing the realized response
to the theoretically expected one as estimated from generation 0, the realized
portion of genetic gain in this trait shown in Table 9 is only 0.25 which is con-
siderably less than 1.87. In this population, the hypothesis that ,8 = 0 for develop-
mental time cannot be rejected at the 0.05 level of probability and verifying the
fact that the adequate restriction can be applied and realized. However, this is
not true for larval weight. The negative realized response of larval weight rather
than no response, as postulated for this population and population I?,?, is probably
a function of the genetic and/or phenotypic covariances of this trait with the
other two primary traits.
The results given in Table 11 do not reveal any additional information except
that the change in the genetic variance for developmental time is positive while
the changes for the other two traits are in the opposite direction with none of
them being statistically significant.

DISCUSSION

The underlying assumptions for most prediction equations in quantitative

genetics dealing with direct and indirect genetic gain, including the KEMP-
THORNE-NORDSKOG restricted index, is that the attribute or attributes under con-
sideration are normally distributed and that their inheritance is additive. How-
ever, this idealized situation is seldom, if ever, confirmed.
Some writers, in the theoretical as well as the applied field of quantitative
genetics, have considered the genetic advances achieved from one stage of arti-
ficial selection (be it tandem or combined selection) as gains to be maintained
over a number of generations (HAZEL and LUSH1942; YOUNG1961). However,
even if both the genetic and phenotypic distributions are normal initially, any
method of directional selection theoretically will soon destroy the normality. As
pointed out by FINNEY (1962), it seems that the distortion will tend toward the
extreme when the heritabilities are high and the proportion of the population to
be selected is small. On the other hand, the genetic recombinations that occur
and the reassemblage of the genotypes and the phenotypes in the new generation,
together with the assumption of newly accumulated nongenetic variations may
tend to partially restore normality.
The general result of deviation from normality (such as skewness), caused by
selection, is usually to overestimate the true genetic gains when one is selecting
84 E. SCHEINBERG et al.
from the right hand side of a skewed distribution in the high direction (FINNEY
1962). This is true at least for uni-stage tandem selection. Theoretically, it is
known that artificial selection applied at any one generation changes the additive
genetic variance of the function under consideration. At the same time, it will
in the long run reduce the additive genetic variance of all correlated traits in
addition to modifying the covariances between these traits. The phenotypic vari-
ances and covariances are also expected to change.
In light of these theoretical considerations, it is easier to evaluate the results
obtained in this study. There are indications that some of the unexpected results
reported here may be due to one or a combination of the following reasons:
( 1 ) Development of skewness as selection progresses, (2) changes in genetic as
well as nongenetic parameters not accounted for in the restricted selection indices
used, ( 3 ) overestimated parameters used to construct the indices in generation 0,
and (4) the possibility of nonadditive components in the models assumed in this
investigation.
The development of skewness resulting from selection has not been checked
directly in the study and no experimental evidence is available from previous
studies. However, a comparison between the expected and realized selection
differentials may hold some useful information related to this source of variation
affecting the genetic gains. The absolute value of the expected selection differ-
ential was assumed to be 0.95 standard units (30.0 selected of 72.1). From Table
12 it is evident that the realized selection differentials were always smaller in
magnitude than the expected ones regardless of the selection method employed.
The deviation of the realized values from the expected ones is generally more
pronounced in populations I, 23, I, 1 3 , and I, Since the genetic gain in a given
trait, especially when one is using a restricted selection index, is a function of the
genetic and phenotypic variances and covariances as well as the selection differ-
entials, one must consider them jointly.
The fact that rather orderly changes occurred in some of the genetic and
phenotypic variances and covariances supports the inclusion of these changes as
a source of variation. This may account for some possible errors of selection and
thus unexpected genetic gains. The comparisons between the theoretical and
observed genetic gains are given in Table 9. These comparisons indicate that

