FEMS Microbiology Ecology, 2022, 98, 1–11

DOI: 10.1093/femsec/fiac090
Advance access publication date: 25 July 2022
Research Article

Conserved developmental trajectories of the cecal

microbiota of broiler chickens in a field study
Jannigje G. Kers1,2,3,* , Francisca C. Velkers1 , Egil A.J. Fischer1 , J. Arjan Stegeman1 , Hauke Smidt2 , Gerben D.A. Hermes2

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

1
Department Population Health Sciences, Faculty of Veterinary Medicine, Division Farm Animal Health, Utrecht University, Yalelaan 7, 3584 CL Utrecht, The
Netherlands
2
Laboratory of Microbiology, Wageningen University & Research, Stippeneng 4, 6708WE Wageningen, The Netherlands
3
Department Population Health Sciences, Institute for Risk Assessment Sciences (IRAS), Utrecht University, Yalelaan 2, 3584 CM Utrecht, The Netherlands
∗
Corresponding author: Department Population Health Sciences, Institute for Risk Assessment Sciences (IRAS), Utrecht University, Yalelaan 2, 3584 CM Utrecht,
The Netherlands. E-mail: [email protected]
One sentence summary: Conserved cecal microbiota development trajectories across different commercial broiler flocks despite the known strong influence of
environmental factors.
Editor: Leluo Guan

Abstract
There is great interest in identifying gut microbiota development patterns and underlying assembly rules that can inform strategies
to improve broiler health and performance. Microbiota stratification using community types helps to simplify complex and dynamic
ecosystem principles of the intestinal microbiota. This study aimed to identify community types to increase insight in intestinal
microbiota variation between broilers and to identify factors that explain this variation. A total of 10 well-performing poultry flocks
on four farms were followed. From each flock, the cecal content of nine broilers was collected at 7, 14, and 35 days posthatch. A total
of two robust community types were observed using different clustering methods, one of which was dominated by 7-day-old broilers,
and one by 35-day-old broilers. Broilers, 14-day-old, were divided across both community types. This is the first study that showed
conserved cecal microbiota development trajectories in commercial broiler flocks. In addition to the temporal development with age,
the cecal microbiota variation between broilers was explained by the flock, body weight, and the different feed components. Our
data support a conserved development of cecal microbiota, despite strong influence of environmental factors. Further investigation
of mechanisms underlying microbiota development and function is required to facilitate intestinal health promoting management,
diagnostics, and nutritional interventions.

Keywords: 16S rRNA, gut, intestinal, microbial community, microbiome, poultry

Introduction the intestinal microbiota is unique per individual, conserved com-

The intestinal microbiota is associated with the health and positional patterns, termed enterotypes, were discovered across
production performance of broiler chickens (Stanley et al. 2014a, human adults, independent of age, gender, cultural background,
2016, Han et al. 2016, Johnson et al. 2018). Therefore, there is and geography (Arumugam et al. 2011). This enterotype concept
great interest in identifying the biological principles that underlie was further refined, acknowledging that statistical support for ro-
the structure and function of these microbiological ecosystems. bust clusters was variable and a range of confounding factors
This knowledge can contribute to the development of beneficial could affect the initially defined discrete clusters (Costea et al.
nutritional management as well as diagnostic tools, to improve 2018). Nevertheless, stratification using cluster analysis can still
broiler health. However, several studies in broilers have described serve as a powerful tool to reduce the complexity of the micro-
the intestinal microbiota as highly variable within and between biota community landscape (Costea et al. 2018). In human infants
repeated experiments (Stanley et al. 2013, Thibodeau et al. 2015, and adults, community types have been associated with differ-
Cuperus et al. 2018). Feed, antimicrobial products, host, and envi- ences in microbial functionality, diseases, and with differences in
ronmental factors have been shown to attribute to the variation diet (Arumugam et al. 2011, Costea et al. 2018, Borewicz et al.
in intestinal microbiota (Apajalahti et al. 2001, Borda-Molina et 2019, Zhong et al. 2019). There have also been attempts to de-
al. 2018, Kers et al. 2019). The effect of factors can overlap as we fine microbiota community types in poultry. In 31 broilers aged 56
showed that the effect of a feed intervention on the composition days, the fecal microbiota was classified into four potential com-
of broiler microbiota was highly dependent on the environment munity types based on principal component analysis (Kaakoush
(Kers et al. 2019). In humans, it has also been shown that similar et al. 2014). These community types or clusters were defined as
foods can have different effects on the microbiota (Johnson et al. dominated by Firmicutes alone, or in combination with Proteobac-
2019). Therefore, it is important to identify which factors influence teria, Actinobacteria, or Bacteroidetes, respectively (Kaakoush et al.
the microbiota composition of broilers, and to which extent. 2014). Another study observed three community types in duo-
It has been proposed that microbial communities coevolve with denal content of broilers aged 77 days. One cluster was domi-
their hosts (Dethlefsen et al. 2008, De Filippo et al. 2010, Uhr et al. nated by Bacteroides and Escherichia–Shigella, one by Ochrobactrum
2019). Although it is generally accepted that the composition of and Rhodococcus, and the third by Bacillus and Akkermansia (Yuan

Received: February 9, 2022. Revised: June 1, 2022. Accepted: July 21, 2022

C The Author(s) 2022. Published by Oxford University Press on behalf of FEMS. This is an Open Access article distributed under the terms of the Creative

Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any
medium, provided the original work is properly cited.
2 | FEMS Microbiology Ecology, 2022, Vol. 98, No. 9

(A) (B)

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

(C) (D)

Figure 1. Quality and confidence scores for PAM and DMM clustering of cecal broiler microbiota. The dataset contained 270 broilers of age 7, 14, and 35
days old. All four figures show most support for two clusters. The thresholds for significance of clustering scores are indicated as dashed lines on the
plots. (A) Based on the prediction strength, strong support was observed for the PAM weighted UniFrac (PAM–WUF), and moderate support when using
other distance metrics (BC, JS, and UF). (B) No support for two clusters was observed based on the silhouette index, although the highest score was also
observed for two clusters. (C) All distance metrics showed the highest score for two clusters based on the Calinski–Harabasz score. The PAM–WUF
methods showed again the highest score for two clusters. (D) The DMM cluster score showed also highest evidence for two clusters.

