phenotypes which may represent directexpression of the genotype(s) or, alterna-tively, reflect a greater or lesser influenceof environmental factors [5, 37, 38] or con-stitute products of the norm of reaction [6].The mountain honeybees of Africa supply arich source of these difficulties [13], andform the basis of this study.

To illustrate these difficulties we shallrecapitulate briefly a few examples. In 1961,Smith [35] reported the occurrence of large

1. INTRODUCTION

One of the most intriguing problemsrelated to the delineation of honeybee (Apismellifera L.) populations is that preciselyquantitative definitions of subspecific taxawhich also have biological meaningfulnesshave proven extremely elusive [13]. Thesedifficulties are exacerbated by the constraintsinherent in the typological Linnean systemof nomenclature [3, 4] and the fact that cat-egories have always been defined in terms of

Original article

Mountain honeybees of Africa

H. Randall HEPBURNa*, Sarah E. RADLOFFb, Sunday OGHIAKHEa

a Department of Zoology and Entomology, Rhodes University, Grahamstown, South Africab Department of Statistics, Rhodes University, Grahamstown, South Africa

(Invited paper)

Abstract – The honeybees occurring along transects from low to high altitude were analysed forseven separate mountain systems in Africa using three suites of characters: morphometric characters,flight dimensional measurements and the mitochondrial DNA (mtDNA) restriction length fragmentsderived from the non-coding region of COI-COII by DraI restriction. Morphocluster definition wasconsistent with mtDNA cluster membership but not with flight dimensional data. When all threecharacter suites are combined, six different kinds of unrelated mountain bees are obtained. The onlycommonality among the mountain bees is that they are larger than those of lower altitudes. Becauseof fundamental differences in the restriction length fragments and other clusters obtained, it is con-cluded that mountain bees should probably be regarded as ecotypically differentiated populations ofthe subspecies surrounding each particular mountain.

Apis mellifera/ Africa / mountain / morphometrics / flight dimension / mtDNA

Apidologie 31 (2000) 205–221 205© INRA/DIB-AGIB/EDP Sciences

* Correspondence and reprintsE-mail: r.hepburn@ru.ac.za

H.R. Hepburn et al.206

and dark bees at about 2 000 m on the slopesof Mounts Kilimanjaro and Meru in Tanza-nia, which he named Apis mellifera monti-cola. There were, however, forms of inter-mediate size between these bees and othersmaller bees lower down the mountainswhich were named A. m. scutellata. Ruttner[31, 32], Meixner [18] and Meixner et al.[20] subsequently extended these studieswith multivariate techniques and concludedthat “the distinct monticolaareas of todayrepresent the refugia of a former much largercoherent distribution across all of the high-lands of East Africa” [19, 31]. This equatesto a concept of an archipelago of relictualbees constituting the subspecies A. m. mon-ticola. This interpretation was supported byadditional analyses of allozymes coupled tomorphometric analyses [19, 20]. However,evidence for such an archipelago (asopposed to the Tanzanian case) of A. m.monticolatype bees using allozymic [21,33], morphometric and mitochondrial DNA(mtDNA) [8, 10, 33, 34] analyses of hon-eybees of other mountainous systems inAfrica were equivocal in this regard.

Put another way, one can ask the questionwhether there is a shared genetic heritageamong honeybee populations at high alti-tudes in the different mountain systems ofthe African continent that would justify con-sidering them more closely related to eachother than to their surrounding or nearbylowland neighbours. This is the purpose ofthe present paper.

2. MATERIALS AND METHODS

2.1. Honeybees

Recently (1997) the morphometricdatabases on honeybees of the Institut fürBienenkunde (Ruttner Collection at Oberursel,Germany) and of the Apiculture Group atRhodes University (Hepburn and RadloffCollection, Grahamstown, South Africa)were amalgamated to form a single databasefor the continent of Africa. These combined

data, augmented by more recent measure-ments of honeybees from Zimbabwe,Malawi, Lesotho and Tanzania, were used toanalyse the honeybee populations in twodifferent ways. First, a full morphometricanalysis of 14 973 worker bees from825 colonies at 193 localities in East Africaextending from South Africa to Ethiopiawas made [26]. But because of the severeconsequences of sample size and samplingdistance limitations (see Discussion) inmasking small biometric groups [25], a morerestricted analysis was made on a transectbasis. The localities and transects are shownin Figure 1.

Each of the transects was an average of1 000 km in length and extended (usually)from near sea level to above 2 000 m altitudein the mountain systems of transect 1(Ethiopia), transect 2 (Cameroon), transect 3(Tanzania), transect 4 (Malawi), transect 5(Zimbabwe/Mozambique), transect 6 (SouthAfrica/Lesotho) and transect 7 (Namibia).Some of the transects simply extend fromlow to higher altitudes, but others extendfrom low to higher and then low altitudesagain on the opposite side of the mountainsystem (Tab. I, Fig. 1). All of the workerbees used in this part of the study were sam-pled from small-scale, fixed-site beekeep-ing colonies at 31 localities covering allmajor regions of Africa except the Maghrebin the northwest of the continent. While‘captive’ colonies were used, it must beunderstood that these were simply beesattracted to empty hives from the wild pop-ulation. In most parts of Africa bees are sel-dom transported and bee breeding is virtu-ally non-existent. Thus the samples used inthis study are authentic samples of unadul-terated wild honeybees typical of the areasunder consideration.

2.2. Measurements

Three distinct and independent sets ofmeasurements were made: a series of mor-phometric measurements to determine

Mountain honeybees

COII region of cytochrome oxidase inmtDNA.

For the morphometric studies the samenine characters used in previous studies ofhoneybees in Africa were measured [2, 13,

morphocluster membership and affinity ona continental and regionally restricted basis,aerodynamic measurements of flight-relatedparameters which provide a power effi-ciency index, and analyses of the restrictionlength fragments for non-coding COI and

207

Figure 1.Distribution of the seven transects through different mountainous regions of Africa and thelocalities in them that were sampled: Localities given by reference numbers are as follows: Tran-sect 6(South Africa): 1. Port Alfred, 2. Grahamstown, 3. Hofmeyr, 4. Tarkastad, 5. Queenstown,6. Dordrecht, 7. Sterkstroom, 8. Burgersdorp; (Lesotho): 9. Quthing, 10. Thaba-Tseka, 11. Semonkong,12. Mokhotlong; Transect 7(Namibia): 13. Keetmanshoop, 14. Mariental, 15. Windhoek, 16. Oka-handja; Transect 5(Mozambique): 17. Beira; (Zimbabwe): 18. Mutare, 19. Harare, 20. Karoi; Tran-sect 4(Malawi): 21. Chikwawa, 22. Thyolo, 23. Rumphi, 24. Chilinda, 25. Chitipa; Transect 3(Tanzania): 26. Tanga, 27. Kasungu, 28. Mt. Meru; Transect 1(Ethiopia): 29. Holeta, 30. DebreMarkos, 31. Bahir Dar, 32. Gonder, 33. Adi Arkay; Transect 2(Cameroon): 34. Mamfe, 35. Bamenda,36. Kumbo, 37. Banyo, 38. Gouna.

