Citation
Fully automated software for quantitiative measurements of mitochondrial morphology

Material Information

Title:
Fully automated software for quantitiative measurements of mitochondrial morphology
Creator:
McClatchey, Penn Mason ( author )
Language:
English
Physical Description:
1 electronic file (82 pages) : ;

Subjects

Subjects / Keywords:
Mitochondrial pathology ( lcsh )
Morphology -- Mathematical models ( lcsh )
Mitochondria -- Formation ( lcsh )
Mitochondria -- Formation ( fast )
Mitochondrial pathology ( fast )
Morphology -- Mathematical models ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Abstract:
Mitochondria undergo dynamic changes in morphology in order to adapt to changes in nutrient and oxygen availability, communicate with the nucleus, and modulate intracellular calcium dynamics. Many recent papers have been published assessing mitochondrial morphology endpoints. Although these studies have yielded valuable insights, contemporary assessment of mitochondrial morphology is typically subjective and qualitative, precluding direct comparison of outcomes between different studies and likely missing many subtle effects. In this paper, we describe a novel algorithm for measuring the average length, average width, and spatial density of mitochondria in a fluorescent microscope image. This method was applied to distinguish baseline characteristics of Human Umbilical Vein Endothelial Cells (HUVECs), primary Goto-Kakizaki rat aortic smooth muscle cells (GK vSMCs), primary Wistar rat aortic smooth muscle cells (Wistar vSMCs), and SH-SY5Ys (human neuroblastoma cell line). Consistent with direct observation, our algorithms found SH-SY5Ys to have the greatest mitochondrial density and mitochondrial width, while HUVECs were found to have the longest, thinnest mitochondria. Mitochondrial morphology responses to temperature, nutrient, and oxidative stressors were characterized to test algorithm performance. Large morphology changes recorded by the software agreed with direct observation, and subtle but consistent morphology changes were found that would not otherwise have been detected. Endpoints were consistent between experimental repetitions with high quality microscope images (R=0.96 for length, R=0.77 for width), and maintained reasonable agreement even when the microscope used was intentionally de-tuned to produce grainy, low-resolution images (R=0.70 for length, R=0.84 for width). In addition, the correlation between high and low quality images was found to be similar to the self-agreement of the low quality images (R=0.82 for length, R=0.72 for width), indicating that image quality rather than algorithm function was the limiting factor in reproducibility. Finally, our algorithm was compared to a simple thresholding technique and was found to measure a far greater fraction of the mitochondrial network reliably (p<0.001), primarily by enabling reliable simultaneous detection of perinuclear and peripheral mitochondria. These results indicate that the automated software described herein allows quantitative and objective characterization of mitochondrial morphology from fluorescent microscope images.
Thesis:
Thesis (M.S.) - University of Colorado Denver.
Bibliography:
Includes bibliographic references.
System Details:
System requirements: Adobe Reader.
General Note:
Department of Bioengineering
Statement of Responsibility:
by Penn Mason McClatchey Jr.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
946534840 ( OCLC )
ocn946534840
Classification:
LD1193.E56 2015m M33 ( lcc )

Downloads

This item has the following downloads:


Full Text
FULLY AUTOMATED SOFTWARE FOR QUANTITATIVE MEASUREMENTS
OF MITOCHONDRIAL MORPHOLOGY
by
PENN MASON MCCLATCHEY JR
B.S., Georgia Institute of Technology, 2013
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
Master of Science
Bioengineering
2015


This thesis for the Master of Science degree by
Penn Mason McClatchey, Jr.
has been approved for the
Bioengineering Program
by
Jane E.B. Reusch, Advisor
Richard K.P. Benninger, Chair
Kendall S. Hunter
November 20, 2015