et al. 2020). In addition to community types to observe certain de- provide useful insights in the intestinal microbiota development,
velopmental patterns, maturation patterns have been described as well as into factors that can affect this development and their
before. An experimental study showed the maturation of the fecal respective importance.
microbiota until day 30 (Gao et al., 2017), however, if this process To this end, this study aimed to explore whether stratification
contains different phases and is influenced by factors other than of the cecal microbiota into community types could provide in-
prebiotics or antibiotics is unknown. sight into the developmental patterns of cecal microbiota. In ad-
Although the cecal microbiota has been widely investigated be- dition, we aimed to identify variables impacting this development
cause of its functionality, which in broilers is especially related within and between broiler flocks. A longitudinal study in four
to the fermentation of feed (Stanley et al. 2014b, Svihus 2014), well-performing broiler farms with a total of 10 flocks was per-
the factors that contribute to the normal compositional variation formed. From each flock, nine individual broilers were sampled
in healthy broiler populations have remained under-investigated. at an age of 7, 14, and 35 days posthatch. In total, the cecal mi-
Therefore, the main drivers of cecal microbiota variation remain crobiota of 270 broilers was determined by 16S ribosomal RNA
unknown. In humans, factors such as stool consistency and med- (rRNA) gene amplicon sequencing. The outcomes of community
ication have been shown to be important determinants of adult type analyses have been shown to be highly dependent on the ap-
fecal microbiota variation (Falony et al. 2016, Müller et al. 2020). plied clustering algorithms (Koren et al. 2013, Costea et al. 2018).
However, broilers have a short life span, which increases the diffi- To robustly define community types within the cecal microbiota,
culty to identify the factors that influence their microbiota and two clustering methods were used; partitioning around medoid
disentangle these from temporal variation, because the micro- (PAM) with four beta diversity metrics, and Dirichlet Multinomial
biota is most likely still developing at the end of their lives be- Mixtures (DMM). In addition to the community types, host char-
cause broilers do not reach adulthood. In human the principles acteristics, environmental factors, and feed components were in-
of community types have been studied before. Infants aged 3 to cluded in multivariate distance-based redundancy analysis (db-
46 months, ten community types were observed and described in RDA), to study the relative impact of these factors on the variation
a transition model consisting of three phases: a developmental, in microbiota composition between broilers.
transitional and stable phase (Stewart et al. 2018). Another study
showed that infants belonging to different microbial clusters also
Results
had different degradation patterns of human milk oligosaccha-
rides (Borewicz et al. 2019). Although there is a large difference Microbiota stratification into community types
between the birth of a child and hatching of commercial chick- Identifying community types in compositional datasets not only
ens in terms of exposure, the concept of microbial clusters can depends on the data itself, but can also be sensitive to the ap-
 Kers et al. | 3

plied methods, therefore, different methods are applied; PAM- Table 1. The output of the different models based on UF db-RDA.
based methods using amplicon sequence variants (ASVs) Jensen– The results are the final models. VIF scores are given for variables
Shannon divergence (PAM–JSD), Bray–Curtis dissimilarity (PAM– excluded from the full model.
BC), unweighted UniFrac (PAM–UF), or weighted UniFrac (PAM–
Model 1 Df AIC F Pr (> F)
WUF), and DMM clustering. Nonetheless, all five applied cluster-
ing methods were in concordance, indicating an optimum of two Flock 9 903.10 4.51 0.005
clusters within the 270 broiler chickens aged 7, 14, and 35 days Community 1 893.32 13.40 0.005
(Fig. 1). Across all five methods, the two clusters were associated types
with age. Cluster 1 contained almost all 7-day-old broilers, and Body weight 1 889.68 9.76 0.005
cluster 2 contained all 35-day-old broilers, whereas the 14-day-old Surface 1 888.40 1.20 0.215

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

broilers were distributed across both clusters. However, depending Flock size 1 882.70 0.91 0.560
Sex 1 883.65 0.96 0.580
on the applied method individual broilers were classified in dif-
Feed producer 2 881.65
ferent clusters (Fig. 2). For example, on day 14 of Farm 1, PAM–UF
Litter type 2 881.65
clustering resulted in 11/18 broilers in cluster 1, PAM–WUF clus- VIF scores: age parent stock (1191), age (164), density (22), farm (11),
tering resulted in 6/18 broilers in cluster 1, and DMM clustering hatchery (11), and antibiotic use (6).
resulted in 9/18 broilers in cluster 1 (Fig. 2).
Model 2 Df AIC F Pr (> F)
The predominant families were Lachnospiraceae in cluster 1 and
Community 1 898.65 13.26 0.005
Ruminococcaceae in cluster 2. The predominant genera were Ru-
types
minococcus torques group in cluster 1 and Faecalibacterium in clus- Body weight 1 894.62 9.16 0.005
ter 2 (Figure S1, Supporting Information). The top 10 ASVs, that Rapeseed meal % 1 894.18 8.71 0.005
significantly differed in relative abundance between the two clus- Sunflower seed 1 893.14 7.67 0.005
ters, were independent of the applied clustering method (Table S1, meal %
Supporting Information). Fish oil % 1 890.60 5.12 0.005
Commercial broilers have a short life span between hatch and VIF scores: soybean meal % (1580), farmers wheat % (309),
slaughter of approximately 5–8 weeks, and during this period their metabolizable energy(AME)·kg−1 (153), fecal digestible lysine g·kg−1
(13 449), maize % (195), oats % (2998), phosphorous (22), wheat % (42),
intestinal microbiota composition changes rapidly. Therefore, the
potato protein % (9), methionine + cysteine (5), and flock (4)
cluster analyses were also stratified by age, to identify poten-
tial clusters within age group. These analyses showed that differ- Model 3 Df AIC F Pr (> F)
ent clustering algorithms resulted in differences in optimal clus- Flock 9 887.41 4.13 0.005
Body weight 1 877.32 10.26 0.005
ter structures, which suggests there were no robust age-specified
Community 1 876.49 9.44 0.005
clusters within 7-, 14-, or 35-day-old broilers with the number of
types
samples included in this study (Fig. 3).
Rapeseed meal % 1 873.11 6.16 0.005
Sunflower seed 1 873.07 6.11 0.005
meal %
Diversity and developmental trajectories of cecal Fish oil % 1 870.75 3.89 0.005
microbiota VIF scores: soybean meal % (1580), farmers wheat % (309),
To assess whether the two identified community types differed metabolizable energy (AME)·kg−1 (153), fecal digestible lysine g·kg−1
in composition and alpha diversity, DMM-, PAM–UF- and PAM– (13 449), maize % (195), oats % (2998), phosphorous (22), wheat % (42),
WUF-derived clusters were compared. Within sample (alpha) di- potato protein % (9), and methionine + cysteine (5)

