microCT for Systematic Mouse Phenotyping

Frantisek Spoutil, Michaela Prochazkova, Tereza Michalcikova, Ivana Uramova, Sarah Clewell,

Vendula Novosadova and Jan Prochazka

Czech Centre for Phenogenomics, BIOCEV - IMG, Prumyslova 595, Vestec, Czech Republic

Keywords: 3D Imaging, Animal Model, Body Composition, Bone Density, Embryology, Machine Learning,

Morphology, Mus.

Abstract: Phenotyping of mouse mutants is one of the crucial methods for uncovering genetic network at the level of a

whole organism which could help us to understand origin of rare diseases, developmental malformations, but

also the process of mammalian evolution. For studying morphological aspects of either embryos or adults, the

X-ray computed microtomography (microCT) has become a gold standard within the last years. The three-

dimensional (3D) context, availability of data to additional analysis (e.g. volumetric, bone density, or body

composition), and in-vivo approaches in the case of adults are the main advantages when compared to classic

histology and bone morphology. On the other hand, the amount of data is enormous making the data storage

and analysis the bottle-neck of the microCT method. To overcome this obstacle, we cooperate with

bioinformatics experts to set up automation of the process at maximal possible level. Nevertheless, knowledge

and experience of a specialist remain indispensable.

1 INTRODUCTION

The aim to understand and fully annotate mammalian

genome led to establishment of the International

Mouse Phenotype Consortium (IMPC) for the

systematic generation and analysis of all coding

sequence mutations in mice. On the basis of IMPC,

we have built an advanced phenotyping pipeline

using in-vivo microCT scanning technology for adult

mutant mouse cohorts. The primary challenge of

implementing 3D phenotyping of embryos using

microCT technology was to create standardization

across all stages of embryonic development.

However, the advanced, high resolution microCT

scanning allows for detailed morphological analysis

of embryos from the earliest developmental stages up

to perinatal period. The standardization of the whole

process, as established by IMPC, is critical for

relevant comparison and reproducibility of data

between research centers.

The advantages of 3D data generation compared

to more conservative approaches, such as plain X-ray

imaging, for adult skeletons or histological analysis

of embryos are obvious. While the products of

microCT imaging are incredibly detailed, the system

requires significant and time-consuming efforts in

order to process and further analyse the large 3D

datasets. This article will provide an overview of the

standardised morphological phenotyping pipeline, as

well as outline the data analysis in further detail.

Here we present the contemporary, state-of-the-

art high-throughput embryo and bone morphology

pipelines of mouse phenotyping used in our

department with the help of microCT technology, as

well as the obstacles we try to solve to reduce the

disadvantages of the method. We hope this approach

can make the microCT method more accessible to

broader spectrum of researchers, better their results

significantly, and reduce amount of animals used in

the experiments in the future.

2 METHODS

2.1 Embryology

The standardised embryo phenotyping pipeline

contains multiple critical steps for reproducibility and

reliability of data: proper and unified embryo

breeding, embryo harvest and fixation, contrasting of

soft tissues, microCT scanning with metadata

recording, data reconstruction and processing, data

upload to public open database mousephenotype.org.

144

Spoutil, F., Prochazkova, M., Michalcikova, T., Uramova, I., Clewell, S., Novosadova, V. and Prochazka, J.

microCT for Systematic Mouse Phenotyping.

DOI: 10.5220/0007570701440149

In Proceedings of the 12th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2019), pages 144-149

ISBN: 978-989-758-353-7

2.1.1 Embryo Phenotyping Workflow

To determine the effects of embryonic lethal gene

mutations on development, embryos are harvested

from pregnant females a strictly systematic way,

leading to identification and characterisation of lethal

phenotypes. The initial stage for investigation is

E12.5. In case there are no living knockout (KO)

embryos at this stage the E9.5 embryos are analysed.

If KO embryos are absent at E9.5, earlier

developmental stages are then analysed individually.

In case there are living KO embryos at E12.5, the

subsequent developmental stages (E14.5 and E18.5)

are analysed.

After mating, females are visually examined

every morning for a presence of the vaginal plug,

which indicates embryonic day 0.5 (E0.5) of

development.

Gravidity is confirmed by a weight gain or at

earlier stages by ultrasonography. During embryo

harvest on the desired day of embryonic

development, yolk sacks are collected for genotyping

and embryos are fixed with 4% paraformaldehyde

(PFA).

