Morphometrics Honeybee: Himalayan Region
Morphometrics Honeybee: Himalayan Region
Morphometrics Honeybee: Himalayan Region
Summary — Multivariate statistical analyses of 55 morphometric characters were made for collections
of Apis cerana from 20 localities in the Himachal Pradesh and Jammu and Kashmir states of the
northwest Himalayas. Two distinct biometric groups were observed after discriminant function analy-
sis and cluster analysis. The groups were associated with differences in the climates of the Himachal
and Kashmir regions. Previous observations were confirmed that some characters were associated with
altitude and rainfall such that bees were larger and darker at higher altitudes and lower rainfall. The phe-
netic clustering of samples within each region corresponded in general to physiography. Principal
component analysis indicated minimal contributions of general size components to overall covariation
in each of the regions.
*
Correspondence and reprints
MATERIALS AND METHODS among the 20 localities (Sneath and Sokal, 1973).
On this basis, two major biometric groups were
Collections of worker bees were made from 4-5 recognized.
wild colonies at each of 20 localities. The total The purpose of the second phase of analy-
number of bees from each locality and other infor- ses was to evaluate the discrimination between
mation are given in table I (see also Mattu and the two major biometric groups using all 55 char-
Verma, 1983). The 55 characters were those for acters. The measurements for the collections
which univariate analyses were previously were regrouped into the two biometric groups
reported by Mattu and Verma (1983, 1984a, b). and entered in a second series of DFAs. ANOVA
Code numbers for the characters and brief for each character was performed again, now for
descriptions of the measurements are provided in the two groups. Two DFA methods were used:
table II. the ’direct’ method in which all characters were
included and the ’stepwise with Wilks criterion’
Computations were performed on an IBM
method in which certain characters were auto-
3090 computer at the University of California,
Berkeley. For the discriminant analyses the pro- matically eliminated if they failed to meet certain
critical values. The discrimination between the
grams DISCRIMINANT (prior probabilities equal;
minimum tolerance in stepwise procedure was groups was evaluated by reclassifying each bee
into one of the two biometric groups. With two
0.001) and FACTOR from the Statistical Pack-
age for the Social Sciences (Norusis, 1985) was groups of equal prior probability, 50% of the bees
used. The program NTSYS-PC (Rohlf, 1987) was could be correctly classified by chance alone.
used to perform unweighted pair-group cluster Finally, the odd numbered samples were held as
a test set and only the even numbered samples
analyses with arithmetic averages (UPGMA) as
described by Sneath and Sokal (1973). All other were used as data for the DFA. Then all sam-
computations were made with special programs ples were classified into the two groups. Next,
written by HVD and performed on an IBM-AT per- the even numbered samples were held as a test
sonal set and so forth. The classification of the test sets
computer.
in each case provided further evidence of the
The statistical analyses were performed in
effectiveness of discrimination.
four phases in a manner similar to that described
The purpose of the third phase of the analyses
by Singh et al (1990). The purpose of the first
was to evaluate the contributions of the different
phase was to make an initial DFA of the 20 local-
ities to determine how the collections of bees characters to the discrimination of the two bio-
from the 20 localities could be clustered into a metric groups. This was done by two methods:
smaller number of groups. The measurements 1) by ranking of the absolute values of the stan-
of all characters for the collections were grouped dardized discriminant function coefficients; and
by locality and entered as 20 groups in a DFA. 2) ranking the absolute values of the correlations
The program included a univariate ANOVA for of the 55 characters with the single discriminant
each character over the 20 groups. The extent function. As a result of these comparisons, a
to which the DFA discriminated among the local- reduced number of characters was selected to
ities was evaluated by reclassifying the 1184 indi- provide a quick method for identification of the
vidual bees with the coefficients from the analysis. biometric groups. The final choice was supported
Each bee was assigned to the group for which by the procedure described above where odd or
the bee had the highest probability of member- even samples were held as test sets. ANOVAs
ship. With 20 groups of equal prior probability, were also made for the 55 characters within each
5% could be correctly classified by chance alone. biometric group to examine the differences in the
The mean value of each character at each characters within the two groups.
