Results
High-resolution population
network
The high-resolution population network clearly divided the honey bees
into two distinct population clusters (CAR and MEL), while five MEL
colonies (one SL_CH, and four CS_FR) were allocated to CAR (Figure 2A,
dashed circle). The hub between CAR and MEL particularly included CS_FR
and SL_CH colonies that were not clustering with their respective
patrilines. The topology of the network additionally revealed that
further substructures exist within MEL, while CAR colonies built a tight
population cluster, despite the different geographical sample origin.
The most evident substructures within MEL corresponded to CS_CH
colonies and two patrilines of the selection programme (P1 and P2). It
was interesting to see that four CS_FR colonies were directly connected
with two CS_CH colonies, while the remaining CS_FR colonies were
distributed over the network. Furthermore, CS_CH colonies were the
nearest neighbours of four SL_CH colonies originating from three
different patrilines (P1, P3 and P5), while P1 showed the strongest
genetic relationship. Compared to P1 and P2, the remaining three
patrilines (P3-P5) did not build a distinct population cluster, P3
colonies were distributed over the network without a discernible
pattern, while the majority of P4 and P5 colonies built a common
cluster. The association of the node size with FROHillustrates that with the exception of four colonies, all CAR showed
higher FROH than MEL. Furthermore, it can be noted that
CAR located in the neighbourhood of MEL (and vice versa), as well as
colonies not clustering with the respective patrilines show lower
FROH in general (Figure 2A).
Admixture
Based on the cross-validation error estimation, an optimal cluster
solution at K = 5 was determined increasing K from 2 to 10
(Supplementary Figure 2A). The first level (K = 2) of model-based
clustering clearly differentiated CAR from MEL with a
FST of 0.26 (Supplementary Figure 2B). The
cluster solution simultaneously highlighted that except for CS_CH and
P1, all MEL subpopulations contained admixed colonies, whereas CS_FR
showed the highest percentage of highly admixed colonies. At the second
(K = 3) and third level (K = 4) of clustering, P2 and CS_FR colonies
built a distinct cluster, respectively. At the optimal cluster solution
(K = 5), P4 and P5 colonies were allocated in a common population
cluster. At the additional two levels of clustering (K = 7 and K = 8),
some CAR colonies from the United States of America were allocated in a
distinct cluster and the aforementioned common population cluster was
further sub-structured, without separating P4 from P5 colonies.
Therefore, the hierarchical population clustering (increasing K from 2
to 5) confirmed the findings of the network-based population structure,
with the only exception to clearly differentiate P1 colonies from
CS_CH. This high agreement between the two applied population structure
methods also became visible by integrating the admixture levels at K = 5
in the high-resolution population network (Figure 2B), which
simultaneously revealed that colonies not clustering with their
respective patrilines and having low FROH were highly
admixed. Based on this observation, we removed five MEL and two CAR
outliers from downstream ROH analyses, while 38 MEL colonies with an
admixture level greater than 5% (K = 2) were summarized in a distinct
population cluster (ADMEL) and excluded from the identification of
homozygosity islands.
Runs of homozygosity
The ROH analysis recapitulates the results of the population structure
analyses (Table 2). The CAR sample had nearly twice as many ROH segments
(NROH = 23.28 ±4.95) than MEL (NROH =
13.60 ± 1.66) and ADMEL (NROH = 12.42 ± 2.62).
Concurrently, the total length of ROH was also twice as high in CAR
(SROH = 11.65 Mb ± 0.28) compared to the other two
population cluster (4.96 Mb ± 0.91 to 5.50 Mb ± 0.50), while the mean
segment length (LROH) remained similar between the three
population clusters (between 0.40 Mb and 0.50 Mb). As expected, the
lowest ROH values (NROH of 12.42 ± 2.62,
SROH of 4.96 Mb ± 0.91, and LROH of 0.40
Mb ± 0.05) were observed for ADMEL. Similarly, the ADMEL group had the
lowest mean FROH (2.20%), followed by MEL (2.50%) and
CAR (5.30%).
Summarizing the ROH results within MEL according to the a prioridefined subpopulations (patrilines and conservation areas) revealed that
CS_FR, including the highest proportion of admixed colonies (62%), was
the subpopulation with the lowest ROH values (NROH =
12.62 ± 2.50, SROH = 5.00 Mb ± 0.86,
LROH = 0.40 Mb ± 0.04, FROH = 2.30% ±
0.40) (Supplementary Table 1). The two subpopulations without any
admixed colonies, P1 and CS_CH, showed the highest minimal
FROH (2.30% and 2.10%, respectively), while the
highest maximal FROH (3.20%) was identified within P4.
Furthermore, P4 was associated with the highest maximal
NROH (18.00) and mean SROH (5.62 Mb ±
0.59), while P3 had the highest mean NROH (14.00 ± 1.73)
and shortest mean LROH (0.38 Mb ± 0.03).
The FPED of the 74 SL_CH colonies including all
patrilines except P4 ranged from 0.00 to 11.26%, with a mean
FPED of 3.13% ± 0.26, whereas FROHranged from 1.40 to 3.10%, with a mean FROH of 2.40% ±
0.28. The association between the two inbreeding coefficient measures
was low (R2 = 0.02). Including the individual
admixture levels, as a covariate, in the linear regression model
improved the concordance between FPED and
FROH (R2 = 0.10). Absolute differences
between FPED and FROH were higher in the
ADMEL (15 colonies > 5% CAR admixture, mean absolute
difference = 2.84%) compared to the remaining colonies (59 colonies
< 5% CAR admixture, mean absolute difference = 1.82%).
Homozygosity islands
We identified 24 CAR-specific (private) homozygosity islands (mean
length= 0.53 ± 0.69 Mb) distributed over nine chromosomes, while MEL had
17 homozygosity islands (mean length= 0.35 ± 0.19 Mb) over seven
chromosomes (Figure 3, Supplementary Table 2). Chromosomes 14 and 16 did
not bear any homozygosity islands. The largest homozygosity island was
identified for CAR on chromosome 11 at 3,737,355 bp to 7,286,231 bp,
which covers 3.55 Mb, or approximately 20% of the entire chromosome.
The shortest homozygosity island was private for MEL, spanning just 465
bp on chromosome 7. The majority of homozygosity islands were located at
the starting end of the chromosomes. Four chromosomes (6, 9, 12 and 15)
comprised homozygosity islands for both subspecies, but only one segment
on chromosome 12 overlapped between CAR and MEL, spanning from 3,124,678
bp to 3,335,849 bp, which did not contain any annotated genes.
Only 11 out of 24 homozygosity islands for CAR and 5 out of 17 for MEL
contained annotated genes (Table 3), and one CAR homozygosity island did
not even contain any uncharacterised loci (Supplementary Table 2). There
were substantially more uncharacterised loci than annotated genes within
the homozygosity islands (i.e. 788 uncharacterised loci and 24
characterised genes in CAR).
The 24 annotated genes embedded within CAR-specific homozygosity islands
clustered into six functional groups with high significance levels
(Bonferroni-adjusted p-value < 0.05, Supplementary Table 3):Neuroactive ligand-receptor interaction , Insect cuticle
protein , Structural constituent of cuticle , Ion transport,
cell junction and Synapse . Half of the characterised genes (12
out of 24) could not be regrouped by function, namely Ndufs1,
PHRF1, Rep, Snf, Chmp1, Crh-BP, Gpdh, Grp, Rga, RpL35, Tpx-4 and
Uqcr11 . The five genes embedded in MEL-specific homozygosity islands
(GstS1, Pban, WRNexo, Uvop and Mad ) did not share
functional terms.