2.1.2 Embryo Contrasting

High-resolution microCT provides an opportunity to

visualize embryos at various stages of development in

3D. Due to weak tissue mineralisation, a contrast

agent must be applied to all specimens.

Smaller samples, e.g. E9.5 embryos, are fixed for

24 hours in 4% PFA and stained with 1% PTA

(phosphotungstic acid) for up to 2 weeks. Larger

samples, e.g. E18.5 embryos, are fixed for 1 week in

4% PFA and stained with Lugol’s Iodine solution for

2 weeks or longer. Lugol’s stock solution (10g KI and

5g I

in 100ml H

O;) is diluted to 25% working

solution in H

O to achieve neutral osmotic pressure

to avoid tissue distortion.

2.1.3 Embryo Scanning and Reconstruction

Stained specimens are removed from the contrast

agent, rinsed with PBS and embedded in 2.5% low-

gelling temperature agarose in tubes. Tubes of

various sizes are used, depending on embryo size,

(single 0.2ml PCR microtubes, 2ml microtubes with

caps or 15ml falcon tubes cut to desired length). All

specimens have to be wrapped in Parafilm to prevent

evaporation.

Depending on the embryo size, SkyScan 1272

high-resolution microCT (Bruker, Belgium) is set up

for voxel size 0.2 - 7µm, and 0.5 or 1 mm Al filter.

360° scan with 0.200° rotation step and 3 frames

averaging setup is used for scanning. Scanning takes

from 5 to 20 hours per one sample, depending on size.

Automated scanning of multiple samples is acquired

by using sample carousel.

InstaRecon CBR Premium software (InstaRecon,

USA) is used for reconstruction. The setup of

reconstruction parameters such as smoothing, ring

artefacts correction, beam hardening and intensities

depends on the embryonic stage.

2.1.4 Embryo Phenotyping

Within the embryonic lethal screen, three knockout

embryos and one littermate, wild-type embryo are

scanned and their phenotypes are evaluated. Gross

morphology (growth retardation, development of

limbs, formation of orofacial area, etc.) is assessed in

whole-mount images and defects of inner organs

(positioning and size of organs, tooth development,

presence of cleft palate, etc.) are examined in the

virtual sections.

2.2 Adult Morphology

For standard morphological phenotyping, a cohort of

28 mice (14 wild types and 14 mutants, composed

from 7 males and 7 females each) at the age of 13

weeks is scanned in-vivo.

2.2.1 Adult Morphology Workflow

Each mouse is anesthetized by Zoletile injection,

arranged in natural position and scanned in SkyScan

1176 in-vivo microCT (Bruker, Belgium). After

scanning, the mouse is weighed and some basic body

measurements are obtained with digital dial calliper.

Two reconstructions are produced from the

primary data: i) for skeletal morphology and bone

mineral density (BMD), and ii) for body composition

analysis.

While the bone morphology is evaluated directly

from its reconstruction files, for body composition

analysis the core body is selected first before entering

the analysis. Volume of interest (VOI) from body

composition analysis is used also for BMD analysis

excluding skull, tail and distal limbs.

The whole process takes about 34 working hours

and 0.314 TB per cohort. See Figure 1 for a detail.

2.2.2 Mouse Scanning and Reconstruction

SkyScan 1176 in-vivo microCT (Bruker, Belgium) is

set up for voxel size 35 µm, voltage of 50 kV, current

of 160 µA, and 0.5 mm aluminium filter with 180°

rotation. When using these parameters, scanning one

microCT for Systematic Mouse Phenotyping

145

Figure 1: Scheme of adult mouse morphology and body composition workflow. Black: data acquisition; Red: data

reconstruction; Orange: data preparation; Green: analysis; Blue: data storage; Grey: Excel output. smooth. = smoothing; ring

art. corr. = ring artifact correction; hard. = beam hardening; def. px. mask = defect pixel masking; CS = border intensities for

reconstruction.

mouse takes about 16 minutes and the mouse takes

about 55.4 mGy/min.

For reconstruction InstaRecon CBR Premium

software (InstaRecon, USA) is used. In the case of

bone morphology, the reconstruction parameters, as

recommended by Bruker microCT (Belgium) are set

up as follows: smoothing = 3, ring artefact correction

= 4, beam hardening = 36%, intensities = 0.0047 –

0.1230. For body composition analysis the values are

changed to: smoothing = 7, ring artefacts correction =

= 5, beam hardening = 10%, defect pixel masking =

5%, intensities = 0.0040 – 0.0200.