H.R. Hepburn et al.

22]. Their Ruttner numbers [31] are given inbrackets as follows: length of cover hair ontergite 5 (1), width of wax plate on sternite3 (11), transverse length of wax plate onsternite 3 (13), pigmentation of scutellum(35), pigmentation of scutellar plate (36),pigmentation of tergite 2 (32), wing angleB4 (22), wing angle N23 (30) and wingangle O26 (31).

As noted above, Ruttner [31] concludedthat as few as 10 characters would be suffi-cient to morphometrically discriminateAfrican races of honeybees, and this wasalso borne out by Crewe et al. [2]. Ruttner[31] demonstrated that between 10 and20 bees would be a sufficient sample sizefor morphometric statistical analysis. Thus,for each colony, 20 bees served as the stan-dard sample size.

Flight-related parameters from which apower efficiency index was calculated weremeasured as follows. Worker bees that hadbeen collected in alcohol were subsequentlydissected to separate the wings from the tho-rax and the latter from the other body parts,after which all were weighed on a microbal-ance to constant dry mass. On dissection,the digestive system was discarded andreplaced by a ‘clean’ dry weight gut measure(details in [14, 15]). The four wings of eachbee were slide-projected on a digitizingtablet and scanned to measure total wingsurface area. Finally, values for wing sur-face area (S), whole body mass (M), wingloading (W = M/S), thorax mass (m) toM ratio (r = m/M) and an excess powerindex (EPI) were calculated. The EPI forhoneybees is a measure of the maximumpower available to the bee over that requiredto maintain equilibrium in steady level flight,and is defined as: EPI = √(r2/W) and wasderived for honeybees from the general the-ory of flight [17]. Flight dimensional mea-surements were taken on 12 bees per colony.

2.2.1. mtDNA analysis

Due to a limited availability of speci-mens, only some transects (South Africa,

Zimbabwe, Malawi and Ethiopia) were sub-jected to DNA analysis.

2.2.2. DNA extraction

Genomic DNA was phenol-extractedfrom individual worker thoraxes (n = 5 percolony) following a modified protocol [7].Ethanol-preserved workers were initiallyincubated with agitation in insect Ringersolution (127 mM NaCl, 1.5 mM CaCl2,5 mM KCl, pH 7.4 with NaOH) for 5, 10and 15 min at room temperature beforeextraction. Then, DNA was phenol-extractedfrom the individual thoraxes. Individualworker DNA was resuspended in 30 µLDDH2O. To account for unspecific restric-tions, DNA was electrophoresed on stan-dard 1% agarose gels. The individual DNAsamples were then pooled for each colony.

2.2.3. PCR conditions and DraIrestrictions

The mtDNA fragments (including theCOI-COII intergenic region) were ampli-fied using the previously reported proce-dures with primers E2 and H2 [9, 10].Amplification products were elec-trophoresed on standard 1% agarose gels toseize the fragments. Then, the amplifica-tion products were restricted with DraI fol-lowing routine protocols [9, 10]. Restric-tion fragments were separated in 5 and 10%acrylamide gels and UV-visualised afterstaining in ethidium bromide. Restrictionfragment patterns were classed based on thesize of the fragments. Fragments smallerthan 100 base pairs (bp) were not consid-ered.

2.3. Data processing

The colony means of the morphometricand flight dimensional data were analysedusing factor analysis and discriminant anal-ysis procedures. Wilks’ lambda statistic wasused to test for significant differencesbetween the vector of means of the

208

Mountain honeybees

Namibia and Zimbabwe/Zambia transects[24, 28, 29].

Because smaller biometric groups areoften swamped in large regional analysesand because additional material has becomeavailable from transect 3 (Tanzania), tran-sect 4 (Malawi) and transect 6 (Lesotho),new morphometric analyses were performedon the honeybees of these new transects.That from sea level in South Africa to thehigh mountains of the Drakensberg inLesotho (transect 6) yielded two quite dis-tinct morphoclusters, with a 100% correctclassification for the bees of Lesotho on theone hand and those within South Africa onthe other [27]. In the case of Malawi (tran-sect 4), two morphoclusters were found, andthe mountain form at Chilinda yielded a cor-rect classification of 92.0% and the beesfrom neighbouring lower altitudes with91.3%. Here we present results of the anal-yses of morphometric data from all sevenmountain transects thus far examined.

A factor analysis using the colony meansof the nine morphometric charactersof worker honeybees from 193 coloniesfrom the localities shown in Figure 1 andin Table I isolated four factors with eigen-values greater than 1: factor 1, hair length ontergite 5, transverse length of wax plate onsternite 3, wing angle N23 and pigmenta-tion of tergite 2; factor 2, width of wax plateon sternite 3, pigmentation of scutellum andscutellar plate; factor 3, wing angle O26;factor 4, wing angle B4. These four factorsaccounted for 72% of the variance in thedata. The loading for each character had anabsolute value greater than 0.6. The factorscores graph revealed three morphoclusters,those colonies from transect 1 (Ethiopia)forming one cluster, those colonies fromtransects 2 (Cameroon), 3 (Tanzania),4 (Malawi), 5 (Zimbabwe/Mozambique),6 (South Africa) and 7 (Namibia) forming asecond cluster, and those from transect 6(Lesotho) forming a third cluster (Fig. 2).

A discriminant analysis confirmedthe three morphoclusters with 93.8%

characters entered into the discriminant func-tions [17]. The intercolonial variances withintransects were tested for heterogeneity bymeans of Levene’s F-statistic. Analysis ofvariance (ANOVA) and Kruskal-Wallis pro-cedures followed by Scheffé’s multiple com-parison tests were used to test for significantdifferences in the means of the flight dimen-sional measurements between localities.