versity defined as phylogenetic diversity was higher in cluster 2

compared to cluster 1, independent of clustering method (Fig. 4). variables on microbiota composition during the development of
Alternative alpha diversity metrics (ASV richness and Shannon cecal microbiota. Model 1 is was the most parsimonious UF-db-
diversity) confirmed these results (Table S2, Supporting Informa- RDA model (after testing and adjusting for collinearity), and con-
tion). As the 14-day-old broilers were distributed across both clus- sisted of three significant explanatory variables: flock, the DMM-
ters we also tested whether, within this age category, a difference based community types, and body weight (Table 1). This model
between clusters could be identified. The phylogenetic diversity explained 31.7% of the cecal microbiota variation, which was very
of 14-day-old broilers in DMM, PAM–UF-, or PAM–WUF cluster 1 similar to the analysis based on WUF (30.6%) with the same vari-
was indeed lower compared to that of 14-day-old broilers in DMM ables (Fig. 5A and D). Only minor differences were observed with
cluster 2 (Fig. 4). the PAM based clusters (33.3% and 31.0%). It should be noted that
We used PERMANOVA, based on UF and WUF distances, to as- the two community types were related to age, and that age and
sess the percentage of total microbiota variation the community body weight are highly correlated in these fast growing chickens.
types accounted for. This was 16.4% and 18.1% for DMM cluster- Variation partitioning visualized with Venn diagrams, shows this
ing, 16.5% and 17.6% for UF–PAM clustering, and 14.1% and 22.1% strong collinearity between cluster, body weight, and age (Figure
for WUF–PAM clustering. In addition to the two community types, S3, Supporting Information; 11% and 9%). Because of the collinear-
13 host and environmental characteristics that might affect com- ity between these variables, we cannot unequivocally conclude
position were tested for their effect on microbiota composition us- that the relative abundance of members of the genus Faecalibac-
ing UF and WUF db-RDA (Table S3, Supporting Information). These terium was only associated with higher body weight (Fig. 5A and
microbiota covariates included flock size, surface (poultry house D), and whether the relative abundance of a member of the genus
in m2 ), bird density per m2 , litter type, age of the parent stock, Ruminococcus torques group was strongly associated with the ordi-
hatchery, feed producer, antibiotics use, farm, and flock, as well nation in broilers within cluster, as both were also associated with
as characteristics at the individual animal level, including body the age of the broiler (Fig. 5A and D).
weight, age, and sex (Table S3, Supporting Information). The db- To assess the impact of differences in feed components on the
RDA analysis allowed us to determine the relative effect of these cecal microbiota, 13 feed components were added as explana-
4 | FEMS Microbiology Ecology, 2022, Vol. 98, No. 9

(A)

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

(B)

(C)

Figure 2. Distribution of clusters across farms and animal age. (A) Clusters were assigned using PAM–UF. (B) Clusters were assigned using PAM–WUF
clustering. (C) Clusters were assigned using DMM clustering. The variation of individual samples through different cluster methods stratified per farm.
Farm 1 is light blue, Farm 2 is dark green and light green (a and b are different production cycles), Farm 3 is red, and Farm 4 is yellow.

tory variables (Table S3, Supporting Information). In model 2, only at all farms was relatively similar on day 7, except for Farm 3
body weight, sunflower seed meal %, rapeseed meal %, and fish oil (Fig. 6A), which remained distinct, due to the presence of certain
% were related to microbiota composition and together with the Lactobacillus ASV (Fig. 6A). The WUF based PRC, which also con-
DMM community types explained 26.4% (UF-db-RDA) and 27.9% siders the abundance of ASV, showed that Farms 3 and 4 started
(WUF-db-RDA) of cecal microbiota variation (Table 1, Fig. 5B and with a deviant composition from Farms 1 and 2. Through time the
E). Model 2 did not contain flock anymore, however, when flock composition of Farm 3 converged, but the composition of Farm
was included in the model (model 3), it did increase the explana- 4 did not (Fig. 6B). Taken together, we observed that feed com-
tory power to 37.0% and 36.7% (UF and WUF; Fig. 5C and F). The ponents, flock, and body weight had an effect on the cecal mi-
addition of flock resulted in a variance inflation factor (VIF) value crobiota composition. Importantly, we observed a convergence of
of 4.2, suggesting that the variable flock contains limited or no cecal broiler microbiota through time, showing a conserved de-
unique information. In this dataset, it is difficult to disentangle the velopmental pattern exemplified by the identified clusters. How-
contribution of the different feed components from flock and farm ever, especially feed associated variables were associated with the
effects, because dietary components were strongly correlated farm, therefore, the independent effects of these variables on mi-
with farms. For instance, Fig. 5(C) and (D), show that e.g. fish oil % crobiota composition could not be disentangled further.
was associated with Farm 1 and rapeseed meal % with Farm 2.
Finally, to determine the development of cecal microbiota com-
position of poultry flocks in different farms over time was visual-
Discussion
ized using Principle Response Curve (PRC; Fig. 6). When only con- This study showed that cecal microbiota of broilers from com-
sidering community membership using UF, the cecal microbiota mercial broiler farms could be stratified into community types
 Kers et al. | 5

(A) (B) (C) (D)

(E) (F) (G) (H)

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

(I) (J) (K) (L)