McClatchey, Penn Mason (M.S. Bioengineering)
Fully Automated Software for Quantitative Measurements of Mitochondrial
Morphology
Thesis directed by Professor Jane E.B. Reusch
ABSTRACT
Mitochondria undergo dynamic changes in morphology in order to adapt
to changes in nutrient and oxygen availability, communicate with the nucleus,
and modulate intracellular calcium dynamics. Many recent papers have been
published assessing mitochondrial morphology endpoints. Although these
studies have yielded valuable insights, contemporary assessment of
mitochondrial morphology is typically subjective and qualitative, precluding
direct comparison of outcomes between different studies and likely missing
many subtle effects. In this paper, we describe a novel algorithm for measuring
the average length, average width, and spatial density of mitochondria in a
fluorescent microscope image. This method was applied to distinguish baseline
characteristics of Human Umbilical Vein Endothelial Cells (HUVECs), primary
Goto-Kakizaki rat aortic smooth muscle cells (GK vSMCs), primary Wistar rat


aortic smooth muscle cells (Wistar vSMCs), and SH-SY5Ys (human
neuroblastoma cell line). Consistent with direct observation, our algorithms
found SH-SY5Ys to have the greatest mitochondrial density and mitochondrial
width, while HUVECs were found to have the longest, thinnest mitochondria.
Mitochondrial morphology responses to temperature, nutrient, and oxidative
stressors were characterized to test algorithm performance. Large morphology
changes recorded by the software agreed with direct observation, and subtle
but consistent morphology changes were found that would not otherwise have
been detected. Endpoints were consistent between experimental repetitions
with high quality microscope images (R=0.96for length, R=0.77 for width), and
maintained reasonable agreement even when the microscope used was
intentionally de-tuned to produce grainy, low-resolution images (R=0.70 for
length, R=0.84 for width). In addition, the correlation between high and low
quality images was found to be similar to the self-agreement of the low quality
images (R=0.82 for length, R=0.72 for width), indicating that image quality
rather than algorithm function was the limiting factor in reproducibility. Finally,
our algorithm was compared to a simple thresholding technique and was found
to measure a far greater fraction of the mitochondrial network reliably
(p<0.001), primarily by enabling reliable simultaneous detection of perinuclear
and peripheral mitochondria. These results indicate that the automated
IV


software described herein allows quantitative and objective characterization of
mitochondrial morphology from fluorescent microscope images.
The form and content of this abstract are approved. I recommend its
publication.
Approved: Jane E.B. Reusch
v


ACKNOWLEDGEMENT
I would like to acknowledge each of my thesis committee members for their
contributions to this project and for the mentorship they provided me during my
pursuit of this degree.
In addition, I would like to acknowledge the work done by Amy Keller, Ph.D.,
Leslie Knaub, M.S., Ron Bouchard, M.A., and Chelsea Connon, B.S., which
enabled the investigation documented herein.
VI


TABLE OF CONTENTS
Chapter
1. Introduction to Mitochondrial Morphology................................1
1.1 The Role of Mitochondria in ATP Production..............................1
1.2 The Role of Mitochondria in Intracellular Signaling....................3
1.3 Structure-Function Coupling in the Mitochondrial Network...............4
1.4 Scientific Justification of the Project................................5
2. Methodological Design of Mitochondrial Measurement Software.............9
2.1 Reagents................................................................9
2.2 Cell Culture...........................................................9
2.3 Fixation and Staining.................................................10
2.4 Microscopy and Imaging Techniques.....................................11
2.5 Image Binarization for Computational Measurements.....................12
2.6 Computation of Raw Mitochondrial Morphology Parameters................17
2.7 Correction of Raw Mitochondrial Morphology Parameters.................19
2.8 Post-Processing of Data...............................................22
2.9 Statistics and Software Packages......................................22
3. Internal Validation of Mitochondrial Measurement Software...............24
3.1 Baseline Cell Line Characteristics.....................................24
vii


3.2 Performance of Custom Algorithms and Simple Thresholding.........27
3.3 Consistency of Measures Between Microscopes......................30
3.4 Consistency of Measures Between Experimental Repetitions.........32
4. Cellular Perturbation Measurement Test Cases.......................35
4.1 Effects of Fixation Temperature in HUVEC Cells....................35
4.2 High Glucose Exposure in Primary Rat vSMCs.......................38
4.3 Effects of Hydrogen Peroxide Exposure in SHSY-5Y Neurons.........41
5. Exporting this Technology for Use by other Laboratories............43
5.1 Software Testing with Independently Produced Images...............43
5.2 Software Testing with Uncontrolled Sample Images..................46
6. Conclusions and Future Directions...................................48
6.1 Discussion of Results..............................................48
6.2 Likely Barriers to Implementation.................................54
6.3 Potential future applications.....................................56
Literature Cited.......................................................58
viii