Chi-square tests using Greenacre’smethod were used to test for significant het-erogeneity in the frequency distributions ofthe restriction length fragments [12]. Thismethod tests for heterogeneity between clus-ters using the frequency distributions of thefragment types in a two-way contingencytable with r rows (transects) and c columns(patterns of restriction length fragments).The cut-off point for significant clustering isfound from the largest eigenvalue of aWishart matrix variate Wk(s) where the orderk = min{r-1, c-1} and the degrees of free-dom s= max{r-1, c-1} [12].

3. RESULTS

3.1. Morphometric analysis

Morphometric analyses of mountain hon-eybees present some special problems withrespect to sampling distance and size of thearea considered. For example, in a previousstudy of honeybees collected along a transectthrough Cameroon, a very distinct form wascollected at Bamenda in the Adamaouamountains which was described as ‘monti-cola-like’ [23]. Likewise, in previous mor-phometric analyses of honeybees from EastAfrica (including Burundi, Rwanda, Uganda,Kenya, Tanzania, Malawi, Zambia andnorthern Mozambique) the factor and dis-criminant analyses revealed yet other ‘mon-ticola-like’ morphoclusters, one consistingof large black mountain bees in Kenya andTanzania, the other of large yellowish beesat Chilinda on the Nyika plateau of Malawi[13]. However, no ‘monticola-like’ beesemerged in earlier studies of Ethiopia,

209

H.R

. Hepburn et al.

210Table I. Means and standard deviations (sd) of whole body massM, wing surface area S, wing loading factor W, thorax/whole body mass ratio r, excesspower index and morphometric variances (Var.) for honeybees sampled at differing altitudes in seven transects through mountainous areas of Africa#.Reference numbers to localities are shown in Figure 1.

Transect No., Map Altitude Sample M (mg) S(mm2) W r EPI Var. country ref. No. (m) sizeand locality

Mean sd Mean sd Mean sd Mean sd Mean sd

1 EthiopiaAdi ArkayGonderBahir Dar

2 CameroonMamfeBamendaKumboBanyoGouna

3 TanzaniaTangaArushaMt. Meru

4 MalawiChikwawaThyoloRumphiChitipaChilinda

333231

3435363738

262828

2122232524

9502 1212 400

1502 5002 1001050400

01 3903 000

1009001 0501 3002 600

565F2,13Pr

34224F4,10Pr

234F2,6Pr

66666F4,25Pr

15.68a17.16c17.46c3.90.0470.61*

17.24a19.67c19.07c16.31a16.34a3.60.0440.48

14.58a15.27a15.91a1.2ns0.53*

17.40a18.59a18.82a16.15a20.28c3.90.0140.37*

1.030.991.240.3ns

0.182.070.940.780.672.2ns

0.321.530.602.9ns

1.040.862.141.193.31b

2.1ns

47.54a50.44c47.59a5.50.0180.25

46.37a49.33c50.66c48.74a47.07a9.00.0020.85**

50.16a49.42a52.67a3.1ns0.63*

50.07a52.29a54.19c50.82a54.52c17.9< 0.0010.61**

2.131.770.790.7ns

1.030.911.910.110.543.4ns

2.311.801.610.3ns

0.931.311.150.951.310.7ns

0.33a

0.34a

0.37c

5.70.0170.55*

0.37a

0.39a

0.38a

0.34a

0.35a

2.5ns0.19

0.29a

0.31a

0.30a

0.6ns0.68*

0.35a

0.36a

0.35a

0.32a

0.37a

1.9ns0.19

0.010.010.03w

4.20.040

0.010.040.000.020.011.7ns

0.010.020.020.7ns

0.020.020.040.020.06b

1.8ns

0.54a

0.55a

0.54a

0.9ns0.22

0.53a

0.53a

0.56a

0.55a

0.55a

0.9ns0.26

0.52a

0.51a

0.55a

3.8ns0.57

0.51a

0.52a

0.50a

0.55c

0.49a

6.00.0020.19

0.020.010.011.6ns

0.020.030.010.000.002.0ns

0.030.020.012.4ns

0.030.010.010.020.04w

4.90.005

0.93a

0.94a

0.89a

3.4ns0.34

0.87a

0.86a

0.91a

0.96a

0.93a

1.7ns0.03

0.98a

0.92a

0.99a

3.6ns0.40

0.86a

0.88a

0.86a

0.98c

0.81d

4.40.0080.16

0.020.020.05w

4.10.041

0.030.09b

0.020.020.021.9ns

0.040.050.031.3ns

0.050.040.040.050.14wb

4.70.006

58.4b

33.021.32.1ns

28.226.618.447.723.54.2ns

39.510.6-1.3ns

51.546.549.654.883.8wb

5.4< 0.001

Mountain honeybees

211Table I. (continued).