Figure 3. Quality and confidence scores of cluster analyses stratified by sampling day using different metrics. (A) In the cecal content of 7-day-old
broilers, the PAM–WUF method showed near-moderate support (0.78, threshold at 0.80) for two clusters, however, the other methods showed little or
no support for significance based on the prediction strength. Although there was no support for clusters, the prediction strength based on the PAM–UF
method was higher for four clusters instead of two clusters. (B) and (C) Based on the silhouette index and the Calinski–Harabasz score, the PAM–UF
distance metrics resulted in three clusters, and all other distance metrics in two clusters. (D) Based on DMM, no clusters were observed. This indicated
that depending on the clustering method, the number of clusters varied in 7-day-old broilers. (E) On day 14, little to no support was observed for any of
the distance metrics based on the prediction strength, however, the highest support was for two clusters independent of distance metrics. (F) No
support for clusters was observed based on the silhouette index as well. (G) PAM–BC, –JS, and –WUF showed the highest Calinski–Harabasz score for
three clusters, and PAM–UF the highest Calinski–Harabasz score for two clusters. (H) The DMM method resulted in two clusters. (I) and (J) In the
broilers of 35 days old, also little to no support was observed for any of the distance metrics, although again the highest support was for two clusters
(prediction strength, I). (K) The Calinski–Harabasz score showed that distance metrics BC and JS resulted in two clusters, and distance metrics UF and
WUF resulted in three clusters. (L) The DMM method resulted in two clusters.

and that cecal microbiota development patterns are conserved In general, identifying community types is sensitive to the ap-
between farms and flock across age, despite the use of different plied methodologies (Koren et al. 2013). The highest prediction
farms and flocks. We identified, two robust community types; one strength, silhouette index, and Calinski–Harabasz score were ob-
dominated by broilers of 7 days old, and one dominated by broil- served with the PAM–WUF method. This is in line with previous
ers of 35 days old. Broilers, 14-day-old, were divided across both suggestions that WUF distance metric might be the best choice
community types. This indicates that the development of the ce- for cluster or enterotyping structures (Koren et al. 2013). Our re-
cal microbiota followed a general trajectory across farms, and that sults showed high to moderate support for two community types
a transition occurred around the second week of life. within the total dataset based on the prediction strength, but the
The development of the intestinal microbiota toward a ma- silhouette index did not support this observation. This trend of
tured intestinal microbiota in chickens is not fully understood (Yin high to moderate support based on the prediction strength and a
et al. 2010, Ballou et al. 2016, Donaldson et al. 2017, Jurburg et al. limited support based on the silhouette index has been observed
2019, Kers et al. 2019). Early life colonization, feed additives, and before. It has been noticed that prediction strength is most sen-
antimicrobial drugs play a large role in microbiota composition sitive to observe community types (Koren et al. 2013, Yuan et al.
and affect the development (Ballou et al. 2016, Gao et al. 2017). 2020). In previous studies, four fecal community types and three
Gao et al., 2017 showed maturation of the fecal microbiota until duodenum community types were identified in broiler chickens of
day 30. Our results indicate that the temporal cecal microbiota 57 days (n = 31) and 77 days old (n = 206) (Kaakoush et al. 2014,
development of broiler chickens undergoes two distinct phases, Yuan et al. 2020). In contrast, we observed no robust community
independent of factors as farm or flock, while in human infants, types in age-stratified analyses of 7-, 14-, or 35-day-old broilers.
10 clusters, and three phases have been observed (Stewart et al. This might be because in our study cecal content was used, and
2018). The observation that the broilers of 14 days old were divided the broilers were much younger. In humans, the establishment of
across both clusters suggests that the period around the second intestinal microbiota cluster structures has been estimated to oc-
week of life is an important transitional phase. In contrast to other cur between the age of 9 and 36 months (Bergström et al. 2014,
studies that suggest stabilization of the community richness (al- Zhong et al. 2019). Furthermore, the limited number of individ-
pha diversity) by day 14 in cecal or fecal content (Lu et al. 2003, ual samples in the stratified data analysis (n = 90) might also be
Jurburg et al. 2019), our data showed that community richness was an explanation why we did not observe robust community types
still increasing around this age. This may suggest that in the other within age groups.
experimental studies this transition phase was earlier, compared Previous research in broilers showed that continuous supply
to our field study. However, considering the large time interval be- of in-feed antibiotics decreased the maturation of the intestinal
tween days 14 and 35 the exact moment of stabilization could not microbiota, while feed with the probiotic bacterium Lactobacillus
be determined in this study and may have occurred shortly after plantarum accelerated the development of intestinal microbiota
day 14. (Gao et al. 2017). On days 3 until 6 of the production cycle of Farm
6 | FEMS Microbiology Ecology, 2022, Vol. 98, No. 9

(A) viance from other farms was more related to lower chick health
on Farm 3, and not to antibiotic treatment. We cannot rule out
that the effects on microbiota were mostly caused by a lower chick
quality and disease in both houses than by the use of antibiotics.
In both houses on the farm, the day old chicks were obtained from
the same hatchery and parent flock and both had relatively high
early life mortality due to omphalitis. In house 2, slightly higher
mortality was observed compared to house 2, which prompted the
antibiotic use in this house alone.
In addition to the observed community types, body weight,

