Anal Bioanal Chem (2017) 409:1185–1194

DOI 10.1007/s00216-016-0116-6

RESEARCH PAPER

In ovo sexing of chicken eggs by fluorescence spectroscopy

Roberta Galli 1 & Grit Preusse 1 & Ortrud Uckermann 2 & Thomas Bartels 3 &
Maria-Elisabeth Krautwald-Junghanns 3 & Edmund Koch 1 & Gerald Steiner 1,4

Received: 6 October 2016 / Revised: 11 November 2016 / Accepted: 24 November 2016 / Published online: 14 December 2016
# Springer-Verlag Berlin Heidelberg 2016

Abstract Culling of day-old male chicks in production of
laying hen strains involves several millions of animals every
year worldwide and is ethically controversial. In an attempt to
provide an alternative, optical spectroscopy was investigated to
determine nondestructively in ovo the sex of early embryos of
the domestic chicken. The extraembryonic blood circulation
system was accessed by producing a window in the egg shell
and the flowing blood was illuminated with a near-infrared
laser. The strong fluorescence and the weak Raman signals
were acquired and spectroscopically analyzed between 800
and 1000 nm. The increase of fluorescence intensity between
3.5 and 11.5 days of incubation was found to be in agreement
with the erythropoietic stages, thus enabling to identify hemo-
globin as fluorescence source. Sex-related differences in the
fluorescence spectrum were found at day 3.5, and principal
Dedicated to Professor Reiner Salzer on the occasion of his 75th birthday.
Professor Reiner Salzer is an internationally recognized leader in
analytical chemistry and pioneered spectroscopic methods.
* Roberta Galli
roberta.galli@tu-dresden.de
* Gerald Steiner
gerald.steiner@tu-dresden.de
1
Faculty of Medicine, Anesthesiology and Intensive Care Medicine,
Clinical Sensoring and Monitoring, Technische Universität Dresden,
Fetscherstrasse 74, 01307 Dresden, Germany
2
Faculty of Medicine, University Hospital Carl Gustav Carus,
Neurosurgery, Technische Universität Dresden, Fetscherstrasse 74,
01307 Dresden, Germany
3
Faculty of Veterinary Medicine, Clinic for Birds and Reptiles,
University of Leipzig, An den Tierkliniken 17,
04103 Leipzig, Germany
4
Faculty of Physics, Vilnius University, Sauletekio av. 9 bl. 3,
10222 Vilnius, Lithuania
component (PC) analysis showed that the blood of males was
characterized by a specific fluorescence band located at
∼910 nm. Supervised classification of the PC scores enabled
the determination of the sex of 380 eggs at day 3.5 of incubation
with a correct rate up to 93% by combining the information
derived from both fluorescence and Raman scattering.
Keywords Optical spectroscopy . Fluorescence . Raman
scattering . Chicken embryo . Sexing . In ovo
Introduction
The possibility to determine the sex of birds Bin ovo^ is driv-
ing increasing attention as a potential method to overcome the
culling of day-old male chicks in poultry industry. Laying hen
strains of modern breeding differ from broiler strains, so that
male birds of the laying strain are not profitable for meat
production. Therefore, day-old cockerels are culled directly
in the hatchery. This practice involves a very large number
of animals: approximately 370 million in North America and
420 million in Europe every year [1]. Only in Germany, 40
millions of chicks are killed every year according to animal
welfare legislation by asphyxiation with carbon dioxide or by
grinding [2]. Killing of day-old chicks is considered as prob-
lematic and ethical issues have triggered increasing research
aimed to provide alternatives [3].
The optical spectrum contains information about the bio-
chemical composition and/or the structure of a biological sam-
ple and can provide the information about the sex as well.
Vibrational spectroscopic techniques have been applied to
the sexing of birds, either of hatched birds as well as of incu-
bated and unincubated eggs. UV resonance Raman spectros-
copy [4] and Fourier-transform infrared absorption spectros-
copy [5, 6] were used to retrieve the sex of hatched animals
1186
based on DNA differences by analyzing cell material extract-
ed from the feather pulp. The use of Fourier-transform infrared
spectroscopy was also reported earlier by our group for sex
recognition based on the spectrum of germinal cells obtained
from unincubated eggs [7].
Optical methods for in ovo sexing have the advantage of
being applicable in situ without taking samples, and can pro-
vide real-time sexing by eliminating the need to wait for the
results of chemical or genetic analyses performed on previ-
ously extracted samples. Therefore, optical methods have
clear advantages toward an industrial exploitation when com-
pared to other non-destructive approaches proposed for in ovo
sexing that are based on measurement of hormone [8–10] and
DNA analysis [11–13], which all require the extraction of egg
material or fluids. Totally non-invasive methods, like the se-
lection of unincubated chicken eggs based on morphometric
parameters, can only bias to a low extent the ratio between
sexes [14]. There is also evidence that egg odor encodes sex
information; however, this was not exploited for sexing [15].