5 Zimbabwe/MozambiqueBeiraMutareHarareKaroi

6 Lesotho/South AfricaPort AlfredGrahamstownQueenstownQuthingMokhotlongSemonkongThaba Tseka

7 NamibiaKeetmanshoopMarientalWindhoekOkahandja

All transectscombined

17181920

1259

121110

13141516

0338

1 4781 251

0525

1 0771 5782 1332 2002 286

1 7731 1801 7791 439

3545

F3,13Pr

5555555

F6,25Pr

4553

F3,13Pr

F30,105P

17.54a23.37c26.93d23.22c

5.2 0.0140.58*

19.88a19.89a21.58a21.59a19.87a20.91a18.71a

1.0ns

0.04

18.74a18.69a19.30a18.53a

0.3ns

0.15

7.93< 0.001

2.141.363.84b4.08b1.2ns

3.47b1.181.942.170.611.810.271.8ns

1.181.481.371.580.1ns

1.760.019

46.65a50.16c51.38d49.95c10.3

< 0.0010.62**

45.58a45.45a46.95a50.36c51.06c51.06c49.99c44.4

< 0.0010.90**

49.28a49.29a49.46a47.30a

2.4ns

0.13

17.43< 0.001

1.140.850.51

1.64w3.5

0.048

0.250.840.470.910.811.07

1.77w3.6

0.011

0.971.661.020.542.1ns

1.48ns

0.38a

0.47c

0.53d

0.46c

3.80.0380.54*

0.43a

0.44a

0.46a

0.43a

0.39a

0.41a

0.37a

2.0ns

0.40*

0.38a

0.38a

0.39a

0.39a

0.3ns

0.12

9.36< 0.001

0.040.030.08b0.07b1.1ns

0.08b

0.030.040.040.010.030.012.1ns

0.020.020.030.030.4ns

1.920.008

0.53a

0.50a

0.50a

0.51a

3.3ns

0.31

0.52a

0.51a

0.51a

0.54a

0.55a

0.55a

0.55a

2.3ns

0.51**

0.55a

0.54a

0.52c

0.50c

32.6< 0.001

0.07

4.29< 0.001

0.010.010.010.022.7ns

0.05wb

0.010.030.020.010.010.032.7

0.039

0.010.000.010.012.2ns

3.11< 0.001

0.87a

0.74c

0.70c

0.76c

5.00.0160.49

0.80a

0.78a

0.76a

0.83c

0.88c

0.86c

0.89c

2.70.0370.49**

0.90a

0.88a

0.84c

0.81c

5.40.0130.01

6.94< 0.001

0.050.040.060.080.6ns

0.12b

0.030.070.060.020.030.062.2ns

0.030.020.030.050.9ns

2.47< 0.001

17.426.125.742.83.2ns

13.269.9wb

34.423.454.654.826.06.4

0.026

28.647.130.832.20.7ns

3.67< 0.001

# Fdf: F-statistic, P: P-value, r : r correlation coefficient; ns: non significant.* (P < 0.05); ** (P < 0.01); b: between transects (P < 0.05); w: within transects (P < 0.05); acd: means with the same do not differ (P > 0.05).

H.R. Hepburn et al.

(1 misclassified) correct classification ofthe colonies from transect 1 (Ethiopia), witha posteriori probabilities P = 1.0 for14 colonies and P = 0.98 for the remaining1 colony; 97.8% (3 misclassified) correctclassification of the colonies from transects2 (Cameroon), 3 (Tanzania), 4 (Malawi),5 (Zimbabwe/Mozambique), 6 (SouthAfrica) and 7 (Namibia) with a posterioriprobabilities P = 1.00 for 97 colonies,0.90 ≤ P ≤ 0.99 for 25 colonies and0.62 ≤ P ≤ 0.89 for the remaining ninecolonies; 95.7% (1 misclassified) correctclassification of the colonies from transect 6(Lesotho only) with a posteriori probabilitiesP = 1.00 for 20 colonies and 0.72 ≤ P ≤ 0.83for the remaining two colonies. The jack-knife procedure gave the same classifica-tion results, except one more colony fromLesotho was classified incorrectly into thesecond morphocluster.

Considering high and low altitude influ-ences on the morphometric characters, nosignificant correlations were found betweenthe scores of factors 1 to 4 and altitude when

using the data from all three clusters. When,however, the analysis was restricted to thescores from clusters 1 and 3, a significantcorrelation was found between factor1 scores and altitude (r = 0.45, P < 0.0001;Fig. 3).

The variance characteristics of the mor-phometric data are given on a transect basisin Table I. Significantly high domains ofmorphometric variance occur in transects 1,4 and 6 at Adi Arkay (Ethiopia), Chilinda(Malawi) and Grahamstown (South Africa)respectively. Otherwise the bees of transects2, 3, 5 and 7 (Cameroon, Tanzania, Zim-babwe/Mozambique and Namibia respec-tively) are morphometrically on the homo-geneous side.

3.2. Flight dimensional analysis

In a factor analysis using the colonymeans of the mass-related characters andthe total wing surface area of worker hon-eybees from 136 colonies, two factors with

212

Figure 2. Factor analysis plot using morphometric characters: cluster 1 comprises colonies fromtransect 1 (Ethiopia), cluster 2 comprises colonies from transects 2, 3, 4, 5, 6 and 7 (Cameroon,Tanzania, Malawi, Zimbabwe/Mozambique, South Africa and Namibia respectively) and cluster 3 com-prises colonies from transect 6 (Lesotho).

Mountain honeybees

remaining 3 colonies; 93.7% (4 misclassi-fied) of the colonies from transects 1,2 and 6 (Ethiopia, Cameroon and Lesothorespectively) with a posteriori probabilitiesP = 1.00 for 11 colonies, 0.90 ≤ P ≤ 0.99for 38 colonies and 0.59 ≤ P ≤ 0.89 for theremaining 10 colonies; 79.4% (7 misclas-sified) of the colonies from transects 6,5 and 7 (South Africa, Zimbabwe/Mozam-bique and Namibia respectively) with a pos-teriori probabilities P = 1.00 for 11 colonies,0.90 ≤ P ≤ 0.99 for 6 colonies and 0.50 ≤P ≤ 0.89 for the remaining 10 colonies.

The jackknife procedure gave the sameclassification results, except one morecolony from cluster 2 was misclassified intocluster 3. A significant difference was foundbetween the means of the three clusters(∆ = 0.1175 with (4, 2, 133) df; F = 62.3with (8, 260) df, P < 0.0001). Consideringhigh and low altitude influences on the flightdimensional characters, a significant corre-lation was found between factor 2 scores(relating to total wing surface area) and alti-tude (r = 0.37, P < 0.0001; Fig. 5).

The means and standard deviations forthorax mass, whole body mass, wing surface

eigenvalues greater than 1 were isolated:factor 1, head, thorax, abdomen and wingmass; factor 2, total wing surface area. Thesetwo factors accounted for 80% of the vari-ance in the data. The loading for each char-acter had an absolute value greater than 0.7.The graph of the factor scores showed threeclusters: colonies from transects 3 and 4(Tanzania and Malawi) forming a cluster,colonies from transects 1, 2 and 6 (Ethiopia,Cameroon and Lesotho respectively) form-ing a second cluster and colonies from tran-sects 6, 5 and 7 (South Africa, Zimbabwe/Mozambique and Namibia respectively)forming a third cluster (Fig. 4). The bees ofthe first and third clusters consist of groupsfrom more or less neighbouring countries;however, the bees of the second cluster con-sist of populations in countries that are at a3 000 to 4 000 km distance from oneanother.

A discriminant analysis confirmed theseparation of the three clusters and correctlyclassified 94.9% (2 misclassified) of thecolonies from transects 3 and 4 (Tanzaniaand Malawi) with a posteriori probabilitiesP = 1.00 for 30 colonies, 0.90 ≤ P ≤ 0.99for 4 colonies and 0.76 ≤ P ≤ 0.89 for the

213

Figure 3. Relationship between the factor 1 scores of clusters 2 and 3 (Fig. 2) using morphometriccharacters and altitude.