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

and the percentage of the feed components sunflower seed meal,
rapeseed meal, and fish oil explained 26%–27% of the cecal mi-
crobiota variation between broilers. However, when flock was
(B) included in the analysis, the explained variation increased by
around 10% but so did the collinearity between variables. This in-
dicates that it is difficult to disentangle the contribution of the
different feed components from the effect of flock and farm, un-
less a larger number of farms, with similar feeds would be in-
cluded in the analysis. Another study identified four fecal com-
munity types in two farms and suggested the grouping did not oc-
cur by chance because considerable microbiota variation between
farms was observed (Kaakoush et al. 2014). In our study, the two
community types were observed across different poultry flocks.
All flocks were located in the Netherlands, where similar feed
compounds are available from different feed suppliers, and all
farms used wheat-based feeds. Therefore, the variation between
feed and flock is relatively small, which may limit the generaliz-
(C) ability of the relevance of the identified feed compounds to other
countries.
In humans, diet can provoke a shift in intestinal microbiota
clusters (Wu et al. 2011, Kovatcheva-Datchary et al. 2015). Al-
though the feed in different farms was obtained from different
feed suppliers, this was not reflected in our developmental trajec-
tory. During the production cycle, feed shifts occurred on days 9
(Farms 1 and 2), 10 (Farms 2 and 4), and 12 (Farm 3), however, with
a retention time of the digestive tract of less than 12 h (McWhorter
et al. 2009, Sundu 2009), we assume that the microbiota was al-
ready adjusted to the feed shift before day 14. Also, if feed or farm
would be the main drivers of the temporal development, then
the optimal number of community types (clusters) would have
been three, associated with broiler age, or four associated with
Figure 4. Cecal microbial alpha diversity of the two clusters stratified by the farm.
age. Phylogenetic diversity (ASV level) of cecal microbiota (n = 270) of Only a few studies have focused on providing insight into the
the two clusters using different clustering methods. (A) DMM cluster factors that influence the temporal development and microbiota
(Kruskal–Wallis χ2 = 161.49, P-value < 2.2e-16). (B) PAM–UF cluster,(χ2 = configurations of broiler chicken intestinal microbiota (Johnson et
151.73, P < 2.2e-16). (C) PAM–WUF cluster (χ2 = 117.35, P < 2.2e-16).
al. 2018). This research remains challenging, because of the wide
Within broilers of 14 days old (DMM, χ2 = 36.60, P-value < 1.5e-09;
PAM–UF, χ2 = 25.04, P-value < 5.6e-07; and PAM–WUF, χ2 = 19.84 array of available approaches, each with their advantages and
P-value < 8.4e-06). Whiskers show 95% interval, box 50% interval. disadvantages (Costea et al. 2018). No single method is perfect,
and although the db-RDA resulted in the same important factors,
3, the flock in poultry house 2 received a 3-day antibiotic treat- a less stringent VIF threshold influences the results and increases
ment (trimethoprim + sulfamethoxazol). This early life antibiotic explained variation (Zuur et al. 2010).
treatment did not result in different community types. Although In summary, stratification of microbiota composition in clus-
the PRC analysis showed that Farm 3 (poultry houses 1 and 2), ters showed two robust community types in the cecal microbiota
was distinct from the other farms on day 7, this was due to the of broiler chickens and with the PRC results, this indicates a con-
higher relative abundance of a member of the genus Lactobacillus served developmental trajectory of the cecal microbiota across 10
(ASV). This higher relative abundance of Lactobacillus was also ob- different commercial broiler flocks on four different farms. This
served on days 14 and 35 in house 2 as well in house 1, where no emphasizes the importance for further investigation of mecha-
treatment with antibiotics was applied. This suggests that the dif- nisms underlying microbiota development and functions that af-
ference between Farm 3 and the other farms could not be solely fect broiler health and performance. This mechanistic knowledge
attributed to the early life treatment with antibiotics, or that the can contribute to the development of new nutritional interven-
houses were not distinct due to transmission of microbes between tions, improved management, as well as better diagnostic tools to
poultry houses. Alternatively, it can also be argued that this de- improve broiler health.
 Kers et al. | 7

(A) (B) (C)

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

(D) (E) (F)

Figure 5. Multivariate effects of microbiota covariates on the cecal microbiota of broilers. Distance based redundancy analysis (db-RDA) triplots of the
cecal microbiota of broilers with weighted and UF. Samples are colored by DMM-based cluster. (A) UF-db-RDA with explanatory variables cluster, flock,
and body weight. (B) UF-db-RDA with cluster, body weight, sunflower seed meal %, rapeseed meal %, and fish oil %. (C) UF-db-RDA with cluster, flock,
body weight, sunflower seed meal %, rapeseed meal %, and fish oil %. (D) WUF-db-RDA with cluster, flock, and body weight. (E) WUF-db-RDA with
cluster, body weight, sunflower seed meal %, rapeseed meal %, and fish oil %. (F) WUF-db-RDA with cluster, flock, body weight, sunflower seed meal %,
rapeseed meal %, and fish oil %.

Table 2. Farm characteristics.

Farm 1 Farm 2 cycle 1 Farm 2 cycle 2 Farm 3 Farm 4

Start production cycle (visit) August 2016 June 2017 June 2017 August 2017 August 2017
Hatchery Hatchery A Hatchery B Hatchery B Hatchery C Hatchery D
Age of the parent flock (weeks) 55 35 42 49 54
Type of litter Wood shavings Peat Peat Straw pellets Straw pellets
Feed supplier Supplier A Supplier B Supplier B Supplier C Supplier B
Poultry house 1 2 1 2 1 2 1 2 1 2
Size poultry house (m2 ) 1313 1313 972 968 972 968 1600 1850 1350 1730
Flock size 28 000 28 000 23 500 23 500 23 500 23 500 32 200 38 400 30 500 39 000
Antibiotic treatment No No No No No No No Yes (day 3) No Yes (day 22)

The broilers of Farm 2 cycle 1 and 2 are of the same parent flock. Each poultry house contains one broiler flock.

Materials and methods dose per animal year on farm level) were recruited for the study.
All selected farms were within the target zone for antimicrobial
Ethical statement use according to national benchmark thresholds for poultry farms
The field study was approved by the Dutch Central Authority for in 2017(SDa 2018). Notwithstanding, two flocks were treated with
Scientific Procedures on Animals and the Animal Experiments antibiotics. The flock in poultry house 2 of Farm 3 received a 3-
Committee (registration number AVD108002016442) and was car- day antibiotic treatment (trimethoprim + sulfamethoxazol) be-
ried out in compliance with all relevant legislation. tween days 3 and 6 of the production cycle, and the flock in poultry
house 2 of Farm 4 received a 3-day antibiotic treatment (amoxi-
Farm selection and data collection cillin) from day 22 onward of the production cycle (Table 2). The
Data for this study were obtained from four broiler farms in the farms were visited on days 7, 14, and 35 of the production cycle.
Netherlands, each with two similar houses. All farms had conven- On one of the farms (Farm 2), data was collected from two consec-
tional Ross 308 broilers, both male and female. The farms were se- utive production cycles. The farms received chicks from different
lected for good production performance, as we were interested in commercial hatcheries. An overview of the different farm charac-
a healthy intestinal microbiota. Also, to reduce the chance of in- teristics of the poultry flocks is provided in Table 2. The diets on all
cluding flocks treated with antibiotics, only farms with an antimi- farms were provided ad libitum and were mostly wheat-based, but
crobial use in the previous months below 15 DDDAf (defined daily there were differences in composition of the feed and feed suppli-
8 | FEMS Microbiology Ecology, 2022, Vol. 98, No. 9