TABLE 12
The m a n standardized selection differentials from the selection phase

Traits
Population Larval weight Develonmental time Puwl weight
SI 0.724 -0.353 0.547
s2 0.443 -4.697 0.250
s, 0.404 -4.193 0.723
11.2, 0.262 0.298 -0.234
12.1, 4.444 -0.078 -0.582
13.12 -0.357 -0.036 0.372
 RESTRICTED INDICES I N TRIBOLIUM 85
when using the selection phase parameter estimates the theoretical gains are in-
variably closer to the observed gains than are those when the parameter estimates
from generation 0 are used. These results and those given in Table 11 imply that
parametric changes occurred during the selection phase and affected the genetic
gains accordingly.
Confirmation that parameters were overestimated in generation 0 is made
when one compares the heritabilities and genetic correlation coefficients given in
Table 13 to those of the same parameters given in Table 6. All the realized esti-
mates are smaller in magnitude than the corresponding ones estimated in genera-
tion 0. The realized heritabilities are about one half the magnitude of those used
to construct the restricted selection indices. The differences between the realized
genetic correlations and those estimated from generation 0 are smaller than the
ones for the heritabilities. This is due to the fact that the genetic variances and
covariances involved are apparently overestimated by a similar multiplicative
factor. Similar values of the realized parameters from selection experiments when
using the same base population were given by ENGLERT and BELL (1964) and
HARDIN and BELL (1967).
The striking differences between estimated and reaiized heritabilities justify
the point brought up earlier in this article that the estimated genetic variances
contain more than just an additive component. Other €indings in this laboratory
(ENGLERT and BELL 1963; HARDIN and BELL 1967) indicate that at least for
larval weight there is considerable dominance variance. MCNARYand BELL
(1962) and BARTLETT, BELL and ANDERSON (1966) reported heritability esti-
mates for pupal weight from data on the same foundation population as used in
this study. Their estimates were devoid of dominance bias and were considerably
lower than those of this study. However, an earlier study, in which this same
population was used (BELLand MOORE 1958) reported heritability estimates for
pupal weight in the same range as found here (Table 6).
It is evident from Tables 8 and 9 that for most cases the realized responses f o r
the nonrestricted traits were smaller than either of the theoretically expected
responses. This should not discredit the model proposed by KEMPTHORNE and
NORDSKOG. On the contrary, it was shown, though not a priori planned, that in
most cases the formulated restriction yielded a negative response. This is true
for larval weight in populations I, and I,.,, and for pupal weight in populations

TABLE 13
Realized heritabilities and genetic correlations estimated from population S,, S, and S,.
Heritability estimates are those with standard errors

-
LW 0.30 f 0.01 -0.56 0.76
DT 0.26 f 0.01 -0.24
PW 0.41 f 0.01

I,\\': larral weight. DT: developmental time. PW: pupal weight.

86 E. SCHEINBERG et al.
 RESTRICTED INDICES I N T R I B O L I U M 87
I, 2 1 and I, A realized negative response with restricted indices was also ob-
served by ABPLANALP et al. (1963). They explained their results as a possible
consequence of changes in genetic parameters under selection rather than error
in their initial assumptions. However, no evidence was given to support this
possibility.
Using the realized estimates obtained from the uni-stage tandem selection new
restricted selection indices can be constructed with the same parametric weight-
ing coefficients previously used and compare them to the initial ones. Since the
genetic variances and covariances were overestimated by a similar multiplicative
factor, one should expect the resulting coefficients of a newly constructed re-
tricted selection index within a population to differ from the coefficients of the
previous index by a common factor. This is confirmed for each one of the restricted
indices as shown in Table 14. This, however, does not hold true when considering
the theoretically expected genetic gains given in Table 15. Since both the genetic
variances and covariances are smaller in magnitude the genetic gain will be
smaller though not necessarily in a multiplicative fashion.
The theoretical genetic gains given in Table 15 further justify the points
brought up earlier. It can be seen, without exception, that the theoretical genetic
gains calculated by the use of the realized parameters and the observed selection
differentials for the nonrestricted trait in population I, 2 3 , I, l i and I, 1, were

TABLE 15
The theoretical genetic gains per generation from the realized genetic parameters

Theoretical genetic gains

With a theoretical With the observed
Population selection differential selection differential