2.2.3 Bone Morphology Phenotyping

CTvox software (Bruker microCT, Belgium) is the

basic tool for bone morphology evaluation used. We

record 53 qualitative and numerical variables

describing axial, brachial, and cranial skeleton to

localize effect of the mutation. 8 standardized views

on the whole skeleton or its selected parts (cranium,

limbs) are taken for each sample, as well as details of

malformations if they occur.

2.2.4 Body Composition Analysis

CT analyzer (CTan: Bruker microCT, Belgium) and

Batch Manager (BatMan: Bruker microCT, Belgium)

software are used for VOI selection as well as

analysis itself.

The analysis is based on different density of bones

(and teeth), lean, and fat (with lungs) which is then

distinguishable by X-rays. Primary data includes

absolute and relative volume of all three parts, which

are recalculated to absolute and relative mass

assumptions of lean and fat.

Machine learning-based procedure is developed

and tested at our centre to automatize the VOI

selection.

2.2.5 Bone Mineral Density

The same software is used also for evaluation of bone

mineral density (BMD) and tissue mineral density

(TMD) within the VOI of body composition analysis.

The values are calculated based on results from

calibrated hydroxyapatite phantoms scanned and

reconstructed under the same conditions.

Two secondary variables are computed from the

basic ones: bone mineral content (BMC), and BMC

per body mass.

2.3 Bioinformation Solution

As one of the primary goals in phenotyping is to

implement a high-throughput approach, maximum

automation of the whole procedure with the help of

machine-learning is optimal for our methodological

development, now and in the future. Although we

want to end up with total automation, we start step by

step by solving the most time-demanding problems,

which are i) VOI selection and ii) extraction of

analysis results, for body composition analysis.

Manual selection in CTan was originally used for

VOI definition. The biggest challenge for automation

is how to train the software to delineate what is the

BIOIMAGING 2019 - 6th International Conference on Bioimaging

146

Figure 2: MicroCT scans of wild type murine embryos at different stages of development. A-C - whole-mount views, A'-C' -

virtual sagittal sections.

relevant specimen (i.e. mouse), as the size, shape, and

even orientation of some parts (e.g. limbs) have

significant variability between samples and slices. To

overcome this huge variability, machine-learning

using U-net deep neural network in PyTorch (Paszke

et al., 2017) with principal component analysis based

reorientation in R (R Core Team, 2015) and RStudio

(RStudio Team, 2016) with EBImage (Pau et al.,

2010), pixmaps (Bivand et al., 2015), and magick

(Ooms, 2018) libraries are applied based on previous

manual selection. Although the algorithm is still

under development and results need to be check by a

specialist, it has already increased the efficiency of

VOI selection 15:1 hour per whole mouse cohort (i.e.

28 mice). The data for VOI selection is rotated first

with principal component analysis (PCA), the starting

virtual section image for VOI is selected based on the

shape similarity in the cervical area, and then the

software crops all images section by section, i.e. in

2D to keep core body only. The accuracy of selecting

the 1

virtual slice is very high and differs a few slides

from selection of specialist, especially in animals

with less standard position (e.g. when shoulders are

moved more cranially). We were testing also VOI

selection based on 3D model, but the results were

comparable to 2D model approach, but computation

time was longer. Moreover, 2D approach enables

better correction of results by specialist. While

training the algorithm for automatic selection of the

areas of interest, we used 100 mice (more than 60

thousands virtual sections), which were manually

corrected. The final algorithm has been now used for

more than 1000 mice, and will be retrained to achieve

increasingly higher accuracy.

CTan saves results of every analysis to .CSV table

format with all procedures data acquired. The ideal

final stage of automation would connect the ROI

selection and data extraction with CTan’s macro for

body composition analysis along with its

sophisticated methods for smoothing, noise reduction

and separation.

However, the real challenge is automation of bone

morphology. We want the software to be able to

distinguish and identify individual bones or parts of

the skeleton (e.g. spine, skull, paws), to compare them

with standard shape variability of baseline, and

highlight any differences worthy of attention of a

morphologist. Although some similar approaches

already exist (e.g. Baiker et al., 2010, Wise et al.,

2013), their results are not fully applicable for our

demands and their optimization or finding of original

solution will be needed.

microCT for Systematic Mouse Phenotyping

147

3 RESULTS & DISCUSSION

3.1 Embryo Morphology

Within the embryology lethal screen, we have at the

moment phenotyped 12 lineages. One of the lineages

was lethal before E9.5, four were perinatal lethal and

two were identified as “subviable” lineages. The rest

of strains were lethal between E9.5 and birth. See

Figure 2 for examples of scans and virtual sections of

wild type embryos at different developmental stages.