locality wasthen used to compute the coordi- The purposes of the fourth phase were: 1) to
nates of the centroids for each locality with respect determine the relationships of the morphometric
to the 19 canonical discriminant functions. From characters to altitude and rainfall; and 2) to deter-
the coordinates, the square roots of the Euclidean mine the contribution of a general size compo-
distances between each pair of localities were nent to overall variation within each biometric
computed. The smaller the distance, the more group. First, univariate product-moment correla-
similar are the biometries of the bees collected tions (r) were computed for altitude and rainfall
at two localities. A hierarchic UPGMA cluster at the localities (table I) versus the values of the
analysis was made of the Euclidean distances 55 morphometric characters for individual bees.
Second, a PCA was performed with altitude and were used. Because the two methods gave
rainfall included as variables with the 55 charac-
essentially the same result, we provide here
ters for individual bees. Because the two bio-
the statistics from the direct method: unstan-
metric groups were so different, the computa-
dardized coefficients for use with raw mea-
tions of correlations and PCA were carried out
within each biometric group rather than overall. surements and standardized coefficients for
use with standardized measurements (table
IV; see table III for total means and stan-
RESULTS dard deviations). The ANOVA of the two
groups gave highly significant F values for
45 of the 55 characters (significance prob-
Discriminant analysis ability < 0.001; see table VI for list).
In the third phase, the characters with
In the first phase, the initial DFA indicated standardized discriminant coefficients of
that a majority of the bees from each of the absolute value > 0.35 (an arbitrarily chosen
20 localities were closer to other bees from value) on the single function were characters
the same locality than to bees from other 48, 25 and 46. These characters also had
localities when placed in a 19-dimensional correlations with the function of absolute
morphometric space. This was shown by value > 0.33 (see table IV). When only char-
the reclassification of the individual bees acters 48 and 25 were used in a DFA
into the 20 groups by two DFAs with differ- between the biometric groups and all bees
ent procedures. The direct method, where all were included as data, the overall percent-
55 characters were included, gave an over- age of correct classification was 99.9%.
all average of 61.15% of bees correctly clas- When samples were split into test and data
sified and the stepwise method gave sets, the percentages of correct classifica-
61.23% correctly classified. The univariate tions for the data sets were 100% (even
ANOVA indicated that 48 of the 55 charac- numbered localities) and 99.8% (odd num-
ters over all 20 localities were highly signi- bered localities). For the corresponding test
ficant (significance probability < 0.001; see sets, 99.8% and 100%, respectively. Table
table VI for list). V provides statistics for a quick classification
The biometric relationships of the bees of new samples with only two characters.
from the 20 localities are shown in the When classification of new samples into the
two biometric groups is critical, then the use
phenogram (fig 1).Two distinct groups are
evident: samples 51 to 62 from Himachal of all 55 measurements is recommended.
region form one biometric group, and sam-
ples 63 to 70 from Kashmir region form a
second biometric group. The two biometric Variation and correlation of characters
groups are so distinct that clearer pictures of
the three-dimensional relationships of the The several ANOVAs that were performed
samples in each region are best shown in for different groupings of the data indicate
separate illustrations (figs 2,3). that forewing characters 10 and 14, tongue
The second phase was a DFA of the two characters 34-36, and antenna character
biometric groups with all 55 characters. 53 were not different among the samples
When all samples were entered as data, or groups, while forewing characters 2, 4,
the overall classification rate was 99.9% 7, 8, and 17, hindwing character 29, and
regardless of whether all 55 characters were abdomen character 41 consistently differed
used in the direct method or just the 39 char- between the regions and among the sam-
acters selected by the stepwise procedure ples within each region (table VI). ANOVAs
within each region indicated that 31 char- antenna characters. Most correlations with
acters of bees of the Himachal region and altitude were positive and most with rainfall
16 characters of bees from the Kashmir were negative. In other words, higher alti-
region were different among the samples. tude and lower rainfall were correlated with
Bees within the latter region, therefore, larger size within each region. Twenty-two
exhibited the least differences among the characters were consistently correlated with
samples. altitude in both groups, but only character
The correlations of rainfall and altitude 4 was correlated with rainfall in both groups.