We recently reported near-infrared Raman spectroscopy for
in ovo sexing of incubated chicken eggs, showing that it pro-
vides correct sexing up to 90% without hindering embryo
development [16]. We demonstrated that, already during the
fourth day of incubation (i.e., 84 ± 4 h), Raman measurements
can be performed directly on the blood that flows in the ex-
traembryonic vessels of the vitelline circulation avoiding all
damage to the embryo. The Raman spectra delivered the bio-
chemical fingerprint of the embryonic blood, from which the
sex information was obtained by use of supervised classifica-
tion algorithms. In this study, we observed that the fluores-
cence background intensity was significantly different be-
tween sexes, but this information was not exploited because
of the very high overlap between sexes.
In Raman spectroscopy, the presence of an intense fluores-
cence background is normally considered a pitfall rather than a
source of information and thus suppressed or removed during
data processing [17]. Fluorescence spectra of biological tissue
are broad compared with Raman and FT-IR spectra, and they do
not carry such detailed biochemical information. However,
spectral fluorescence-based methods have continuously evolved
and improved enabling to address new cellular features [18].
Here, we show that spectral analysis of the near-infrared
fluorescence signal of the blood flowing in the extraembryon-
ic vessels can indeed provide sex information of the domestic
chicken eggs (Gallus gallus f. dom.). For this purpose, we
analyzed spectroscopically the backscattered radiation of
blood illuminated with a 785 nm laser, repeating the measure-
ments at different time points of the incubation and comparing
with the time course of erythropoiesis to identify the source of
the fluorescence signal. Afterwards, we exploited the fluores-
cence spectral profile for sex determination and compared the
results with the combination of fluorescence and Raman
scattering.
R. Galli et al.
Methods
Egg handling
Fertilized eggs of a white layer strain (LSL—Lohmann
Selected Leghorn) were used in all experiments. They were
obtained from Lohmann Tierzucht GmbH (Cuxhaven,
Germany). The freshly laid eggs were inspected for any shell
damage and then stored at approximately 14 °C. Immediately
before starting the incubation, egg shells were windowed
using a 30 W CO2 laser (Firestar v30, Synrad, Mukilteo,
Washington, USA) equipped with scanning head (FH Flyer,
Synrad, Mukilteo, Washington, USA). The shells were laser-
scribed at the pointed end along a round path of 12 mm diam-
eter, without removing the shell window. The incubation was
performed in an automatic egg incubator (Favorit Olymp 192,
Heka-Brutgeräte, Rietberg, Germany) at 37.8 °C and humidity
of 53%, in vertical position with the pointed end downward. A
±45° tilting at an interval of 3 h was applied during the incu-
bation time until day 3.5 (i.e., 84 h). At this time point, the
eggs were subjected to the measurement.
The shell window was gently removed using a scalpel and
the spectrum was acquired. After measurement, embryonic
tissue samples were isolated for subsequent molecular sexing,
or the shell windows were closed using a biocompatible ad-
hesive tape (Leukosilk, BNS Medical GmbH, Hamburg,
Germany) and further incubated to perform time-course ex-
periments. Measurements were repeated at day 4.5, 5.5, 6.5,
7.5, 9.5, and 11.5. At each time, the tape was removed before
the measurement and reapplied afterwards. The incubation
between the measurements was performed with the egg in
upright position to maintain the embryo and the main blood
vessels optically accessible. Before measurement, the eggs
were inspected and the spectrum was acquired only on blood
vessels of vital embryos. Otherwise, samples of embryonic
tissue were isolated and kept frozen at −80 °C for subsequent
sex determination. After the last measurement at day 11.5, all
embryos were isolated and kept frozen for sex determination
as well.
Molecular sexing
Reference sexing was obtained with genetic analysis on em-
bryonic tissue based on polymerase chain reaction (PCR).
Alkaline extraction of DNA was performed as described else-
where [19]. The samples were treated in NaOH for 20 min at
75 °C and subsequently neutralized with Tris–HCl (pH = 7.5).
Afterwards, the samples were centrifuged for 10 min at
14,000 rpm and the supernatants transferred to reaction tubes.
The DNA content of the supernatant was measured using the
Genesys 10 Bio UV–vis spectrophotometer (Thermo Fisher
Scientific Inc., Waltham, MA, USA). The amplification of the
CHD-1 gene was performed in the PCR cycler T Gradient
In ovo sexing of chicken eggs by fluorescence spectroscopy
(Biometra GmbH, Göttingen, Germany). The primers and the
temperature profile are described elsewhere [20, 21]. Finally,
the amplified PCR products were separated by ethidium bro-
mide agarose gel electrophoresis and visualized by UV light.
Spectroscopy
Spectroscopy was performed with a spectrometer RamanRxn
(Kaiser Optical Systems Inc., Ann Arbor, USA). The excita-
tion was obtained with a diode laser emitting at a wavelength
of 785 nm (Invictus 785-nm NIR, Kaiser Optical Systems
Inc., Ann Arbor, USA). A fiber optic probe (MR-Probe-785,
Kaiser Optical Systems Inc., Ann Arbor, USA) was used in
the experiments. The excitation fiber had a core diameter of
62.5 μm and the collection fiber of 100 μm. The fiber probe
was coupled to a self-build microscopy system that enabled
coaxial vision. The microscopy system was composed by
commercial elements and it is pictured in Fig. 1.