H.R. Hepburn et al.

area, wing loading, thorax/whole body massratio and excess power index for workerhoneybees at each locality are given in Table I.ANOVA and nonparametric Kruskal-Wallis

procedures used to test for significant dif-ferences in means between localitiesrevealed significant differences for all theflight dimensional measurements (Tab. I).

214

Figure 4. Factor analysis plot using flight dimensional characters: cluster 1 comprises colonies fromtransects 3 and 4 (Tanzania and Malawi), cluster 2 comprises colonies from transects 1, 2 and 6(Ethiopia, Cameroon and Lesotho respectively) and cluster 3 comprises colonies from transects 5,6 and 7 (Zimbabwe/Mozambique, South Africa and Namibia respectively).

Figure 5. Relationship between the factor 2 scores using flight dimensional characters and altitude.

Mountain honeybees

morphoclusters obtained in the morphome-tric analysis.

Considering each mtDNA cluster sepa-rately, we see that within the mtDNA clus-ter of Ethiopia in transect 1, that Holeta,Gonder and Bahir Dar have a common DraIpattern. Different patterns are found at AdiArkay and Debre Markos. Similarly, withinthe mtDNA cluster of transects 4 and 5(Malawi and Zimbabwe/Mozambique) dif-ferent DraI patterns are found at Chilindaand Mutare respectively, and finally withinthe mtDNA cluster of transect 6 (Lesotho/South Africa) only one DraI pattern waspresent at all localities. The single colonyfrom Chilinda may be an incomplete digestof the more typical pattern, given the469 bp size increase of the amplified frag-ment over the standard size total found inTable II.

4. DISCUSSION

The morphoclusters obtained in the dis-criminant analysis of the honeybees of allseven transects considered jointly yieldedgroups which partially accord with otherrecently published assessments. That thehoneybees of transects 6 and 1 (Lesotho andEthiopia) are quite distinct from other loweraltitude bees surrounding them supportsrecent interpretations [13, 26, 27]. However,in this present analysis the largest cluster,comprised of bees from transects 4, 3, 5 and7 (Malawi, Tanzania, Zimbabwe/Mozam-bique and Namibia respectively) on the onehand and of transect 2 (Cameroon) on theother, is at odds with recent views [13]. Theformer have previously been classified asA. m. scutellataand the latter as A. m. adan-sonii.

Other discrepancies concern the high alti-tude mountain bees themselves, especiallythose regarded as A. m. monticola[1, 19,20, 30, 31]. The first point is that those highaltitude bees of Tanzania and Kenya wereundetectable as a distinct group in other

The correlation coefficients and corre-sponding levels of significance betweenflight dimensional properties and altitudeare given in Table I. In several transects theresults clearly show that there are signifi-cant positive correlations between certainflight properties and altitude which indicatethat larger honeybees are found at higheraltitudes.

The variance characteristics for the fiveflight dimensional properties of the honey-bees are indicated in Table I. Tests for theequality of the variances showed that thereare significantly higher variances at locali-ties at higher altitudes for certain proper-ties, i.e., significantly higher variations intransects 4 at Chilinda (Malawi), 1 at BahirDar (Ethiopia), 2 at Bamenda (Cameroon),5 at Karoi (Zimbabwe) and 6 at Thaba Tseka(Lesotho). Not surprisingly, transect 7(Namibia) is fairly homogeneous in thisregard, but the homogeneity in the case oftransect 3 (Tanzania) was unexpected.

3.3. mtDNA analysis

A total of eight different DraI restrictionfragment patterns were present in the sam-ples analysed from transects 1, 4, 5, and 6(Ethiopia, Malawi, Zimbabwe/Mozambiqueand Lesotho/South Africa respectively).Four of these patterns were restricted to tran-sect 1 (Ethiopia) alone (Tab. II). AnotherDraI pattern was common to the samplesanalysed from transects 4, 5 and 6 (Malawi,Zimbabwe/Mozambique and Lesotho/SouthAfrica respectively). The remaining threepatterns were distributed among the sam-ples from transects 4 and 5 (Malawi and Zim-babwe). Three significantly different mtDNAclusters were found using Greenacre’s chi-square method [12] (χ2 = 99.6, 4 df,P < 0.0001); 1) Ethiopia; 2) Malawi andZimbabwe (χ2 for difference = 0.180, ns);3) Lesotho and South Africa (χ2 for differ-ence = 0.001, ns). The mtDNA cluster ofEthiopia and the mtDNA cluster ofLesotho/South Africa matched the two

215

H.R. Hepburn et al.

geographically large-scale morphometricanalyses, and this was the result of samplingdistance resolution [25]. This is because thegreater the distance between samples, themore distinct the morphoclusters and, whenbetween-group variation is considerablylarger than within-group variation, small

biometric groups may be obscured entirely.On the other hand, the high mountain bees ofboth Lesotho and Ethiopia (at the oppositeend of the geographical spectrum consid-ered) remain intact as distinct populationsand are fundamentally different from theother high altitude mountain bees. So on

216

Table II. Frequency Distribution of eight different mitochondrial DNA restriction length patternsobtained from the non-coding region of COI-COII by a DraI restriction. Reference numbers tolocalities are shown in Figure 1.

Country and Map No. of Restriction patternslocalities ref. No. colonies

South AfricaQueenstown 5 4 550 209 100Port Alfred 1 2 550 209 100Dordrecht 6 3 550 209 100Hofmeyr 3 3 550 209 100Tarkastad 4 3 550 209 100Sterkstroom 7 2 550 209 100Burgersdorp 8 3 550 209 100

LesothoQuthing 19 6 550 209 100Mokhotlong 12 6 550 209 100Semonkong 11 6 550 209 100Thaba Tseka 10 2 550 209 100

ZimbabweHarare 19 5 550 209 100Mutare 18 4 550 209 100

1 550 1001 550 191 209

MalawiChitipa 25 5 550 209 100

1 550 100Rumphi 23 4 550 209 100

2 550 100 Chilinda 24 5 550 209 100

1 550 240 229 209 100

EthiopiaAdi Arkay 33 3 550 138 100

1 468 132 115Holeta 29 1 550 138 100Gonder 32 5 550 138 100Bahir Dar 31 1 550 138 100Debre Markos 30 1 550 138 100

2 468 132 1153 550 1382 550 216 100

Mountain honeybees

ANOVA for the whole data set per-formed independently of transect groupingsshow that there are significant differencesin all means of the flight-related variables.Only one locality (Chilinda, Malawi)exhibits high variance values for flightdimensions as well as morphometrically.Continuing the comparisons, high inter-locality variances for flight are independentof high variance domains for morphomet-ric characters for several localities (BahirDar, Ethiopia; Bamenda, Cameroon; PortAlfred, South Africa) which exhibit lowintercolonial morphometric variance.Although flight-related variables wereshown to lack subspecific discriminatorypower [15], they are highly effective indelineating altitudinal clusters [14]. Thereis no correspondence between those clus-ters derived from flight dimensions andthose defined by traditional morphometricmethods or by mtDNA cluster membershipin conjunction with the flight clusters.