with morphometrics of individual bees were As observed previously, bees of the
examined separately within each region by Kashmir region are larger in length and width
univariate and multivariate analysis (tables measurements than those of the Himachal
VI, VII). A univariate correlation existed region (Mattu and Verma, 1983, 1984a, b).
between rainfall versus altitude among local- This was confirmed by comparing the
ities of the Himachal region (r = -0.38, sig- means of the characters (table III) after
nificance probability = 0.00), but no corre-
excluding characters 9-19 which are angles
lation existed among the localities of and characters 10, 14, 18, 21, 26, 34, 35,
Kashmir region (r= 0.02, significance prob-
36, 43, and 53 which are not significantly
ability 0.33). These relationships also
=
different between the groups. The excep-
appeared in the PCAs where altitude and tional characters were 38, 40, and 48, where
rainfall jointly contributed strongly to PC1 on the average, bees of the Himachal region
in the analysis of the Himachal region, but
were unrelated and divided in their contri-
are larger.
butions to PC1-PC3 in the analysis of the The PCAs of morphometrics within each
Kashmir region (table VII). region (table VII) shared certain features in
common: about 4% of the variation was
For bees of the Himachal region, 35 mor-
accounted for by PC1 in each analysis and
phometric characters had significant corre- < 2% was accounted for by each of the
lations with altitude (two-tailed significance
probability < 0.01).All but two of these were remaining components. As a result, the
cumulative percentages of variances
positive. The highest correlations (> 0.3, an
arbitrary value) were found for forewing explained by components 1-23 were only
characters 2 and 5; hindwing character 24; 55.9-58.9%. Altitude had the highest cor-
hind leg character 30; and abdomen char- relations of all variables with PC1. Morpho-
acters 41, 50, and 51. Fifteen characters metric variables that had higher correlations
had significant, negative correlations with with PC1 in both analyses were forewing
rainfall of which abdomen characters 41 and characters1, 5 and 29; hind leg characters
50 were < -0.17. 30 and 31; and abdomen character 41.
For bees of the Kashmir region, 28 char- Otherwise, the two PCAs shared little in
common with regard to the distribution of
acters were correlated with altitude; all but
three were positive. Among the highest were the higher correlations among the compo-
nents.
forewing character 7 and hindwing charac-
ter 26 at > 0.29. Seven characters were cor-
related with rainfall, of which five were neg-
ative. Forewing character 6 and abdomen DISCUSSION
character 47 were highest at < -0.18.
In summary, most correlations with alti- Two major biometric clusters emerged when
tude and rainfall within a region were found all 20 samples were analyzed in a single
in wing, leg, and abdomen characters, while DFA (fig 1).The 8 samples from the Kash-
relatively few were found among tongue and mir region formed one biometric group and
the 12 samples from 1 the Himachal region in which bees of the Kashmir region are
formed the second group. The morphome- smaller. Bees of the Kashmir region also
tric ’gap’ between the groups was quite dis- had wider bands of dark color on the third
tinct. Samples from localities along the and fourth terga and a narrower light band
boundary between the regions did not on the third tergite. However, the charac-
appear to form a transitional cline. For exam- ters of the tongue were complex: length and
ple, bees from sample 63 (table I) taken at breadth of prementum and total tongue
a lower elevation in western Kashmir region length were not different between the
did not cluster with bees taken at lower ele- groups, length of labial palp and glossa were
vations in the western Himachal region. Vice larger in the Kashmir region, while length
versa, bees from sample 58 taken at a of the postmentum was smaller in the Kash-
higher elevation in the northern Himachal mir region.