A Keplerian telescopic system was used as beam-expander
to collimate the laser beam and to match the diameter to the
objective pupil. The beam-expander was connected to a 45°
filter cube that contained a short-pass dichroic mirror with
edge at 670 nm (FF670-SDi01, Semrock Inc., Rochester,
New York, USA). The mirror reflected to 90° the laser exci-
tation in the microscope objective ×20/0.4NA Plan Apo NIR
(Mitutoyo Corp., Kanagawa, Japan), which was also mounted
on the filter cube. The laser spot in the focus had a diameter of
∼55 μm and the measured laser power was 160 mW. The light
was collected in reflection mode by the objective. The collect-
ed near-infrared light with wavelength above 670 nm was
reflected by the dichroic mirror back to the fiber probe and
propagated to the spectrograph. The collected visible light was
transmitted through the dichroic mirror and used to obtain the
Fig. 1 Schema of the microscopy
system with inset showing the egg
shell window and the main
vitelline vessels suited for in ovo
spectroscopic measurement
1187
image of the sample. For this purpose, a CCD camera with
50 mm focal length lens and a NIR hot mirror were mounted
on the aperture of the filter cube that faces the objective. In
order to maximize the visibility of perfused blood vessels, side
illumination with green LEDs was employed. A motorized x-
y-z micrometer stage was used to move the egg and bring a
blood vessel in the laser spot. Blood vessels with diameter
larger than 100 μm were manually chosen for the measure-
ment. An autofocus system based on blood flow detection in
the camera images was used to set the laser focus inside the
blood vessel. A tracking system remained active during the
whole acquisition to compensate all movements of the blood
vessel. The autofocus and tracking software modules are de-
scribed elsewhere [22]. An enclosure of the system provided
rejection of ambient light and protection from reflected or
scattered laser light during acquisition.
Total acquisition time was set to 40 s (20 accumulations of
2 s). The acquired spectral range was from 794 to 1054 nm
(i.e., from 150 to 3250 rel. cm−1) and the spectral resolution
was 0.3 nm (i.e., approximately 4 cm−1).
Data analyses
Spectroscopic data were analyzed using the MATLAB pack-
age (MathWorks Inc., Natick, USA) and statistics were calcu-
lated with Prism 6.0 (Graph Pad Software Inc., La Jolla, CA,
USA).
The intensity of the fluorescence was calculated as sum
area under the spectra in the range 820–1000 nm. Principal
component analysis was performed on the raw spectra by
using the MATLAB function princomp. Classification was
performed using supporting vector machine. The MATLAB
functions svmtrain was employed to train the classifier, by
1188 R. Galli et al.

using a quadratic kernel and least-square method to find the

separating hyperplane. Afterwards, the function svmclassify
was employed in order to retrieve the classification.

Results and discussion

Spectral analysis of blood fluorescence

In the chicken egg, blood islands appear toward the end of the
first day of incubation and already at day 2 of incubation the
blood circulation starts driven by the primitive heart. Between
day 3 and 4, the vascularized area of the yolk sac is roughly
circular and reaches a diameter between 3 and 4 cm at day 4.
The major blood vessels are the paired lateral vitelline arteries
and veins, and the unpaired anterior and posterior vitelline
veins. The branching pattern of arteries and veins is dichoto-
mous, creating a treelike topology. Around day 8, the respira-
tory function is transferred to the blood vessels of the newly
formed chorioallantoic membrane [23]. When the egg is
brought in vertical position at day 3.5 of incubation, the
vascularized area moves on the top and, after opening of the
shell window, it remains on the surface, so that the vessels can
be optically sampled through the shell window (Fig. 1, inset).
The position of measurement was always chosen on a large
vessel (diameter larger than 100 μm), preferably in main lat-
eral or anterior/posterior veins.
The backscattered signal generated upon irradiation of ex-
traembryonic blood vessels with a laser beam at 785 nm was
spectroscopically analyzed daily on a set of 27 eggs from 3.5
to 7.5 days of incubation, and then at 9.5 and 11.5 days. Six
embryos died during the time course of experiments because
of egg incubation in upright position without the required
periodic tilting, repeated egg removal from the incubator, or
opening of the shell window in order to perform the measure-
ments. The mean spectra calculated for each day of measure-
ment are shown in Fig. 2a. A weak Raman signal is
superimposed on the strong fluorescence. The fluorescence
intensity was calculated as area under the curves and is shown
in Fig. 2b. The signal above 1000 nm (2740 rel. cm−1) was not
included in the calculation as it is dominated by the Raman Fig. 2 a Mean blood spectra acquired in ovo from day 3.5 until day 11.5
bands of CH and OH. The fluorescence increased with the of incubation. b Time course of total blood fluorescence acquired in ovo
in the range 820–1000 nm (mean ± SD). The number of eggs included in
incubation time until day 7.5 and tended to decrease the statistics is indicated. c Spectra of plasma and erythrocytes (red blood
afterwards. cells—RBC) separated by centrifugation of embryonic chicken blood
Different erythropoietic phases take place starting at day 2. extracted at day 11.5 of incubation; the RBC spectrum was acquired
The primitive erythrocytes undergo six rounds of mitosis until using a twentieth of the laser power to avoid thermal damage
day 5. In the middle of day 5, the definitive erythrocytes start
entering the circulation and gradually replace the primitive increase of hemoglobin content was reported until day 5,
line by day 7. By day 8–9, the erythrocytes have completed followed by a plateau between day 5 and 6, corresponding
their maturation [24]. The time course of fluorescence is in to the transition from the proliferative to the post mitotic phase
agreement with the time course of hematocrit observed during of primitive erythrocytes. This plateau can be seen also in the
erythropoiesis [25], and mimics the increase of hemoglobin time course of fluorescence. Afterwards, the hemoglobin con-
measured in embryonic and mature erythrocytes [26]. A linear tent increases again until day 9 and slightly decreases
In ovo sexing of chicken eggs by fluorescence spectroscopy
afterwards, when the erythrocyte maturation is concluded.