LIST OF FIGURES
Figure
1. Outline of Cellular Energetics and Mitochondrial Respiration..........2
2. Method Used to Binarize Raw Microscope Images.........................14
3. Calibration Using Fluorescent Beads of Known Size.....................21
4. Baseline Cell Line Characteristics....................................26
5. Performance of Custom Algorithm Relative to Thresholding..............29
6. Consistency of Measures Using Different Microscopes...................32
7. Consistency of Experimental Repetitions...............................34
8. Effects of Temperature During Fixation on Mitochondrial Morphology....37
9. Mitochondrial Network Response to Incubation in High Glucose..........40
10. Effects of Flydrogen Peroxide on Mitochondrial Morphology............42
11. Effects of vSMC Conditioned Media on Mitochondrial Morphology........45
12. Algorithm Performance with an Uncontrolled Image Sample..............47
IX


LIST OF TABLES
Table
1. Microscope Settings Used in This Study................................12
2. Baseline Cell Line Characteristics....................................26
3. Effects of Temperature During Fixation on Mitochondrial Morphology....38
4. Effects of Hydrogen Peroxide on Mitochondrial Morphology..............44
5. Effects of vSMC Conditioned Media on Mitochondrial Morphology.........48
x


1. Introduction to Mitochondrial Morphology
1.1 The Role of Mitochondria in ATP Production
The primary role of the mitochondrial network is to produce ATP for the
cell. Raw macronutrients such as glucose and fatty acids undergo glycolysis
and beta oxidation, respectively, to prepare them for the citric acid cycle (TCA).
The TCA serves to reduce NAD+ to NADH, which acts as an electron donor in
the electron transport chain (ETC), where a majority of intracellular ATP
production occurs. An outline of these pathways is shown in Figure 1.
Eukaryotic cells derive some energy from mitochondrial respiration (TCA and
ETC), which generates 36 ATP molecules per glucose molecule and 131 ATP
molecules per fat molecule, and some from glycolysis, which generates 2 ATP
molecules per glucose molecule. Accordingly, when oxygen is not rate limiting
to mitochondrial respiration (as is typically the case), the TCA is the dominant
source of energy for cellular processes. Highly metabolically active tissues
such as the brain or skeletal muscle generally have high mitochondrial content
to accommodate their unusually high energetic needs.
1


Cytoplasm
Mitochondria
FIGURE 1: OUTLINE OF CELLULAR ENERGETICS AND MITOCHONDRIAL
RESPIRATION. Glycolysis and beta oxidation prepare glucose and fatty acids
for mitochondrial respiration, respectively. Pyruvate produced through
glycolysis is oxidized to acetyl CoA before entry into the citric acid cycle (TCA).
Glycolysis, beta oxidation, and the TCA are all involved in reduction of NAD+
and FAD to NADH and FADH2, respectively, which serve as electron donors in
the electron transport chain (ETC), where most ATP production occurs. Note
that many intermediates are neglected in this figure, and the consumption of
oxygen required for these processes is not illustrated here.
2