Turning to the results of the mtDNA anal-ysis, three distinct clusters were formed,each of which closely corresponds with therespective morphoclusters for the samelocalities (Tab. III). However, when all threecluster sets (morphometric, flight andmtDNA) are jointly considered, it is quiteapparent that the mountain honeybees inves-tigated here consist of at least six differentpopulations (Tab. III). Put another way, thissimply means that there are at least sixentirely different kinds of mountain honey-bee populations within Africa. In theabsence of relevant mtDNA information, itis not possible to comment on the mountainbees of Kenya.

The only mountain bees for which thereis a one-to-one correspondence for the mor-phoclusters, flight clusters and mtDNA clus-ters are those of the Lesotho/South Africatransect. While these are the most homoge-neous mountain bees of the continent, thoseof Ethiopia are the most heterogeneous.Also, the latter exhibit the highest degreeof intracolonial and intercolonial high

morphometric grounds alone, the idea ofan archipelago of one subspecies or mor-phocluster of mountain bees designated as‘A. m. monticola’ is not supported.

These results do not put into question thefindings that the high altitude bees of Kenyaand Tanzania (the original A. m. monticola)can be morphometrically and allozymicallydistinguished from their lower altitudeneighbours in very fine space scale studies[18-20]. Rather, the general case is simplythat the magnitude of difference betweenthe high and low altitude bees of transects 1and 6 (Ethiopia and Lesotho) happen to besignificantly greater than is the case in Tan-zania or Kenya. Indeed Mounts Kenya,Elgon, Meru and Kilimanjaro shared palaeo-climatic conditions during the Quaternarythat were vastly different from all otherAfrican mountain regions [11, 16, 36]. Thus,while the monticola/scutellataseparation inTanzania and Kenya may well be relictual aspreviously suggested [18–20], the morpho-metric, flight dimensional and mtDNArestriction length pattern data (cf. below)unequivocally exclude the possibility of anarchipelago of related, high altitude beesthroughout the mountain systems of the con-tinent.

Specific details in which there is generalagreement in the data for high altitude beesis simply that they tend to be larger thanbees of lower altitudes. That is the only com-monality that holds for the honeybees of allseven mountain systems. Pigmentation isinteresting but also problematical, becausethe high altitude bees of Tanzania and Kenyaare more darkly pigmented than the beesimmediately surrounding them at lower alti-tudes, which is precisely the obverse of whatoccurs in all of the other transects. This alsoposes physiological questions as to the sig-nificance the pigmentation may hold forthermoregulation. In any event, the mountainbees of Lesotho and Ethiopia are far moredistinct morphometrically than are any ofthe high altitude mountain bees of the othertransect countries.

217

H.R. Hepburn et al.

variance for morphometric domains, andare also the most variable with respect tomtDNA restriction length patterns. It canbe noted in passing that when the bees oftransect 1 (Ethiopia) are considered with-out reference to other countries, the honey-bees can be differentiated into three quitedistinct populations [13, 24].

There is no commonality of high altituderestriction length fragments of mtDNA thatcould support the idea of an archipelago ofthe same subspecies of honeybees. It is aninteresting possibility to consider whetherthe mountain bees within coherent moun-tain systems are more related to each otherthan to bees in intrasystem comparisons.That this is actually the case is shown incomparisons of the total set of discriminantanalysis clusters for all the mountain sys-tems. The most parsimonious interpretationwould be that each of the high altitudegroups of mountain bees are nothing more orless than local adaptations of the predomi-nant lower altitude bees surrounding a par-ticular mountain. As such, all of these moun-tain bees would best be regarded as ecotypesof the prevailing subspecies in the area ofthe mountain under consideration.

The mtDNA analysis yielded significantheterogeneity among the three mtDNA clus-ters that were formed (Tab. II). When all ofthe data are combined the evidence is

unequivocal that there are at least six dif-ferent recognisable kinds of ‘mountain’ bees(Tab. III) that are more closely related tothe local, lower altitude bees surroundingthem than to each other on very distantmountain systems. We conclude that each ofthese kinds of mountain bees is an ecotypi-cally adapted subpopulation of the sub-species prevailing in the area where theyare found.

ACKNOWLEDGMENT

We thank P. Neumann for discussions regard-ing this manuscript.

Résumé – Les abeilles de montagned’Afrique. Il est un problème particulière-ment difficile dans la taxonomie et la bio-géographie des abeilles domestiques (ApismelliferaL.): définir des taxons infraspéci-fiques qui soient quantitativement précis etrecouvrent aussi une réalité biologique. Lesabeilles de montagne d’Afrique illustrenttrès bien cette difficulté. Plusieurs formesdifférentes « de montagne » ont été identi-fiées ces dernières années par l’analyse mul-tivariée des caractères morphométriques,étayée parfois par l’analyse des allozymes oude l’ADNmt. Nous avons prélevé desabeilles le long de sept transects différents à

218

Table III. Variations in cluster formation in discriminant analyses of morphometric, flight dimen-sional and mitochondrial DNA characters for the mountain honeybees of Africa. The flight dimen-sional clusters are given different numbers because they are not concordant with the morphometricclusters but form entirely new groups. Transect numbers shown in map of Figure 1.

Country Transect No. Morphoclusters Flight clusters mtDNA clusters

Ethiopia 1 1 4 1Malawi 4 2 5 2Tanzania 3 2 5 –Cameroon 2 2 4 –Namibia 7 2 6 –Zimbabwe 5 2 6 2Lesotho 6 3 4 3South Africa 6 2 6 3

Mountain honeybees

de montagne est une sous population éco-logiquement adaptée de la sous-espèce quiprédomine dans la région où ils ont été trou-vés.