Fluorescence intensity displays a similar trend at day 9.5 and
11.5 as well. The comparison between the time courses of
fluorescence intensity and hemoglobin content indicates that
hemoglobin is the main source of the observed NIR
fluorescence.
Heme in red blood cells is a molecule that belongs to the
family of porphyrin complexes. Porphyrins normally display
strong reddish-orange fluorescence. However, heme consti-
tutes an exception, as the fluorescence of the porphyrin ring
is quenched by the coordinated iron [27]. Steady state fluores-
cence of heme in the UV range is nevertheless observed and
attributed to fluctuations in the protein structure that partly
neutralize the quenching. The source of steady state fluores-
cence is identified in the tryptophan residues [28]. Moreover,
it was already shown in vitro that excitation at 785 nm of
methemoglobin over the concentration range spanning the
normal capillary blood range produces a fluorescence that
increases linearly with increasing hemoglobin volume percent
[29], although fluorescence detected in vivo was attributed to
both plasma and hemoglobin [30].
In our experiments, the fluorescence was observed only
during irradiation of blood (i.e., acquiring the signal from
perfused blood vessels and from the heart chambers), while
other embryonic tissue, egg components (egg white and yolk),
as well as extraembryonic tissue (vitelline and corioallantoic
membranes) did not display any fluorescence, but only Raman
scattering. Moreover, blood taken from embryos at day 11.5 of
incubation was centrifuged to separate erythrocytes from plas-
ma and then the two components were subjected to spectros-
copy. While the erythrocytes displayed strong fluorescence
signal, plasma displayed Raman scattering only (Fig. 2c).
It is also worth to note that any bleaching of fluorescence
during irradiation of perfused blood vessels is prevented by
the blood flow itself. The blood flow velocity in the center of
large vitelline vessels is around 1 mm/s already at early devel-
opment stages [31], and therefore the exposure time of blood
cells traveling through the laser spot is as short as 0.05 s.
The embryos were sexed by PCR and the time course of the
fluorescence for both sexes was analyzed on a subset of eggs
that survived until the end of experiments. The mean spectra
for both sexes are shown in Fig. 3a from day 3.5 until day 11.5
on incubation. Male and female egg fluorescence exhibited
the largest difference and lowest variability at day 3.5
(Fig. 3b). Starting at day 4.5, the intensity difference progres-
sively declined. Male blood was characterized by more in-
tense fluorescence until day 7.5. While the fluorescence inten-
sity of female blood steadily increased with the incubation
time, fluorescence intensity of male blood did not increase
from day 3.5 to day 4.5. At this point of the incubation, the
main changes in the fluorescence of male blood affected the
spectral shape. After day 9.5 of incubation, the mean fluores-
cence intensities of both sexes became similar. Higher
1189
fluorescence of male blood at early incubation stages may be
likely related to faster erythropoiesis in male embryos. It is
known that in the first phases of incubation (∼30 h), male
embryos display faster development [32]. Moreover, it was
reported that at later incubation stages (after day 13), male
embryos possess higher hematocrit [33]. Our data evidenced
that developmental differences between sexes exist for the
whole first third of incubation.
The spectral differences between male and female blood
fluorescence detected at day 3.5 may be used for in ovo
sexing. We also have shown previously that at this time point,
it is possible to measure in ovo without impairing embryo
development [16].
In ovo sexing by means of fluorescence
In order to verify whether the differences between the fluores-
cence spectra of male and female blood at day 3.5 of incuba-
tion enable a reliable sexing of eggs, the measurements were
performed on 380 fertilized eggs incubated until 84 ± 4 h.
Reference PCR showed that 199 eggs contained a male and
181 eggs a female embryo. Figure 4a shows the mean spectra
and the corresponding standard deviations. The spectral dif-
ferences of the fluorescence between the sexes were con-
firmed, being the average intensity of fluorescence much
stronger for males. However, there is overlap between the
two groups. In Fig. 4b, the signal intensity calculated as area
under the spectra in the range 820–1000 nm is shown for all
measured eggs, together with the mean value and the standard
deviation. It becomes evident from these data that the lower
signal intensities are rather close for both sexes, while higher
intensities are characteristics of males, with only one excep-
tion in the female group. This indicates that, although carrying
sex information, the evaluation of the fluorescence intensity
alone is not sufficient for sex recognition with high correct
rate. Moreover, the intensity of males is affected by a larger
variability compared to the one of females, as indicated by the
larger standard deviation (SD).