1.2 The Role of Mitochondria in Intracellular Signaling
In addition to providing most of the ATP consumed in a typical
Eukaryotic cell, mitochondria also serve as multi-modal hubs of intracellular
signaling. The mitochondrial network helps to buffer cytoplasmic calcium
oscillations [1, 2] and regulate increases in cytosolic calcium content during
apoptosis [3], Since calcium dynamics are critically involved in cellular
processes ranging from muscle contraction [4, 5] to immune cell activation [6],
the role of mitochondria in regulating calcium dynamics is likely critical for
normal cellular function. The role of mitochondria in regulation of cytosolic
calcium signaling has been extensively reviewed elsewhere [7],
Likewise, because the ETC is the primary site of oxidation/reduction
reactions within the cell, mitochondria are also a major source of free radicals.
Although mitochondrial free radical production can play a pathological role [8],
mitochondrial free radicals are also involved in normal cellular signaling
processes, typically in the context of a cellular stress response [9], The role of
mitochondrial superoxide (the primary free radical produced by mitochondria)
in cellular signaling has been extensively reviewed elsewhere [10], An
emerging area in mitochondrial signaling is possible communication between
mitochondria and the nucleus. It has been shown that certain stimuli can cause
perinuclear gathering of mitochondria [1_1_, 12] and others can result in
accumulation of mitochondrial proteins in the nucleus [13], but the meaning of
3


these findings is not yet clear. A detailed review of mitochondrial involvement
in intracellular signaling processes is beyond the scope of this introduction; for
purposes of this manuscript it is sufficient to recognize that visible changes in
the structure of the mitochondrial network may indicate changes in intracellular
signaling processes as well as changes in cellular energetics.
1.3 Structure-Function Coupling in the Mitochondrial Network
In order to perform its many functions in cellular signaling and energy
homeostasis, the mitochondrial network must be highly dynamic. One
important manifestation of this dynamism lies in dynamic changes in network
morphology during certain physiologic responses. Mitochondria operate more
efficiently when networked [14], and this increase in efficiency is thought to be
mediated by allowing each individual organelles flaws to be diluted in the larger
network. Mitochondrial fission allows separation of damaged and intact
mitochondrial fragments for subsequent autophagy or re-integration into the
mitochondrial network, respectively [15], In addition to preserving the quality
and efficiency of the mitochondrial network, changes in mitochondrial
morphology also modulate mitochondrial calcium signaling [7] and redox
signaling [16], Mitochondrial dynamics are altered during cellular activation
and/or proliferation [17], and nutrient stressors tend to induce dynamic
mitochondrial responses in cells such as pancreatic beta cells that work as
4


metabolic sensors [18], Consistent with a role for mitochondrial dynamics in
these crucial cellular processes, mutations in genes regulating mitochondrial
morphology result in disease and many disease states involve changes in
mitochondrial morphology [191.
1.4 Scientific Justification of the Project
The study of mitochondrial morphology has become a topic of intensive
investigation in recent years. Whereas a PubMed search for mitochondrial
morphology returns only 222 results from the year 2000 or before, the same
search returned 1331 results as of 4/14/2015, and many of these papers are
published in high profile journals. This vast body of research is unsurprising in
light of recent reports indicating that mitochondrial dynamics are involved in
regulating cellular respiration [14], calcium dynamics [7], free radical production
[161, substrate catabolism [48], cell survival [20], cell activation [21], and
perhaps epigenetics [22]. Alterations in mitochondrial dynamics have been
reported in a wide variety of disease states including diabetes [23], obesity [241,
aging [25], vascular proliferative disorders [26], and several genetic conditions
that directly influence regulation of mitochondrial morphology [19], Clearly,
mitochondrial dynamics are critically involved in human physiology and their
study is thus warranted.
5


A great deal has been learned about mitochondrial morphology and
dynamics in recent years, but the current state of the art for assessing
mitochondrial morphology leaves much to be desired. Of the first 50 results on
a PubMed search for mitochondrial morphology sorted by relevance (as of
4/14/2015), 31 are primary research articles assessing mitochondrial
networking. Of these, 3 used molecular or histologic markers as surrogates for
mitochondrial morphology measurements, 18 reported subjective assessments
of morphology, 4 performed morphological measurements by hand, and 6 used
some manner of software measurement. It is troubling that most common
methods for assessing mitochondrial morphology involve no objective
measurements of mitochondrial morphology. Subjective assessment is ill-
suited to comparison of findings between studies or between cell types, and
molecular markers do not necessarily relate to actual morphology in an intuitive
manner. Because these methods are relatively insensitive, the interventions
employed to perturb mitochondrial morphology are often extreme relative to
what a cell might encounter in-vivo. Taking measurements by hand avoids
these pitfalls, but it is very slow, thus limiting sample sizes and increasing time
per experimental iteration. A quantitative, objective, and consistent software
method of measuring mitochondrial morphology is attractive in that it allows
comparison between studies or cell types, might detect more subtle changes
6