Apis mellifera/ Afrique / biogéographie /morphométrie / caractéristique liée auvol / ADNmt / montagne

Zusammenfassung – Bergbienen vonAfrika. Es ist ein überaus schwieriges Pro-blem in der Taxonomie und Biogeographieder Honigbienen, quantitative exakte Defi-nitionen von innerartlichen Einheiten (Taxa)zu erstellen, denen gleichzeitig eine biolo-gische Bedeutung zukommen. Bei den Berg-honigbienen von Afrika zeigt sich dieseSchwierigkeit besonders deutlich. In denletzten Jahren wurden anhand multivaria-ter Analysen morphometrischer Eigen-schaften mehrere verschiedene “Berg” for-men identifiziert, die manchmal durchErgebnisse mit Allozymen oder mtDNAgestützt wurden. Wir haben Honigbienenentlang sieben verschiedener Schnittliniendurch Afrika analysiert, und zwar mit einerReihe von Techniken: Morphometrie, Datenüber Flugeigenschaften und der Analysevon mtDNA in der nicht kodierendenRegion von COI-COII. Die Bergsystemeund Orte, von denen die Bienen gesammeltwurden sind in Tabelle I und Abbildung 1dargestellt.Eine Faktorenanalyse ergab drei Mor-phocluster (Punktwolken von Proben), dieanschlieβend durch eine Diskriminanzana-lyse und dem “jackknife” Verfahren bestätigtwurden. Die Bienen von der Schnittlinie 1(Äthiopien) bilden eine Gruppe, die derSchnittlinien 4, 3, 2, 7, 6 und 5 ( die jeweilsMalawi, Tansania, Kamerun, Namibia, Süd-afrika und Simbabwe / Mosambik entspre-chen) die 2. Gruppe und eine 3. Gruppe liegtentlang der Schnittlinie 6 (Lesotho) (Abb. 2). Nach einer Faktorenanalyse der Flugeigen-schaften gefolgt von Diskriminanzanalyseund dem “jackknife” Verfahren ergaben sichwieder drei Cluster, die aber verschieden

travers l’Afrique et les avons analysées àl’aide d’un ensemble de techniques : mor-phométrie, mesures des caractéristiques liéesau vol (telles que surface des ailes, poidscorporel, etc.) et analyse de l’ADNmt dansla région non codante de COI-COII. Letableau I et la figure 1 indiquent les chaînesde montagne et les localités où les abeillesont été prélevées.L’analyse factorielle a fourni trois morpho-groupes qui ont été ensuite confirmés parl’analyse discriminante et les procédures de« jackknife ». Le premier groupe comprendles abeilles du transect 1 (Ethiopie), lesecond groupe celles des transects 4, 3, 2, 7,6 et 5 (Malawi, Tanzanie, Cameroun, Nami-bie, Afrique du Sud, et Zimbabwe/Mozam-bique respectivement) et le troisième groupeles abeilles du transect 6 (Lesotho) (Fig. 2).L’analyse factorielle des caractéristiquesliées au vol suivie d’une analyse discrimi-nante et d’une procédure de jackknife afourni trois groupes distincts : le premiercomprend les abeilles des transects 4 et 3(Malawi et Tanzanie), le second les abeillesdes transects 1, 2 et 6 (Ethiopie, Camerounet Lesotho, respectivement) et le troisièmeles abeilles des transects 6, 5 et 7 (Afrique duSud, Zimbabwe/Mozambique et Namibie,respectivement) (Fig. 4). Dans plusieurstransects il y avait des corrélations positivesentre certaines caractéristiques liées au volet l’altitude, qui montraient que les plusgrosses abeilles se trouvaient aux altitudesles plus élevées. Les abeilles de haute alti-tude présentaient aussi des variances signi-ficativement plus élevées que celles des alti-tudes plus basses.L’analyse de l’ADNmt a montré une diver-sité significative parmi les trois groupesd’ADNmt formés (Tab. II). La combinai-son de toutes les données prouvent sansambigüité qu’il existe au moins six typesreconnaissables d’abeilles « de montagne »(Tab. III). Chaque type se rapproche plusdes abeilles locales et de faible altitude quil’entourent que des abeilles des autreschaînes de montagne très éloignées. Nousconcluons que chacun de ces types d’abeilles

219

H.R. Hepburn et al.

zu den vorherigen sind: die Bienen von derSchnittlinie 4 und 3 (Malawi und Tansania)bilden eine Gruppe, eine 2. Gruppe umfas-sen die Bienen der Schnittlinien 1, 2 und 6(Äthiopien, Kamerun und Lesotho) und die3. Gruppe bilden die Bienen der Schnittli-nien 6, 5 und 7 (Südafrika, Simbabwe /Mosambik und Namibia; Abb. 4). In ver-schiedenen Schnittlinien ergaben sich posi-tive Korrelationen zwischen einigen Flug-eigenschaften und der Höhe, die dasVorkommen von gröβeren Bienen in grö-βeren Höhen anzeigen. Auβerdem weisenBergbienen auch signifikant höhere Vari-anzen auf als Bienen von geringeren Höhen.Die mtDNA Analysen ergaben eine signi-fikante Verschiedenartigkeit zwischen dendrei mtDNA Clustern, die in Tabelle II dar-gestellt sind. Nach der Kombination allerDaten läβt sich eindeutig feststellen, dass esmindestens sechs als verschieden anzuse-hende Arten von Bergbienen gibt (Tab. III),die näher mit den angrenzenden Flachland-bienen verwandt sind als mit den Bienender weit entfernten anderen Bergmassive.Wir schlieβen daraus, dass jede dieser Artenvon Bergbienen ökologisch angepasste Sub-populationen der Unterarten sind, die in demGebiet vorherrschen, in dem sie gefundenwurden.

Apis mellifera/ Afrika / Berg / Morpho-metrie / Flugeigenschaft / mtDNA

REFERENCES

[1] Bro A., L’abeilles noire d’Afrique, Rev. Fr. Apic.475 (1988) 294–296.

[2] Crewe R.M., Hepburn H.R., Moritz R.F.A., Mor-phometric analysis of 2 southern African races ofhoneybees, Apidologie 25 (1994) 61–70.

[3] DuPraw E.J., Non-Linnean taxonomy, Nature202 (1964) 849–852.

[4] DuPraw E.J., Non-Linnean taxonomy and thesystematics of honeybees, Syst. Zool. 14 (1965)1–24.

[5] Falconer D.S., Introduction to QuantitativeGenetics, 3rd ed., Longman, Harlow, 1989.

[6] Futuyma D.J., Evolutionary Biology, SinauerAssoc. Inc., Sunderland, MA, 1997.

[7] Garnery L., Vautrin D., Cornuet J.-M., SolignacM., Phylogenetic relationships in the genus Apisinferred from mitochondrial DNA sequence data,Apidologie 22 (1991) 87–92.