Principal component analysis was applied in order to gain
insights in the different spectral contributions that are related
to the sex. Mathematically, principal component analyses en-
ables extraction of the most important information from the
observation data table, by performing a procedure of linear
decomposition and computing new variables called principal
components (PC). The values of these new variables are called
scores, and can be interpreted as the projections of the obser-
vations onto the PCs [34]. This approach allows discrimina-
tion among spectral groups using scatter plots of scores, and
the loading vectors convey the spectral variations that differ-
entiate the data [35].
Score intensities for the PC #1 to #8 are shown in
Fig. 5a, and the corresponding loading vectors in Fig. 5b.
The mean scores of PC #1, #2, #3, and #7 are significantly
1190 R. Galli et al.

Fig. 3 a Mean blood spectra of

male (blue) and female (red)
embryos (n = 7 for each sex)
acquired in ovo from day 3.5 until
day 11.5 of incubation. b Intensity
difference (male − female), mean
and SD

different between the sexes (two-tailed t test with Welch
correction, p < 0.001). The first PC score is higher for
males and accounts for the overall higher fluorescence in-
tensity. The loading vector of PC #1 represents the mean
Fig. 4 a Mean spectra of blood acquired from female and male eggs in
the range 820–1000 nm; the overlapping range of SD is indicated in gray.
b Total area intensity calculated in the same range as shown in (a) with
mean value and SD
spectrum and includes both fluorescence and Raman scat-
tering signals. The higher components describe the
Bspectral^ variance of the data. For instance, PC #2 ac-
counts for a variance as high as 97.5%. The score is higher
for males and the loading vector represents a fluorescence
band which is typical of male blood. PC #3 accounts for a
variance of 1.3%; the score is higher for females and ac-
counts for differences in the Raman signal. The Raman
bands of the loading vector indicate presence of proteins
(at 1003, 1085, 1304, 1445, and 1665 rel. cm−1 [36]) and
of nucleic acids (at 780 and 826 rel. cm−1 [36]), while
bands of hemoglobin were not observed. Therefore, this
component might be interpreted as blood cells different
from erythrocytes. The overall spectrum appears very sim-
ilar to the one reported for blood immune cells [37].
Although the immune system of the embryo is not yet
developed, early presence at least of macrophages is
known [38]. Moreover, hematopoietic stem cells and pro-
genitor cells of myeloid and lymphoid lineages were found
in the embryonic blood during the fourth day of incubation
[39]. Therefore, the interpretation of this component
remained tentative. PCs #4 and #5 account for a variance
of 0.6 and 0.2%, respectively. Both loading vectors of PC
#4 and PC #5 are dominated by Raman bands of unsatu-
rated lipids at 1300, 1440, 1656, and 1748 rel. cm−1 [36].
For instance, the loading vector of PC #5 was found fully
consistent with the Raman spectrum of yolk, which has
been shown earlier [16]. This component likely accounts
for sampling of egg material outside the blood vessel and is
justified by the large laser spot and non-confocal micro-
scope configuration that was used in the experiments. PCs
#6 and #7 account for variances close to 0.1% and were
interpreted as variations of the protein profile: the loading
vector of PC #6 is characterized by bands typical of pro-
teins with cyclic ring at 1003, 1210, 1545, and 1606 rel.
cm−1, and the one of PC #7 by Raman bands of amide
In ovo sexing of chicken eggs by fluorescence spectroscopy
Fig. 5 a First eight PC scores;
two-tailed t test with Welch cor-
rection, ***p < 0.001. b First
eight PC loading vectors; the po-
sition of Raman bands discussed
in the text is indicated in rel. cm−1
vibrations at 1225, 1563, and 1637 rel. cm−1 [36]. Finally,
the loading vector of PC #8, which accounts for a variance
lower than 0.1%, contains vibration bands of protein cyclic
rings at 1210 and 1545 rel. cm−1 [36] and thus represents
changes in the protein profile, too.
In order to test a classification for in ovo sexing based
on PC scores, the dataset was randomly split in two groups,
which were used as training set (n = 190) and test set
(n = 190). The training set was used to create the
1191
classification model, while the test set was used to deter-
mine model performances. Supporting vector machine
(SVM) was used as solution to the two-class (that is to
say, sexes) problem. SVM creates a boundary in the form
of a hyperplane between two groups that maximize the
margin between the most similar samples in each group
[40]. For instance, SVM was applied on the PC scores to
find the quadratic hyperplane that separates the sexes of the
training set. By using the scores of the first two PCs only—
1192
i.e., the fluorescence information alone (Fig. 6a)—the
training set was reclassified with a correct rate of 81%
(females 76/91, males 78/99), and the test set was classi-
fied with a correct rate of 85% (females 75/90, males 86/
100). By including higher PCs, the correct rate of classifi-
cation increased with the dimensionality (Fig. 6b), and the
best performances were attained by using the first eight
PCs. In this case, the training set was reclassified with a
correct rate of 93% (females 85/91, males 91/99), and the
test set was classified with a correct rate of 91% (females
81/90, males 91/100). By further increasing the dimension-
ality, the algorithm became overfitted: the classification
rate of the training set further increased, while the one of
the test set decreased.