in morphology than subjective assessment, and allows for efficient analysis of
a large sample size.
Some investigators have begun to use software tools, such as the
Mitochondrial Morphology plugin in ImageJ [27, 28, 29]. This software allows
introduction of investigator bias since the user selects the thresholding
parameters, and the endpoints reported to the user are rather limited. In
addition, our results indicate that thresholding allows reliable measurement of
only a portion of the mitochondrial network. The Fourier-based algorithm of
Koopman et al reliably detects mitochondria [30], but implementation of this
algorithm is technically sophisticated, likely accounting for the limited use of
this method in the nine years since its publication. We propose that a practical
software solution for the measurement of mitochondrial morphology should
meet six basic criteria: algorithm performance should not require a specific
microscope, algorithm expertise should not be required to operate the software,
quantitative outcomes should be consistent between different microscopes
when applied to the same samples, endpoints should be consistent between
experimental repetitions, no specific portion of the mitochondrial network
should be excluded from analysis or incorrectly analyzed due to algorithm
performance, and the endpoints reported to the user should hold intuitive and
tangible meanings. The final point may seem soft on first glance, but we
7


contend that if the output does not align with visual impressions, it will be
difficult to trust the results.
These six criteria were used to guide our experimental design.
Comparison of multiple cell types on two different microscopes yielded
consistent measurements within each cell type and between microscopes and
definitively detected cell type differences in network morphology. Automation
of the analysis precludes the need for software expertise on the part of the end
user, and the methods developed for removing aberrantly detected
mitochondria successfully removed mitochondrial bodies which were obviously
detected incorrectly. A variety of experimental perturbations were used to test
the ability of this software to track dynamic changes in network morphology,
and dynamic changes were in fact detected in with each perturbation.
8


2. Methodological Design of Mitochondrial Measurement Software
2.1 Reagents
Dulbecco's Modified Eagles Medium (DMEM) 5 mM and 25 mM glucose
and non-essential amino acids and Laughlins F12 Medium were obtained from
Thermo Scientific Hyclone, and trypsin and trypsin inhibitor were purchased
from Fisher Scientific. Fetal bovine serum (FBS) was procured from Gemini
Bioproducts. Flank's Balanced Salt Solution (FIBSS) was purchased from
Corning Life Sciences. Secondary detection antibodies Alex Fluor 488 and 546
were purchased from Life Technologies. Antibodies to TOM20 (rabbit) and
nitrotyrosine (mouse) were procured from Santa Cruz Biotechnology.
2.2 Cell Culture
Primary rat vascular smooth muscle cells (vSMCs) were cultured in low
glucose (5 mM) DMEM with 10% fetal bovine serum (FBS), 1% L-glutamine,
1% non-essential amino acid blend, and 1% Pen/Strep, all expressed as % by
volume. For vSMC starvation media, the same proportions were maintained,
except that FBS was reduced to 0.1%. Human umbilical vein endothelial cells
(HUVECs) were cultured in F-12k medium (Hyclone #SH30526.01) with 10%
9