For embryos older than E9.5, we used the Lugol’s

solution as contrast agent. Noteworthy, dilution of

working solution in water and not PBS turned out to

be methodologically crucial as this approach does not

cause tissue shrinkage.

We have observed various pathologies, such as

growth retardation, short face, and heart and intestinal

dysmorphology in the embryos. We could evaluate

the latter mentioned phenotypes thanks to the high

resolution microCT scanning. It would be very

difficult or even impossible to get this data from

classical histology, especially in cases where we

assessed the length and shape of the inner organs (e.g.

in case of embryonic intestine).

Our next goal in embryo screen is the adoption of

an atlas-based approach of organ recognition shared

among the IMPC centers (eg. Brown et al., 2018),

which will point out even slighter differences in organ

shape and position to a researcher, and quantify the

volume.

3.2 Adult Morphology

In-vivo microCT use in adult mouse morphology

brings numerous advantages compared to standard,

2D X-ray imaging: the level of detail is much higher,

we are able to select appropriate angle and section of

view to show a structure of interest without

compromise from X-ray shielding by surrounding

tissues. In that way we are able to observe structures

like rib rudiments on cervical vertebra, even slight

opening of skull sutures or dorsal arches of vertebrae,

occurrence of baculum in females, or ossification in

tendons of mouse. See Figure 3. This results in higher

probability an effect of mutation will be uncovered

(especially in heterozygotes) and that we will be able

to separate it from general genetic background of the

mouse strain. We have scanned and analysed 27,825

WT mice of both sexes so far, which serves as

baseline of comparison for the relevance of abnormal

morphology findings in KO mutant mice. There were

784 WT mice (2.74 %) with some abnormal

morphology findings. This number is almost two

times lower than in KO mutant mice (all mutations

together), where some abnormality was recorded in

1680 animals from 33,263 (4.81 %).

The greatest advantage of microCT though, is the

possibility to reuse the data of spatially different X-

ray absorption repeatedly. This quality is used also for

body composition analysis, where double-energy X-

ray analysis (DXA) is standard for IMPC. In the case

of DXA the amount of fat and lean is computed from

differences in absorption of X-rays of two energies.

MicroCT brings another quality: spatial distribution

of fat and lean, which is important especially for fat

tissue, as differences there are clinical differences,

whether the fat is stored more subcutaneously or

viscerally. In our department, significant differences

in baseline WT body composition were found in 17

of 46 KO mutant mouse gene cohorts with this

analysis so far.

Figure 3: Examples of skeletal dysmorphology. A) Rib

osteoneogenesis due to injuries in mouse mutant (compare

to B and D). B) Malformation of scapula and ulna in mouse

mutant (compare to A). C) Malformation of the 1

and 2

thoracic vertebra in wild type mouse: 1

one without fusion

of dorsal arch, 2

one without prolongation of spinal

process (compare to D). D) Cervical rib of 7

cervical

vertebra causing malformation of the 1

thoracic rib in

mouse mutant. E) Extra ossification in pelvic region of

mouse mutant (compare to F). F) 6

lumbar vertebra with

morphology of the 1

sacral one in mouse mutant (compare

to E). S1: 1

sacral vertebra; T1: 1

thoracic vertebra;

arrowheads: pointing to mentioned malformation. Figures

not in the same scale.

Magnetic resonance imaging (MRI) can be used

for body composition and even for general skeletal

morphology too. Its results are even much better for

soft tissue morphology. However, the crucial

disadvantage (besides much higher working costs

compared to microCT) is, that it isn’t able to quantify

bone density.

BIOIMAGING 2019 - 6th International Conference on Bioimaging

148

In the case of microCT, its computational and

time requirements are as summarized in Figure 1.

Increasing of computational power and machine

learning programming, which we are working on and

which was summarised elsewhere (Spoutil et al.,

2018), we will be able to push usability of microCT

for standard phenotyping procedure to broaden the

spectrum of usage and users. In the case of body

composition, the next goal is to teach the software to

differentiate hard particles of food from bone and

remove them from sections, plus smooth artificially-

increased intensities in their surroundings causing

star-like artefacts, which can distort real borders of fat

and lean, and thus their estimated volume. In the case

of bone morphology, we are planning to use a 3D

atlas-based approach similar to embryo screen (e.g.