(A)

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

(B)

Figure 6. Multivariate temporal compositional dynamics of cecal microbiota in different farms and houses. PRC analysis (n = 270) based on
unweighted (A) and weighted (B) UniFrac distances summarizing the multivariate comparison of cecal broiler microbiota in different farms and
houses over time. Each line corresponds to the cecal microbiota of a different poultry flock. The gray straight line at zero represents the temporal
microbiota dynamics of Farm 1, poultry flock 2, and serves as the reference. ASVs with large differences compared to the reference have large effect
sizes while ASVs with equal presence have zero weight. Taxa plotted vertically on the right axis are the main drivers of the differences between flocks.
Those with negative weights follow the negative lines, whereas those with positive weights follow the opposite pattern.

ers between farms. Table 2 provides an overview of the feed details port and Recovery (STAR) buffer (Roche Diagnostics Nederland
per time point. Coccidiostatic drugs were standardly applied in all BV, the Netherlands) and repeated bead beating. The PCR reac-
flocks (Table 2). tions contained 36.5 μl nucleotide free water (Promega, USA), 0.5
μl of 2 U μl−1 polymerase, 10 μl of 5 × HF buffer, 1 μl of 10 μM
stock solutions of each of the forward and reverse primers, 1 μl
Cecal content collection and 16S rRNA gene amplicon 10 mM dNTPs (Promega), and 1 μl template DNA. Reactions were
sequencing held at 98◦ C for 30 s and amplification proceeded for 25 cycles
At each visit, nine broilers were randomly selected for sampling at 98◦ C for 10 s, 42◦ C for 10 s, 72◦ C for 10 s, and a final exten-
from each poultry house. Coveralls, footwear, and all sampling sion of 7 min at 72◦ C. DNA concentration was measured with
materials were changed between sampling of the two poultry a NanoDrop ND-1000 spectrophotometer (NanoDrop® Technolo-
flocks on the same farm. The start of the sampling of broilers gies, USA), and samples were stored at −20◦ C until further use.
took place at least 30 min after the end of a dark-period, to avoid Barcoded amplicons covering the variable regions V5–V6, were
low amounts of content in the intestinal tract at sampling. Broil- generated by PCR amplification of the bacterial 16S rRNA gene us-
ers were individually weighed, checked for abnormalities, and eu- ing barcoded primers 784F and 1064R as described before (Ramiro-
thanized by cervical dislocation. The gastrointestinal tract was Garcia et al. 2016). To ensure high quality sequencing data, syn-
quickly but carefully removed, using a procedure that was as ster- thetic communities of known composition were used as positive
ile as possible, as previously described in detail (Kers et al. 2019). controls (Ramiro-Garcia et al. 2016), and nuclease free water as
The cecal content was stored at −80◦ C before extraction of DNA negative controls. Sequencing of resulting libraries was performed
as previously described (Kers et al. 2019). Briefly, DNA was ex- by GATC GmbH (now part of Eurofins Genomics Germany GmbH,
tracted from 0.25 g cecal content, using 700 μl of Stool Trans- Konstanz, Germany) on an Illumina Hiseq2500 instrument. Raw
 Kers et al. | 9

sequence data was analyzed using NG-tax 2.0 (Poncheewin et al. Authors’ Contributions
2020) using SILVA 128 as 16S rRNA gene reference database to as-
Conceptualization: F.C.V., J.A.S., and H.S.; methodology: F.C.V.,
sign taxonomy (Quast et al. 2013).
J.A.S., E.A.J.F., J.G.K., and H.S.; investigation: J.G.K. and F.C.V.; data
analysis: J.G.K and G.D.A.H; writing original draft: J.G.K.; writing,
Statistical analysis review, and editing: G.D.A.H, F.C.V., J.A.S., E.A.J.F., and H.S.; funding
Statistical analyses were performed in R, version 3.4.3 (R Core acquisition: F.C.V.; and supervision: F.C.V., J.A.S., E.A.J.F., and H.S.
Team 2017), using packages: Phyloseq, Microbiome, Vegan, Dirich-
letMultinomial, and RVAideMemoire (Oksanen 2010, McMurdie
and Holmes 2013, Morgan 2014, Lahti 2017, Pinheiro et al. 2018, Acknowledgments