FBS, 1% Pen/Strep, 0.05 mg/mL endothelial cell growth supplement, and 0.1
mg/mL heparin. SH-SY5Ys (ATTC neuronal cell line) were cultured in 45% F-
12k medium, 45% low glucose DMEM, and 10% FBS. All cells were cultured
at 37 degrees Celsius at 5% CO2. Media was changed at least every three
days, and cells were split 2:1 at 80% confluence. Experiments were performed
on cells at 50%-70% confluence. Unless otherwise specified, cell culture
materials were obtained from Santa Cruz Biotechnology.
2.3 Fixation and Staining
Prior to fixation, phosphate buffer solution (PBS) rinse and
paraformaldehyde solutions were warmed to 37 degrees Celsius (except as
otherwise indicated). Samples were washed 3x in warm PBS prior to incubation
in 4% paraformaldehyde at 37 degrees Celsius for 15 minutes. Following
fixation, samples were washed 3x in warm PBS and quenched in 50 mM NFMCI
for 15 minutes at room temperature. Samples were then rinsed 3x with room
temperature PBS and stored immersed in PBS at 4 degrees Celsius until
staining. All samples were stained within one week of fixation. Prior to staining,
cells were permeablized by incubation in 0.1% TX-1000 at 37 degrees Celsius
for 15 minutes. Following permeabilization, samples were rinsed 3x with room
temperature PBS and then blocked by incubation in 10% fetal bovine serum
(FBS) in PBS for 25 minutes and 37 degrees Celsius. Samples were
10


subsequently incubated in a primary antibody mix in 5% FBS in PBS at 37
degrees Celsius for one hour. Following treatment with primary antibody,
samples were washed 3x with 5% FBS in PBS and incubated in a secondary
antibody mix in 5% FBS in PBS at room temperature for one hour. Samples
were then washed 3x in PBS, incubated in DAPI diluted 1:10,000 in PBS at
room temperature for 10 minutes, washed 3x in PBS again, then washed once
in distilled water. Coverslips were then mounted to microscope slides for
imaging. The primary antibody mix consisted of a Tom20 antibody (Santa Cruz
Biotechnology sc-11415) diluted 400x to stain mitochondria and a nitrotyrosine
antibody (Santa Cruz Biotechnology sc-32757) diluted 100x to stain the
cytoplasm. The secondary antibody mix consisted of Alexafluor 488 (Life
Technologies A-11008) diluted 1000x to react with the Tom20 antibody and
Alexafluor 546 (Life Technologies A-11030) diluted 500x to react with the
nitrotyrosine antibody. Following staining, slides were stored protected from
light at 4 degrees Celsius.
2.4 Microscopy and Imaging Techniques
The primary imaging system used in these experiments was an Olympus
FV1000 confocal microscope provided by the UC Denver Advanced Light
Microscopy Core. A 488 nm laser was used to excite AlexaFluor 488 and a 543
nm laser was used to excite AlexaFluor 546. Built-in filter settings for these
11


fluorophores were used to detect fluorescence. Slides were imaged using a
60x oil immersion objective at a resolution of 206.8 nm/pixel and an image size
of 1024 pixels by 1024 pixels. An exposure time of 15 ms per pixel was used
and the average of three frame and line acquisitions each was used as the raw
microscope image.
In addition to the Olympus FV1000, further imaging was performed on a
Leica SP8 confocal microscope at the Denver VA Medical Center. A white light
laser was tuned to excitation wavelengths of 488 nm and 546 nm for AlexaFluor
488 and 546, respectively. Similar to images acquired on the Olympus system,
a 60x oil immersion objective was used on the Leica system. In order to test
the effects of poor image quality on algorithm performance, images acquired
on the Leica SP8 system used no averaging to reduce noise, an exposure time
of 4.9 us per pixel was employed, and images were acquired at a resolution of
360.4 nm/pixel and an image size of 512 pixels by 512 pixels. Microscope
settings for both sets of imaging experiments are included in Table 1 below.
TABLE 1: MICROSCOPE SETTINGS USED IN THIS STUDY
Olympus FV1000 Leica SP8
Image Size (pixels) 1024x1024 512x512
Pixel Size (nm) 206.8 360.4
Exposure Time (us) 15 4.9
Averaging 3 frames none
Numerical Aperture 1.4 1.4
Special Considerations none intentionally poor image quality
12