region did not cluster with bees from sample Within each region, correlations also
64, also taken at a high elevation, in the existed between morphometric characters
southern Kashmir region. The overall lack of and altitude and rainfall. Larger size in a
similarity in the distribution of correlations majority of length and breadth measure-
with the first three principal components of
ments was associated with higher altitude
morphometric covariation in each region is and lower rainfall. These results agree in
further evidence of the difference between
bees in the two regions. general with previous analyses, but differ
in some details because of differing com-
Although no striking physiographic fea- putational procedures (Mattu and Verma,
ture separates these regions, a major cli-
1983, 1984a, b; Verma and Mattu, 1982;
matic boundary does coincide with the polit- Verma et al, 1984). Except for the correla-
ical boundary between the states. The tions with altitude, and to a lesser extent
rainfall statistics alone do not reveal this cli- with rainfall, covariation among the mor-
mate difference (table I). However, based
phometric characters themselves as
on precipitation effectiveness (total monthly revealed by PCA was remarkably low. This
precipitation divided by total monthly eva- resulted in a minimal contribution of ’gen-
poration), the climate classification of Taka- eral size’ to the overall covariation among
hashi and Arakawa (1981) shows the col-
the samples within each region. The first
lection sites of the Kashmir region are in a
zone of dry sub-humid climate while those of
component usually accounts for a much
the Himachal region are in zones of moist larger percentage of variance in morpho-
metric studies (Daly, 1992).
sub-humid or humid climates. Analysis of
future collections along this climatic bound- Within each biometric group, clusters of
ary will be necessary to discover where the samples in morphometric space were often,
transition from one biometric type to the but not always, close to each other geo-
other occurs or if the two groups coexist graphically. In the Kashmir region, a close
sympatrically. In the latter case, an appro- correspondence existed between the minor
priate taxonomic action would be in order. clusters (figs 1,3) and physiographic fea-
The two biometric groups were distin- tures: sample 68 from the Kishangana Val-
guished by significant differences in 45 of ley, sample 69 from the Sindh Valley, and
the 55 characters. Bees of the Kashmir sample 70 from the Drass Valley, are all
region were larger on the average than from contiguous river valleys at altitudes
those of the Himachal region in most length above 2300 m. Another cluster of samples
and breadth measurements; exceptions 65, 66, and 67 was taken in the Kashmir
were characters 38 (length of postmentum) Valley, just to the south of the first cluster
and 48 (length of wax mirror on 3rd sternite) and at altitudes of 1700-2000 m. All six
samples were united in a larger cluster at ples 52, 53, 55, and 56 was taken in the
phenon 5 and all were taken north of the region of the Beas River and limited on the
Pan Panjal Range which produces a modi- north by the Dhaula Dhar Range. An excep-
fied monsoon climate. Samples 63 and 64 tion was sample 54 which was morphome-
were isolated morphometrically and the col- trically close to sample 56, but was taken
lection sites were separated geographically. much farther south. Likewise, samples 57
However, these two sites share lower alti- and 58 were clustered morphometrically,
tudes and the normal monsoon climate but were not close geographically. Finally,
which prevails south of the Pan Panjal sample 51, isolated from other samples mor-
Range and in the Chenab River Valley that phometrically, was taken at the lowest alti-
penetrates the range. tude.
In the Himachal region, most samples in
the minor clusters can be related to physio-
graphy and the geographic proximity of mor- ACKNOWLEDGMENTS
phometrically related samples, but some
samples cannot. The cluster of samples 59, The authors are extremely pleased to contribute
60, 61, and 62 were taken from a zone of to this special issue in honor of Professor Ru-
humid climate, dominated by the Sutlej River ttner. This research was supported in part under
Grant No: 367-5600-G-SS-1147-00, Programme
valley, and stretching from Simla to the Kin- in Science and Technology Cooperation, Office of
naur region. The remaining samples of 1
Science Adviser, US Agency for International
the Himachal region were from the moist
Development. Funds for computation were pro-
sub-humid zone or along the its boundary vided by the Agricultural Experiment Station of
with the humid zone. The cluster of sam- the University of California.