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

Hervé and Hervé 2020). The authors wish to thank the farmers of the commercial farms
Clustering was performed according to PAM-based protocols participating in this study and their veterinarian Sible West-
using JSD (PAM–JSD) at ASV-level (Arumugam et al. 2011), BC endorp, animal technicians from the Department of Farm Ani-
(PAM–BC), UF (PAM–UF), and WUF (PAM–WUF). The optimal num- mal Health of Utrecht University and the Cargill Animal Nutri-
ber of clusters was calculated using Prediction Strength, Average tion and Thijs Manders and Amerins Mulder for assisting with
Silhouette Width (silhouette index), Calinski–Harabasz index, and the collection of the samples at the farms, and Ineke Heikamp-
Laplace approximation (Holmes et al. 2012, Koren et al. 2013). de Jong, Chrissy Veldhuizen, and Prokopis Konstanti for all the
Also, DMM, a probabilistic model, was applied to cluster the 16S lab work to process the samples. Furthermore, we are grate-
rRNA gene sequence data at ASV-level (Holmes et al. 2012). To test ful to Jean De Oliveira and David Lamot for their input on the
for differences in relative abundance of genera between clusters, manuscript.
Wilcoxon rank-sum test corrected for multiple comparisons using
Benjamini–Hochberg (BH) was used. A corrected P-value (q-value)
of < .05 was considered statistically significant.
Funding
Shannon diversity, ASV richness and Faith’s phylogenetic di- This work was supported by NWO Earth and Life Sciences (ALW)
versity (Faith 2006) were calculated to define microbial alpha di- and Cargill Animal Nutrition [grant number 868.15.020]. The fun-
versity. Differences in alpha diversity were tested with a Kruskal– ders had no role in the design of the study; in the collection, anal-
Wallis test, and pairwise comparisons were tested using Wilcoxon yses, or interpretation of data; in the writing of the manuscript, or
rank-sum tests and corrected for multiple testing with BH. Beta di- in the decision to publish the results.
versity was visualized using principal coordinates analysis (PCoA),
Conflicts of interest statement. None declared.
and nonparametric permutational analysis of variance (PER-
MANOVA) was used to analyze multivariate impact of the clusters
on the microbiota (Anderson 2001) a P-value of < .05 was consid- References
ered significant.
Anderson MJ. A new method for non-parametric multivariate anal-
To examine the multivariate differences between microbiota
ysis of variance. Austral Ecol 2001;26:32–46.
compositions of poultry houses over time, distance based princi-
Apajalahti JHA, Kettunen A, Bedford MR et al. Percent G+C profiling
pal response curve (PRC) analysis was used (Shankar et al. 2017),
accurately reveals diet-related differences in the gastrointestinal
PRC was originally developed to analyze time-series data and car-
microbial community of broiler chickens. Appl Environ Microbiol
ries out partial redundancy analysis (RDA) ordination to obtain
2001;67:5656–67.
estimates of community changes using time as a predictor vari-
Arumugam M, Raes J, Pelletier E et al. Enterotypes of the human gut
able (Van den Brink and Braak 1999).
microbiome. Nature 2011;473:174–80.
In addition, to determine which explanatory variables impact
Ballou AL, Ali RA, Mendoza MA et al. Development of the chick micro-
broiler cecal microbiota, db-RDA, based on WUF and UF distances
biome: how early exposure influences future microbial diversity.
were performed on ASV level data. To determine the most par-
Front Vet Sci 2016;3:2.
simonious model, the model that explained the microbiota vari-
Bergström A, Skov TH, Bahl MI et al. Establishment of intesti-
ation with the least number of explanatory variables, a stepwise
nal microbiota during early life: a longitudinal, explorative
selection (both directions) was used based on the Akaike informa-
study of a large cohort of danish infants. Appl Environ Microbiol
tion criterion (AIC), using 9999 number of iteration steps of drop-
2014;80:2889–900.
ping and adding terms. The most parsimonious model were tested
Borda-Molina D, Seifert J, Camarinha-Silva A. Current perspectives
for collinearity and those with a VIF > 3 were removed from the
of the chicken gastrointestinal tract and its microbiome. Comput
model one by one (Zuur et al. 2010). High VIF values suggest that
Struct Biotechnol J 2018;16:131–9.
the variable contains limited or no unique information, and there-
Borewicz K, Gu F, Saccenti E et al. Correlating Infant fecal microbiota
fore, is redundant in the set of explanatory variables.
composition and human milk oligosaccharide consumption by
microbiota of 1-month-old breastfed infants. Mol Nutr Food Res
2019;63:1801214.
Data availability
Costea PI, Hildebrand F, Arumugam M et al. Enterotypes in the land-
Sequence data was deposited into the Sequence Read Archive scape of gut microbial community composition. Nat Microbiol
(SRA) at the NCBI under accession number PRJNA844268. The data 2018;3:8–16.
files and R-scripts can be accessed through the Github page: https: Cuperus T, Kraaij MD, Zomer AL et al. Immunomodulation and
//github.com/mibwurrepo/Kers_etal_ConservedDevelopment. effects on microbiota after in ovo administration of chicken
cathelicidin-2. PLoS ONE 2018;13:e0198188.
De Filippo C, Cavalieri D, Di Paola M et al. Impact of diet in shaping
Supplementary data gut microbiota revealed by a comparative study in children from
Supplementary data are available at FEMSEC online. Europe and rural Africa. Proc Natl Acad Sci 2010;107:14691–6.
10 | FEMS Microbiology Ecology, 2022, Vol. 98, No. 9

Dethlefsen L, Huse S, Sogin ML et al. The pervasive effects of an an- Bioconductor. 2014. https://rdrr.io/bioc/DirichletMultinomial/ (1
tibiotic on the human gut microbiota, as revealed by deep 16S February 2022, date last accessed).
rRNA sequencing. PLoS Biol 2008;6:e280. Müller M, Hermes GD, Canfora EE et al. Effect of wheat bran de-
Donaldson EE, Stanley D, Hughes RJ et al. The time-course of broiler rived prebiotic supplementation on gastrointestinal transit, gut
intestinal microbiota development after administration of cecal microbiota, and metabolic health: a randomized controlled trial
contents to incubating eggs. PeerJ 2017;5:e3587. in healthy adults with a slow gut transit. Gut Microbes 2020;12:
Faith DP. The role of the phylogenetic diversity measure, PD, in bio- 1–15.
informatics: getting the definition right. Evolut Bioinformatics On- Oksanen J, Blanchet F G, Kindt R et al. Vegan: community ecology
line 2006;2:277–83. package. R Package Version 1. 2010;17–4. https://cran.r-project.o
Falony G, Joossens M, Vieira-Silva S et al. Population-level analysis of rg/web/packages/vegan/vegan.pdf (1 February 2022, date last ac-