Larval weight
Sl 11.462 7.832
s, 3.266 4.680
s3 5.583 8.298
I,.,, 12.264 3.381
12.1, -12.556 -5.872
13.1, 13.771 -5.175
Developmental time
Sl -0.047 4.064
s2 -0.155 -0.122
s3 -0.023 -0.056
11.23 0.228 0.072
I,,,, -0.187 4.015
13.1, 0.180 0.007
Pupal weight
SI 3.199 3.618
s2 0.929 0.815
s3 7.693 6.652
'1.23 8.758 -2.153
12.1, 4.248 -5.051
13.12 8.669 3.398
88 E. SCHEINBERG et al.
invariably smaller than those calculated by the use of the realized parameters
and the theoretical selection differential. This is also true for the restricted pri-
mary traits in these populations. From Tables 8 and 15 it is shown that applying
the realized parameters calculated from populations S,, S, and S3 to populations
I, 2 3 , I, 19 and I,.,,, and the theoretical genetic gains are compared to the observed
gains, the results agree better than any previously attempted comparison.
I n light of the results reported here and the results reported by HARRIS ( 1963)
using a Monte Carlo technique, it seems that the use of a selection index based
on estimated population parameters should be limited to a few immediate gen-
erations. New estimates of the parameters required for index formulation should
be obtained after a few cycles of selection. This should be done even though the
propagated population is of a smaller size than the initial one. A new index
should be constructed and proper provisions for its use should be made after
taking into account the previously used index and allowing for any additional
changes in the expected genetic gains.
From a practical point of view it would be possible to construct restricted
selection indices with somewhat better control over the modification of growth
than was accomplished here. Inclusion of daily or total gain in the time interval
between the scoring of larval weight and pupal weight in the restricted indices
may be helpful. It is possible that erroneous assumptions about any of these
parameters may be less critical for selection decision if additional attributes are
incorporated in the selection criterion. Thus, more stable predictions as well as
better response might be obtained from an index based on more than three traits.
From the results reported for populations I, 23, I, 13, and I,.,, the following
facts can be established. (1 ) The attempted restrictions of larval weight and pupal
weight invariably were not realized. Instead, the realized response was negative.
(2) Developmental time responded fairly well to the restriction in populations
I,,,, while showing a significantly positive response in population I,,,, which is
contrary to the initial formulation of no response. ( 3 ) When comparisons are
made between the realized response for any of the three primary traits and
either one of the theoretically expected ones, no specific pattern can be established
at present. This is evident from the results given in Table 9. (4) The number of
larvae has changed significantly at the 0.05 level of probability in all populations
except I?13. The change was negative in population I,,,, and positive in all others.
Although the changes in this trait w-ere significantly different in five out of the
six populations studied, they were not large enough to cause a fitness problem in
the first ten generations of selection. This is true for the negative realized response
where the change was -0.101 larvae per generation or -1.01 larvae over ten
generations of selection. (5) From Tables 8 and 9, it is evident that the theoretical
genetic gains per generation estimated from the selection phase never exceeded
the theoretical gains estimated in generation 0 for all traits except the restricted
ones.
The authors wish to express their sincere appreciation to: MRS. SHARON DITTMAR,MRS.
NORMA HANCOCK, and MRS.DORIS SHIDELERfor technical assistance and for making the collection
of these data an enjoyable task; MRS. SHIRLEY WOLFEfor advice and many useful suggestions
 RESTRICTED INDICES I N T R I B O L I U M 89
regarding data processing techniques; MR. JOHNSTEELEfor securing computer time and advice
regarding computer systems; MR. STANLEY ROBBINSfor many useful suggestions and instructions
in data processing, and to DR. M. J. LAWRENCE for helpful comments in the final preparation
of this manuscript.