The accuracy is limited by the variability of the data, which
was likely related to intrinsic variability of blood composition.
Both intensity and spectral shape of fluorescence depended
from the development stage of the embryo. The measurements
spanned over ∼8 h, and it is expected that developmental
variations may occur in embryos even though all eggs are
placed in the incubator at the same time. These differences
exist inside the same setting of eggs depending from
Fig. 6 a Scatter plot of PC #2 vs. PC #1 scores and separating quadratic
function found with supporting vector machine. b Classification rate as
function of the dimensionality
R. Galli et al.
differences in the viability and vigor of embryos, size of indi-
vidual eggs, or from local temperature gradients inside the
incubation chamber [41]. Experimental effects related to dif-
ferent volume sampling due to blood vessel dimensions or
variation of the focal position did not affect the classification.
This is proven by lack of any correlation between the score of
PC #5 (representative of yolk) and the misclassified spectra.
The mean score intensity calculated for wrongly classified
spectra is −0.002 ± 0.062 (mean ± SD, n = 31) vs. 0.002 ±
0.050 calculated for the correctly classified ones (mean ±
SD, n = 349). The mean score intensities are not statistically
different (two-tailed t test, p = 0.7).
Compared to our previous research, the present study pro-
vides significant improvements. We first proposed infrared ab-
sorption spectroscopy of the feather pulp to distinguish the sex
of hatched birds [5, 6]. The studies were performed on a small
number of animals (turkeys and pigeons), as pilot studies aimed
to demonstrate that optical spectroscopic methods can retrieve
sex-related biochemical differences in birds. Furthermore, we
investigated infrared spectroscopy of the germinal disk extract-
ed from unincubated chicken eggs [7]. Sexing was achieved
with high accuracy, but the approach is not transferrable in the
practice because windowing of unincubated eggs causes high
occurrence of embryo mortality and morbidity [42, 43]. By
analyzing the Raman spectra of embryonic blood, we achieved
sex determination with an accuracy comprised between 88%
and 90% [16]. Moreover, the use of NIR laser excitation
avoided phototoxic effects on embryos, so that healthy chicks
hatched from the measured eggs. Here, we retained the exper-
imental configuration with NIR excitation in order to rule out
negative effects on embryos, but all spectral features of the
backscattered spectra were used for sexing, thus exploiting
the sex information conveyed by fluorescence too, and increas-
ing the correct classification rate well over 90%.
The exploitation of multiple information contained in the
backscattered light might open new possibilities for improve-
ment of in ovo sexing accuracy. Fluorescence intensity, fluo-
rescence spectral shape, and Raman scattering encode sex-
related information that are associated to different blood com-
ponents. Several approaches of spectral preprocessing could
be applied in order to Bamplify^ a specific feature and after-
wards a multi-parameter classification strategy developed to
better comply with intrinsic data variability, thus further in-
creasing the sexing accuracy. Moreover, the exploitation of
fluorescence may contribute to overcome some disadvantages
associated with Raman spectroscopy of biological samples, as
Raman scattering is a low-intensity process which requires
highly efficient laser sources, low-noise detectors, effective
Rayleigh filters, and high-throughput optics [19]. As fluores-
cence intensity is some orders of magnitude larger than the
Raman scattering signal, it might enable to largely simplify
the experimental setup and facilitate the future industrial de-
ployment of optical techniques for in ovo sexing.
In ovo sexing of chicken eggs by fluorescence spectroscopy
Conclusion
The results show that the sex information can be extracted
from intensity and spectral shape of the near-infrared fluores-
cence of embryonic blood. Source of fluorescence was iden-
tified in the hemoglobin and the spectral features of fluores-
cence depend from the hematopoietic stage. The intrinsic var-
iability of development stages affected the accuracy of sexing
based exclusively on fluorescence. As the Raman scattering
was simultaneously acquired with the fluorescence signal, it
could be employed to further increase the sexing correct rate
well over 90%, with a significant improvement compared to
sex determination based on Raman spectra only.
The ideal criteria for sex determination have been defined
since years and include lack of negative effects on embryo
development and practicability on a large scale [44]. In ovo
sexing based of spectral analysis of the backscattered radiation
satisfies these requirements, as it is non-invasive, does not
require extraction of egg material, and does not use consum-
ables. Moreover, the method is applicable during the fourth
day of incubation, before onset of embryo sensitivity at day 7
[45], thus in agreement with animal welfare.
The exploitation of fluorescence offers the potential to de-
velop industrial systems for egg sexing that are not based on
expensive spectrometers, but just make use of few light detec-
tors with suited bandpass filters to measure the signal intensity
in selected spectral ranges. Afterwards, the sex information
may be retrieved with simple calculations, such as ratio of
intensities in order to make the method robust toward varia-
tions of laser power and detection efficiency. The availability
of a cheap and easier approach alternative to spectroscopy
might contribute to a broader diffusion of optical sexing in
the hatchery practice. On an international scale, development
of a practicable technique for in ovo sex determination has the
potential to contribute to the prevention of annual culling of 7
billion male layer hybrids, whose female siblings produce the
current global demand of about 68.3 million tons of eggs per
year.