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024

gut microbiome variation. Science 2016;352:560–4. cessed).
Gao P, Ma C, Sun Z et al. Feed-additive probiotics accelerate yet antibi- Pinheiro J BD, DebRoy S, Sarkar D. and R Core Team. nlme: linear
otics delay intestinal microbiota maturation in broiler chicken. and nonlinear mixed effects models. R Package Version 3. 2018;
Microbiome 2017;5:91. 1–137. https://cran.r-project.org/web/packages/nlme/nlme (1
Han GG, Kim EB, Lee J et al. Relationship between the microbiota February 2022, date last accessed).
in different sections of the gastrointestinal tract, and the body Poncheewin W, Hermes GD, Van Dam JC et al. NG-Tax 2.0: a seman-
weight of broiler chickens. Springerplus 2016;5:911. tic framework for high-throughput amplicon analysis. Front Genet
Hervé M, Hervé MM. Package ‘RVAideMemoire’. 2020. https://cran.r-p 2020;10:1366.
roject.org/web/packages/RVAideMemoire/index.html (1 Febru- Quast C, Pruesse E, Yilmaz P et al. The SILVA ribosomal RNA gene
ary 2022, date last accessed). database project: improved data processing and web-based tools.
Holmes I, Harris K, Quince C. Dirichlet multinomial mixtures: Nucleic Acids Res 2013;41:D590–596.
generative models for microbial metagenomics. PLoS ONE R Core Team. R: A language and environment for statistical com-
2012;7:e30126. puting. R Foundation for Statistical Computing, Vienna, Austria.
Johnson AJ, Vangay P, Al-Ghalith GA et al. Daily sampling reveals per- 2017. https://www.R-project.org/.
sonalized diet-microbiome associations in humans. Cell Host Mi- Ramiro-Garcia J, Hermes GDA, Giatsis C et al. NG-Tax, a highly
crobe 2019;25:789–802. accurate and validated pipeline for analysis of 16S rRNA
Johnson TJ, Youmans BP, Noll S et al. A consistent and predictable amplicons from complex biomes. F1000Research 2016;5:
commercial broiler chicken bacterial microbiota in antibiotic- 1791.
free production displays strong correlations with performance. SDa. Usage of antibiotics in agricultural livestock in the Netherlands
Appl Environ Microbiol 2018;84:e00362–18. in 2017. Utrecht. 2018.
Jurburg SD, Brouwer MS, Ceccarelli D et al. Patterns of community Shankar V, Agans R, Paliy O. Advantages of phylogenetic distance
assembly in the developing chicken microbiome reveal rapid pri- based constrained ordination analyses for the examination of mi-
mary succession. MicrobiologyOpen 2019;8:e00821. crobial communities. Sci Rep 2017;7:6481.
Kaakoush NO, Sodhi N, Chenu JW et al. The interplay between Stanley D, Geier MS, Hughes RJ et al. Highly variable microbiota
Campylobacter and Helicobacter species and other gastroin- development in the chicken gastrointestinal tract. PLoS ONE
testinal microbiota of commercial broiler chickens. Gut Pathog 2013;8:e84290.
2014;6:18. Stanley D, Hughes RJ, Geier MS et al. Bacteria within the gastroin-
Kers JG, Velkers FC, Fischer EAJ et al. Take care of the environment: testinal tract microbiota correlated with improved growth and
housing conditions affect the interplay of nutritional interven- feed conversion: challenges presented for the identification of
tions and intestinal microbiota in broiler chickens. Anim Micro- performance enhancing probiotic bacteria. Front Microbiol 2016;7:
biome 2019;1:10. 187.
Koren O, Knights D, Gonzalez A et al. A guide to enterotypes Stanley D, Hughes RJ, Moore RJ. Microbiota of the chicken gastroin-
across the human body: meta-analysis of microbial commu- testinal tract: influence on health, productivity and disease. Appl
nity structures in human microbiome datasets. PLoS Comput Biol Microbiol Biotechnol 2014a;98:4301–10.
2013;9:e1002863. Stanley D, Wu SB, Rodgers N et al. Differential responses of ce-
Kovatcheva-Datchary P, Nilsson A, Akrami R et al. Dietary fiber- cal microbiota to fishmeal, Eimeria and Clostridium perfringens
induced improvement in glucose metabolism is associated with in a necrotic enteritis challenge model in chickens. PLoS ONE
increased abundance of prevotella. Cell Metab 2015;22:971–82. 2014b;9:e104739.
Lahti L, Shetty S. Tools for Microbiome Analysis in R. Microbiome Stewart CJ, Ajami NJ, O’Brien JL et al. Temporal development of the
Package Version 1.15.1. 2017. Available online at: https://microb gut microbiome in early childhood from the TEDDY study. Nature
iome.github.io/tutorials/ (1 February 2022, date last accessed). 2018;562:583–8.
Lu J, Idris U, Harmon B et al. Diversity and succession of the intestinal Sundu B. Gastro-intestinal response and passage time of pel-
bacterial community of the maturing broiler chicken. Appl Envi- leted diets in digestive tract of broilers. Int J Poult Sci 2009;8:
ron Microbiol 2003;69:6816–24. 976–9.
McMurdie PJ, Holmes S. phyloseq: an R package for reproducible in- Svihus B. Function of the digestive system. J Appl Poult Res
teractive analysis and graphics of microbiome census data. PLoS 2014;23:306–14.
ONE 2013;8:e61217. Thibodeau A, Fravalo P, Yergeau E et al. Chicken caecal micro-
McWhorter TJ, Caviedes-Vidal E, Karasov WH. The integration of di- biome modifications induced by Campylobacter jejuni coloniza-
gestion and osmoregulation in the avian gut. Biol Rev 2009;84: tion and by a non-antibiotic feed additive. PLoS ONE 2015;10:
533–65. e0131978.
Morgan M. DirichletMultinomial: Dirichlet-Multinomial Mixture Uhr GT, Dohnalová L, Thaiss CA. The dimension of time in host-
Model Machine Learning for Microbiome Data version 1.24.1 from microbiome interactions. Msystems 2019;4:e00216–18.
 Kers et al. | 11

Van den Brink PJ, Braak CJT. Principal response curves: analysis of Yuan Z, Yan W, Wen C et al. Enterotype identification and its influ-
time-dependent multivariate responses of biological community ence on regulating the duodenum metabolism in chickens. Poult
to stress. Environ Toxicol Chem 1999;18:138–48. Sci 2020;99:1515–27.
Wu GD, Chen J, Hoffmann C et al. Linking long-term dietary patterns Zhong H, Penders J, Shi Z et al. Impact of early events and lifestyle on
with gut microbial enterotypes. Science 2011;334:105–8. the gut microbiota and metabolic phenotypes in young school-
Yin Y, Lei F, Zhu L et al. Exposure of different bacterial inocula to age children. Microbiome 2019;7:2.
newborn chicken affects gut microbiota development and ileum Zuur AF, Ieno EN, Elphick CS. A protocol for data exploration to avoid
gene expression. ISME J 2010;4:367–76. common statistical problems. Methods Ecol Evol 2010;1:3–14.

Downloaded from https://academic.oup.com/femsec/article/98/9/fiac090/6649812 by Uluslararas Kbrs University user on 03 April 2024