SUMMARY

Uni-stage tandem selection and restricted selection indices were practiced in

six experimental populations. A random mating control population with no selec-
tion applied was also maintained. These populations, each replicated three times,
were selected for ten generations. The three primary traits measured were: larval
weight, developmental time and pupal weight. I n addition, two secondary traits,
number of larvae per pair mating and daily gain were scored.-The genetic
gains obtained in each population were compared to their theoretical expectations.
The results obtained, in general, confirm the biological validity of the KEMP-
THORNE-NORDSKOG model for restricted selection indices. The attempted restric-
tion for larval weight and pupal weight resulted in negative response in all cases.
Restriction of developmental time was successful in one case.-The genetic gains
were lower than their theoretical expectations in most of the selected populations.
The results are explained on the basis of overestimation of genetic and pheno-
typic parameters used for index construction as well as expected selection differ-
entials. In addition, the change in parameters and the presence of nonadditive
terms in the models used were considered as a cause of some of the unexpected
results.-Natural selection and reproductive fitness were no problem in this study,
as the weighted and unweighted selection differentials were found to be essen-
tially equal. Thus, the realized genetic gains were attributed mainly to the opera-
tion of artificial selection in each one of the observed populations.-Caution
should be taken to account for changes in parameters during selection in addition
to using other designs and methods of estimation iwhich will provide less biased
estimates. Inclusion of more traits in such an index together with partial rather
than complete restriction should be studied.

LITERATURE CITED

ABPL~NALP. H., F. X. OGASAWARA, and V. S. ASMUNDSON, 1963 Influence of selection for body
weight at different ages in turkeys. Brit. Poul. Sci. 4: 71-80.
BARTLETT.A. C., A. E. BELL, and V. L. ANDERSON, 1966 Changes in quantitative traits of
Tribolium under irradiation and selection. Genetics 54: 699-713.
BELL,A. E., and C. H. MOORE,1958 Further comparisons of reciprocal recurrent selection with
conventional methods of selection for the improvement of quantitative characteristics. Proc.
10th Intern. Congr. Genet. 2 : 20-21.
BRAY,D. F., A. E. BELL,and S. C. KING, 1962 The importance of genotype by environment
interaction with reference to control populations. Genet. Res. 3: 282-302.
ENGLERT, D. C., and A. E. BELL,1963 Genetic differences in the growth curve of Tribolium
castaneum. Growth 27: 87-99. - 1964 Selection and growth in Tribolium castaneum
(Abstr.). Genetics 50: 247.
FINNEY,D. G., 1962 Genetic gains under three methods of selection. Genet. Res. 3: 417-423.
90 E. SCHEINBERG et al.
HARDIN, R. T., and A. E. BELL,1967 Two-way selection for body weight in Tribolium on two
levels of nutntion. Genet. Res. 8 (In press).
HARRIS,D. L., 1963 Influence of errors of parameters estimation upon index selection StatisticaZ
Genetics and Plant Breeding. N.A.S.-N.R.C. Publ. 982: 491-500.
HAZEL,L. N., 1943 The genetic basis for constructing selection indices. Genetics 28: 476-490.
HAZEL,L. N., and J. L. LUSH, 1942 The efficiency of three methods of selection. J. Hered.
33: 393-399.
KEMPTHORNE, O., and A. W. NORDSKOG, 1959 Restricted selection indices. Biometrics 15: 10-19.
MCNARY,H. W., and A. E. BELL, 1962 The effect of environment on response to selection for
b3dy weight in Tribolium castaneum (Abstr.) Genetics 47: 969-970.
OSBORNE, R., and W. S. B. PATERSON, 1952 On the sampling variance of heritability estimates
derived from variance analyses. Proc. Roy. Soc. Edinburgh 64: 456461.
RASMUSON, M., 1964 Combined selection for two bristle characters in Drosophila. Hereditas
51: 23-256.
SCHEINBERG, ELIYAHU,1966 The sampljng variance of the correlation coefficients estimated in
genetic experiments. Biometrics 22 : 187-191.
SMITH,H. F., 1936 A discriminant function for plant selection. Ann. Eugen. 7:240-250.
TALLIS, G. M., 1959 Sampling error of genetic correlation coefficients calculated from analyses
of variance and covariance. Australian J. Stat. 1: 3543. __ 1962 A selection index
for optimum genotype. Biometrics 18: 120-122.
YOUNG,S. S. Y., 1961 A further examination of the relative efficiency of three methods of
selection for genetic gain under less restricted conditions. Genet. Res. 2: 106-121.