[8] Garnery L., Cornuet J.-M., Solignac M., Evolu-tionary history of the honey bee Apis melliferainferred from mitochondrial DNA analysis, Mol.Ecol. 1 (1992) 145–154.

[9] Garnery L., Solignac M., Celebrano G., CornuetJ.-M., A simple test using restricted PCR-ampli-fied mitochondrial DNA to study the geneticstructures of Apis melliferaL., Experientia 49(1993) 1016–1021.

[10] Garnery L., Mosshine F.H., Oldroyd B.P.,Cornuet J.-M., Mitochondial DNA variation inMoroccan and Spanish honey bee populations,Mol. Ecol. 4 (1995) 465–471.

[11] Goudie A.S., Climate: past and present, in:Adams W.M., Goudie A.S., Orme A.R. (Eds.),The Physical Geography of Africa, Oxford Univ.Press, Oxford, 1996, pp. 34–59.

[12] Greenacre M.J., Clustering the rows and columnsof a contingency table, J. Classif. 5 (1988) 39–51.

[13] Hepburn H.R., Radloff S.E., Honeybees ofAfrica, Springer-Verlag, Berlin, 1998.

[14] Hepburn H.R., Radloff S.E., Brown R.W.,Youthed C., Illgner P., Production of aerody-namic power in mountain honeybees (Apis mel-lifera), Naturwissenschaften 85 (1998) 389–390.

[15] Hepburn H.R., Radloff S.E., Steele G.R., BrownR.E., Dimensional aspects of flight in honeybeepopulations of Africa, J. Apic. Res. 37 (1998)147–154.

[16] Lovett J.C., Climatic history and forest distri-bution in eastern Africa, in: Lovett J.C., WasserS.K. (Eds.), Biogeography and Ecology of theRain Forests of Eastern Africa, Cambridge Univ.Press, Cambridge, 1993, pp. 23–29.

[17] Mardia K.V., Kent J.T., Bibby J.M., Multivari-ate Analysis, Academic Press, London, 1979.

[18] Meixner M.D., Population genetic analysis ofhybrid zones in East African Apis mellifera, Proc.XIII Int. Congr. IUSSI, Adelaide, Australia,1999, p. 30.

[19] Meixner M., Ruttner F., Koeniger G., The moun-tain bees of the Kilimanjaro region and theirrelation to neighbouring bee populations, Api-dologie 20 (1989) 165–174.

[20] Meixner M., Sheppard W.S., Dietz A., Krell R.,Morphological and allozyme variability in honeybees from Kenya, Apidologie 25 (1994)188–202.

[21] Ndiritu D.W., Mutugi N., Ndung’u S., Varia-tion in malate dehydrogenase allozymes amonghoneybee populations in Kenya, J. Apic. Res.25 (1986) 234–237.

[22] Radloff S.E., Multivariate Analysis of SelectedHoneybee Populations in Africa, PhD thesis,Rhodes Univ., Grahamstown, South Africa,1996.

220

Mountain honeybees

[31] Ruttner F., Biogeography and Taxonomy ofHoneybees, Springer-Verlag, Berlin, 1988.

[32] Ruttner F., Naturgeschichte der Honigbienen,Ehrenwirth, Munich, 1992.

[33] Sheppard W.S., Heuttel M.D., Biochemicalgenetic markers, intraspecific variation, and pop-ulation genetics of the honey bee Apis mellifera,in: Needham G.R., Page R.E., Delfinado-BakerM., Bowman C.E. (Eds.), Africanized HoneyBees and Bee Mites, Ellis Horwood, Chichester,1988, pp. 281–286.

[34] Sheppard W.S., Rinderer T.E., Meixner M.D.,Yoo H.R., Stelzer J.A., Schiff N.M., Kamel S.M.,Krell R., Hinfl variation in mitochondrial DNAof old world honey bee subspecies, J. Hered. 87(1996) 35–40.

[35] Smith F.G., The races of honeybees in Africa,Bee World 42 (1961) 255–260.

[36] Wasser S.K., Lovett J.C., Introduction of thebiogeography and the ecology of the rain forestsof eastern Africa, in: Lovett J.C., Wasser S.K.(Eds.), Biogeography and Ecology of the RainForests of Eastern Africa, Cambridge Univ.Press, Cambridge, 1993, pp. 3–7.

[37] Wright S., Evolution and Genetics of Popula-tions, Vol. 2, The Theory of Gene Frequencies,Univ. Chicago Press, Chicago, IL, 1969.

[38] Wright S., Evolution and Genetics of Popula-tions, Vol. 4, Variability Within and AmongNatural Populations, Univ. Chicago Press,Chicago, IL, 1978.

[23] Radloff S.E., Hepburn H.R., Multivariate anal-ysis of honeybee populations, Apis melliferaLin-naeus (Hymenoptera: Apidae), from westerncentral Africa: morphometrics and pheromones,Afr. Entomol. 5 (1997) 195–204.

[24] Radloff S.E., Hepburn H.R., Multivariate anal-ysis of honeybees, Apis melliferaLinnaeus(Hymenoptera: Apidae), of the Horn of Africa,Afr. Entomol. 5 (1997) 57–64.

[25] Radloff S.E., Hepburn H.R., The matter of sam-pling distance and confidence levels in the sub-specific classification of honeybees, Apis mel-lifera L., Apidologie 29 (1998) 491–501.

[26] Radloff S.E., Hepburn H.R., Population struc-ture and morphometric variance of the Apis mel-lifera scutellatagroup of honeybees in Africa,Genet. Mol. Biol. (in press).

[27] Radloff S.E., Hepburn H.R., Honeybees, ApismelliferaLinnaeus (Hymenoptera: Apidae), ofthe Drakensberg Mountains in relation to neigh-bouring populations, Afr. Entomol. 7 (1999)35–41.

[28] Radloff S.E., Hepburn H.R., Robertson M.P.,van Hille R., Davidson Z., Villet M.H., Dis-criminant analysis of the honeybee populationsof southwestern Africa, Afr. Entomol. 4 (1996)1–6.

[29] Radloff S.E., Hepburn H.R., Villet M.H., Thehoneybees, Apis melliferaLinnaeus (Hymoptera:Apidae), of woodland savanna of southeasternAfrica, Afr. Entomol. 5 (1997) 19–27.

[30] Ruttner F., African races of honeybees, in: Proc.13th Int. Apic. Congr., Apimondia, 1976,pp. 1–20.

221