Deutsche Version
Zusammenfassung — Es wurden Bienenproben der Art Apis cerana an 20 verschiedenen Orten in
den Staaten (Provinzen) des Nordwest Himalaya gesammelt: Himachal Pradesh, Jammu und Kash-
mir. Von jeder Biene wurden 55 Merkmale gemessen und mit multivariaten statistischen Methoden bear-
beitet. Mit Hilfe der Diskriminanz- und Clusteranalyse konnten zwei verschiedene morphometrische Grup-
pen unterschieden werden, die jeweils mit klimatischen Unterschieden assoziiert waren. Bereits
bekannte Befunde konnten bestätigt werden. Bei großer Höhe und geringerem Niederschlag waren die
Bienen größer und dunkler. Die Verteilung der Proben in phenetische Cluster innerhalb jeder Region
zeigte eine allgemeine Übereinstimmung mit den topographischen und klimatischen Bedingungen.
Allgemeine Größenmerkmale trugen, wie die Principal Component Analysis zeigte, nur sehr wenig
zur Gesamtkovarianz in jeder Region bei.
Prozentsatz der richtigen Klassifizierung 0,00). Für Kashmir konnte diese Korrela-
99,9%. Bei Trennung der Proben in Test- tion nicht gezeigt werden (r 0,02; P
= =
und Datensets war die richtige Klassifizie- 0,33). Diese Beziehungen traten deutlich
rung für die Datensets 100% (für gerade bei den PCAs in Erscheinung. Bei der Ana-
Probennummern) und 99,8% (ungerade lyse der Himachalregion trugen Nieder-
Probennummern). Für die zugehörigen Test- schlag und Höhe gemeinsam sehr stark zu
sets errechneten sich 99,8% und 100%. In PC1 bei. In Kashmir waren dagegen beide
Tabelle V finden sich die Angaben für das Faktoren unabhängig voneinander und teil-
statistische Schnellverfahren zur Klassifi- ten sich unter PC1-PC3 auf (Tabelle VII).
zierung von neuen Proben mit nur zwei Für die Bienen der Himachalregion zeig-
Merkmalen. Wenn allerdings die Klassifi- ten 35 Merkmale eine signifikante Korrela-
zierung von neuen Proben in die beiden bio- tion zur Höhenlage (zweiseitige Verteilung,
metrischen Gruppen nicht klar ist, soll eine P < 0,01), wobei mit Ausnahme von zwei
Berücksichtigung von allen 55 Merkmalen Merkmalen alle Korrelationen positiv waren.
empfohlen werden. Die höchsten Korrelationen (> 0,3; willkür-
licher Wert) ergaben sich für den Vorder-
flügel (2 und 5), den Hinterflügel (24), das
Hinterbein (30) und das Abdomen (41, 50
Variationsbreite und Korrelation
und 51).Bei 15 Merkmalen ergaben sich
signifikante negative Korrelationen mit dem
Es wurden ANOVAs mit mehreren ver- Niederschlag, wobei die Abdomenmerkmale
schiedenen Gruppierungen von Merkma- 41 und 50 eine Korrelation < -0,17 zeigten.
len durchgeführt. Dabei stellte sich heraus, Bei den Kashmir-Bienen waren 28 Merk-
daß die Vorderflügelmerkmale 10 und 14, male mit der Höhe korreliert, wobei nur 3
die Rüsselmerkmale 34-36 und das Anten- Merkmale eine negative Korrelation hatten.
nenmerkmal 53 sich nicht zwischen den Die höchsten Werte (> 0,29) errechneten
Proben oder den Gruppen unterschieden. sich für Vorderflügelmerkmal 7 und Hinter-
Dagegen zeigten sich bei folgenden Merk- flügelmerkmal 26. Mit der Niederschlags-
malen konsistente Unterschiede zwischen menge waren 7 Merkmale korreliert, wobei
den Regionen und den Proben innerhalb in 5 Fällen negative Werte gefunden wur-
der Regionen: Vorderflügelmerkmal 2, 4, 7, den. Im Vorderflügel zeigte Merkmal 6 und
8 und 17, Hinterflügelmerkmal 29 und Abdo- am Abdomen Merkmal 47 eine Korrelation