Baiker et al., 2010) able to highlight significant

changes from mean morphology of individual bones,

as well as sections of skeleton.

We have clearly demonstrated that the data

quality of our approach is equal or higher than in the

standard 2D methods used in descriptive morphology

and anatomy of embryos and adults of mice due to

lower tissue deformation, full 3D spatial context, re-

usability of data etc. Replacing the work of specialists

with machine-learning and automation of the

procedure is the way to overcome the biggest

disadvantage of the method time demands. Its

application brought us first significant time savings.

Nevertheless, we still believe, the main role of the

computers in this process should be to help

researchers to focus more on data of their interest,

instead of fully automatic analysis. This is the way we

want to direct our future development of our

procedure.

ACKNOWLEDGEMENTS

This work was supported by RVO 68378050 by the

Academy of Sciences of the Czech Republic,

LM2015040 Czech Centre for Phenogenomics by

MEYS, CZ.02.1.01/0.0/0.0/16_013/0001789

Upgrade of the Czech Centre for Phenogenomics:

developing towards translation research by MEYS

and ERDF, CZ.1.05/2.1.00/19.0395 Higher quality

and capacity for transgenic models by MEYS and

ERDF, and CZ.1.05/1.1.00/02.0109 Biotechnology

and Biomedicine Centre of the Academy of Sciences

and Charles University in Vestec (BIOCEV) by

MEYS and ERDF. We also want thank to Radislav

Sedlacek, director of CCP for his continued support,

and Karla Fejfarova, Frantisek Malinka and Benoit

Piavaux for their expertise in bioinformatics.

REFERENCES

Baiker, M., Milles, J., Dijkstra, J., Henning, T.D., Weber,

A.W., Que, I., Kaijzel, E.L., Löwik, C.W., Reiber, J.H.,

Lelieveldt, B.P., 2010. Atlas-based whole-body

segmentation of mice from low-contrast Micro-CT

data. Med. Image Anal. 14(6). 723-37.

Bivand, R., Leisch, F., Maechler, M., 2015. Bitmap Images

(“Pixel Maps”). https://cran.r-

project.org/web/packages/pixmap.

Brown, J.M., Horner N.R., Lawson, T.N., Fiegel, T.,

Greenaway, S., Morgan, H., Ring, N., Santos, L.,

Sneddon, D., Teboul, L., Vibert, J., Yaikhom, G.,

Westerberg, H., Mallon, A.-M., 2018. A Bioimage

informatics platform for high-throughput embryo

phenotyping. Brief Bioinform. 19(1): 41-51

Bruker microCT, 2013a. Quantifying adipose tissue (fat) in

a mouse or rat by in-vivo microCT. 26 pp.

Bruker microCT, 2013b. Bone mineral density (BMD) and

tissue mineral density (TMD) calibration and

measurement by micro-CT using Bruker-MicroCT CT-

Analyser. 30 pp.

Ooms, J., 2018. Advanced Graphics and Image-Processing

in R. https://cran.r-project.org/web/packages/magick/

Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E.,

DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer,

A, 2017. Automatic differentiation in PyTorch. 31st

Conference on Neural Information Processing Systems

(NIPS 2017), Long Beach, CA, USA. 4 pp.

Pau, G., Fuchs, F., Sklyar, O., Boutros, M., Huber, W.,

2010. EBImage — an R package for image processing

with applications to cellular phenotypes.

Bioinformatics, 26(7), 979–981.

R Core Team, 2015. A language and environment for

statistical computing. R Foundation for Statistical

Computing, Vienna, Austria. http://www.R-

project.org/.

RStudio Team, 2016. RStudio: Integrated Development for

R. RStudio, Inc., Boston, MA URL

http://www.rstudio.com/.

Spoutil, F., Novosadova, V., Fejfarova, K., Malinka, F.,

Piavaux, B., Prochazka, J., 2018. MicroCT for high

throughput phenotyping: perspectives and obstacles.

Micro-CT User Meeting 2018. 187-190.

Wise, L.D., Winkelmann, C.T., Dogbas, B., Baqchi, A.,

2013. Micro-computed tomography imaging and

analysis in developmental biology and toxicology.

Birth Defects Res. C Embryo Today 99(2). 71-82.

microCT for Systematic Mouse Phenotyping

149