Acknowledgments The authors gratefully acknowledge Andrea
Büchner for performing molecular genetic analyses and Lohmann
Tierzucht GmbH (Cuxhaven, Germany) for providing eggs. Special
thanks to Prof. Rudolf Preisinger and Dr. Anke Förster (Lohmann
Tierzucht GmbH, Cuxhaven, Germany) for the insightful discussions
about hatchery practice.
Compliance with ethical standards
Funding This work was financially supported by the German Federal
Ministry of Food, Agriculture, and Consumer Protection (BMELV)
through the Federal Office for Agriculture and Food (BLE), grant no.
511–06.01-28-1-33.010-07.
Conflict of interest Patent applications for in ovo sexing with the
methods described in this paper are pending.
1193
References
1. Turner J. Animal reproduction, human control. In: Bekoff M, editor.
Encyclopedia of animal rights and animal welfare. 2nd ed. Santa
Barbara: ABC-CLIO, LLC; 2010. p. 30–6.
2. Tierschutz-Schlachtverordnung - TierSchlV (BGBl. I S. 2982). 20-
12-2012.
3. Bruijnis MRN, Blok V, Stassen EN, Gremmen HGJ. Moral BLock-
In^ in responsible innovation: the ethical and social aspects of kill-
ing day-old chicks and its alternatives. J Agric Environ Ethics.
2015;28(5):939–60.
4. Harz M, Krause M, Bartels T, Cramer K, Rosch P, Popp J. Minimal
invasive gender determination of birds by means of UV-resonance
Raman spectroscopy. Anal Chem. 2008;80(4):1080–6.
5. Steiner G, Bartels T, Krautwald-Junghanns ME, Boos A, Koch E.
Sexing of turkey poults by Fourier transform infrared spectroscopy.
Anal Bioanal Chem. 2010;396(1):465–70.
6. Steiner G, Preusse G, Zimmerer C, Krautwald-Junghanns ME,
Sablinskas V, Fuhrmann H, et al. Label free molecular sexing of
monomorphic birds using infrared spectroscopic imaging. Talanta.
2016;150:155–61.
7. Steiner G, Bartels T, Stelling A, Krautwald-Junghanns ME,
Fuhrmann H, Sablinskas V, et al. Gender determination of fertilized
unincubated chicken eggs by infrared spectroscopic imaging. Anal
Bioanal Chem. 2011;400(9):2775–82.
8. Tran HT, Ferrell W, Butt TR. An estrogen sensor for poultry sex
sorting. J Anim Sci. 2010;88(4):1358–64.
9. Phelps P, Bhutada A, Bryan S, Chalker A, Ferrell B, Neuman S,
et al. Automated identification of male layer chicks prior to hatch.
World Poult Sci J. 2003;59(1):33–8.
10. Weissmann A, Reitemeier S, Hahn A, Gottschalk J, Einspanier A.
Sexing domestic chicken before hatch: a new method for in ovo
gender identification. Theriogenology. 2013;80(3):199–205.
11. Clinton M, Haines L, Belloir B, McBride D. Sexing chick embryos:
a rapid and simple protocol. Br Poult Sci. 2001;42(1):134–8.
12. Jensen T, Mace M, Durrant B. Sexing of mid-incubation avian
embryos as a management tool for zoological breeding programs.
Zoo Biol. 2012;31(6):694–704.
13. Porat N, Bogdanov K, Danielli A, Arie A, Samina I, Hadani A.
Direct detection of chicken genomic DNA for gender determination
by thymine-DNA glycosylase. Br Poult Sci. 2011;52(1):58–65.
14. Yilmaz-Dikmen B, Dikmen S. A morphometric method of sexing
white layer eggs. Braz J Poult Sci. 2013;15(3):203–10.
15. Webster B, Hayes W, Pike TW. Avian egg odour encodes informa-
tion on embryo sex, fertility and development. Plos One.
2015;10(1):e0116345.
16. Galli R, Preusse G, Uckermann O, Bartels T, Krautwald-Junghanns
ME, Koch E, et al. In ovo sexing of domestic chicken eggs by
Raman spectroscopy. Anal Chem. 2016;88:8657–63.
17. Wei D, Chen S, Liu Q. Review of fluorescence suppression tech-
niques in Raman spectroscopy. Appl Spectrosc Rev. 2015;50(5):
387–406.
18. Gosnell ME, Anwer AG, Mahbub SB, Menon PS, Inglis DW,
Adhikary PP, et al. Quantitative non-invasive cell characterisation
and discrimination based on multispectral autofluorescence fea-
tures. Sci Rep. 2016;6:23453.
19. Rudbeck L, Dissing J. Rapid, simple alkaline extraction of human
genomic DNA from whole blood, buccal epithelial cells, semen and
forensic stains for PCR. Biotechniques. 1998;25(4):588–90.
20. Don RH, Cox PT, Wainwright BJ, Baker K, Mattick JS.
Touchdown PCR to circumvent spurious priming during gene am-
plification. Nucleic Acids Res. 1991;19(14):4008.
21. Fridolfsson AK, Ellegren H. A simple and universal method for
molecular sexing of non-ratite birds. J Avian Biol. 1999;30(1):
116–21.
1194
22. Schart C. Thesis: Entwiklung einer Steuerung für die Fokussierung
und das Tracking embryionaler Blutgefäße im Hühnerei.
Hochschule für Technik und Wirtschaft Dresden. 2015.
23. Baumann R, Meuer HJ. Blood-oxygen transport in the early avian
embryo. Physiol Rev. 1992;72(4):941–65.
24. Sheng G. Primitive and definitive erythropoiesis in the yolk sac: a
bird’s eye view. Int J Dev Biol. 2010;54(6–7):1033–43.
25. Johnston P. Hematocrit values for the chick embryo at various ages.
Am J Physiol. 1955;180(2):361–2.
26. Campbell GL, Weintraub H, Mayhall BH, Holtzer H. Primitive
erythropoiesis in chick embryogenesis. 2. Correlation between
Hemoglobin Synthesis and the Mitotic History. J Cell Physiol.
1971;50(3):669–81.
27. Berezin MY, Achilefu S. Fluorescence lifetime measurements and
biological imaging. Chem Rev. 2010;110(5):2641–84.
28. Hirsch RE. Heme-protein fluorescence. In: Lakowicz JR, editor.
Topics in fluorescence spectroscopy, Vol. 6 Protein Fluorescence.
New York: Kluwer Academic Publishers; 2002. p. 221–47.
29. Chaiken J, Finney WF, Yang X, Knudson PE, Peterson K,
Weinstock RS, et al. Progress in the noninvasive, in vivo, tissue
modulated Raman spectroscopy of human blood. P SPIE.
2001;4254:216–27.
30. Chaiken J, Goodisman J, Deng B, Bussjager RJ, Shaheen G.
Simultaneous, noninvasive observation of elastic scattering, fluo-
rescence and inelastic scattering as a monitor of blood flow and
hematocrit in human fingertip capillary beds. J Biomed Opt.
2009;14(5):050505.
31. Lee JY, Ji HS, Lee SJ. Micro-PIV measurements of blood flow in
extraembryonic blood vessels of chicken embryos. Physiol Meas.
2007;28(10):1149–62.
32. Tagirov M, Golovan S. Sexual dimorphism in the early embryo-
genesis of the chicken (Gallus Gallus domesticus). Mol Reprod
Dev. 2015;82(5):332–43.
33. Morita VS, Boleli IC, Cargnelutti A. Hematological values and
body, heart and liver weights of male and female broiler embryos
R. Galli et al.
of young and old breeder eggs. Braz J Poultry Sci. 2009;11(1):7–
15.
34. Abdi H, Williams LJ. Principal component analysis. WIREs
Comput Stat. 2010;2(4):433–59.
35. Bonnier F, Byrne HJ. Understanding the molecular information
contained in principal component analysis of vibrational spectra
of biological systems. Analyst. 2012;137(2):322–32.
36. Talari ACS, Movasaghi Z, Rehman S, Rehman I. Raman spectros-
copy of biological tissues. Appl Spectrosc Rev. 2015;50(1):46–111.
37. Bankapur A, Zachariah E, Chidangil S, Valiathan M, Mathur D.
Raman tweezers spectroscopy of live, single red and white blood
cells. Plos One. 2010;5(4):e10427.
38. Cuadros MA, Coltey P, Carmen NM, Martin C. Demonstration of a
phagocytic cell system belonging to the hemopoietic lineage and
originating from the yolk sac in the early avian embryo.
Development. 1992;115(1):157–68.
39. McIntyre BA, Alev C, Tarui H, Jakt LM, Sheng G. Expression
profiling of circulating non-red blood cells in embryonic blood.
BMC Dev Biol. 2008;8:21.
40. Brereton RG, Lloyd GR. Support vector machines for classification
and regression. Analyst. 2010;135(2):230–67.
41. Hamburger V, Hamilton HL. A series of normal stages in the de-
velopment of the chick embryo. J Morphol. 1951;88(1):49–92.
42. Speksnijder G, Ivarie R. A modified method of shell windowing for
producing somatic or germline chimeras in fertilized chicken eggs.
Poult Sci. 2000;79(10):1430–3.
43. Fineman RM, Schoenwolf GC, Huff M, Davis PL. Causes of
windowing-induced dysmorphogenesis (neural tube defects and
early amnion deficit spectrum) in chicken embryos. Am J Med
Genet. 1986;25(3):489–505.
44. Kaleta EF, Redmann T. Approaches to determine the sex prior to
and after incubation of chicken eggs and of day-old chicks. World
Poult Sci J. 2008;64(3):391–9.
45. Aleksandrowicz E, Herr I. Ethical euthanasia and short-term an-
esthesia of the chick embryo. Altex-Altern Tierexp. 2015;32(2):
143–7.