2.5 Image Binarization for Computational Measurements
Images were imported into Matlab in 16-bit TIF format and binarized
using custom-built software. The methods used to binarize the Tom20
(mitochondria) and nitrotyrosine (cytoplasm) channels are outlined in Figure 2.
4x4 binning was applied to raw mitochondrial images to reduce the influence
of noise. The noise-reduced images were then normalized so that the minimum
pixel intensity value was 0 and the maximum pixel intensity value was 1. Noting
that the spaces between mitochondria in areas of high mitochondrial density
were often brighter than were mitochondria themselves in areas of lower
mitochondrial density, a Gaussian blur was applied to noise-reduced
mitochondrial images to obtain an approximation of background intensity.
Equation 2.1 was then used to compare noise-reduced and background
intensity images to identify regions of locally increased fluorescence intensity.
In this equation, M refers to the noise-reduced image, S refers to the
background intensity image, and P refers to the resulting intensity prominence
image.
P =
(.M-S)*M
5+1
[2.1]
13


/\ J Raw Image
Noise-Reduced (M)
Blurry (S)
P = S Ss
FIGURE 2: METHOD USED TO BINARIZE RAW MICROSCOPE IMAGES. (A)
Binarization of mitochondrion channel. Raw microscope images are imported
into Matlab in 16-bit TIF format. 4x4 binning is used to reduce the influence of
machine noise. The noise-reduced image is then compared to a Gaussian blur
14


of the same image to determine intensity prominence. A threshold is then
applied to the intensity prominence to detect mitochondria. (B) Binarization of
cytoplasm channel. Cytoplasm is stained using a monoclonal antibody for
nitrotyrosine. The raw microscope image is then blurred twice using a Gaussian
blur, and the difference between the once-blurred and twice-blurred images
thresholded to detect the cells cytoplasm.
The intensity prominence image obtained using Equation 1 was then
processed using the thresholding algorithm outlined in Equations 2.2-2.7 to
maximize agreement between intensity prominence and the resulting binary
image. Threshold values between 0 and 1 were tested with a resolution of 0.02
and the image was then binarized at the threshold value with the lowest error
score such that values above threshold equal 1 and those below threshold
equal 0. In these equations, Tis the binary image obtained at a given threshold
value, pixel refers to a pixel during single-pixel operations, thresh refers to the
threshold value being tested, Err1 is the component of the error score indicating
background pixels detected as mitochondria, Err2 is the component of the error
score indicating mitochondrial pixels detected as background, and Error is the
combined error score.
1 if P(pixel) > thresh
0 else
[2.2]
Err 1 = T P
[2.3]
15


Err 2 = P T
[2.4]
Errl(Errl < 0) = 0
[2.5]
Err2(Err2 < 0) = 0
[2.6]
Error =
sum(l-T) sum(J)
[2.7]
The nitrotyrosine (cytoplasm) channel initially appears as punctate dots
of fluorescence throughout the cytoplasm. The process for binarizing this
channel is outlined in Figure 1B. Prior to processing, images were normalized
such that the minimum pixel intensity was 0 and the maximum pixel intensity
was 1. A Gaussian blur was used to transform the punctate distribution of
fluorescence into a more uniform distribution of fluorescence within the
cytoplasm. It was observed that more peripheral regions of the cytoplasm are
often more similar in pixel intensity to background than to pixel intensity in
perinuclear regions of the cytoplasm. A second Gaussian blur was therefore
used to obtain an approximation of background pixel intensity, similar to the
method employed in binarization of the mitochondrial channel. The difference
16


in pixel intensities between the once-smoothed and twice-smoothed images
was then used to construct an intensity prominence image. Threshold values
between 0 and 1 were tested with a resolution of 0.02. The threshold value
which minimized the ratio of edge length to background area was then selected
to binarize the image. Finally, the DAPI (nucleus) channel was binarized
directly by thresholding using the same algorithm as was used on the intensity
prominence image from the mitochondrial channel.
2.6 Computation of Raw Mitochondrial Morphology Parameters
Raw mitochondrial and cytoplasm images were binarized as described
in the previous sections. The number of illuminated pixels in each binary
mitochondrial image was then multiplied by the area of a single pixel in pm2 to
obtain a metric of mitochondrial network size. The number of illuminated pixels
bordering non-illuminated pixels was multiplied by the pixel edge length in pm
to obtain a metric of total mitochondrial network perimeter. The number of
mitochondrial bodies was detected using a built-in function in Matlab (see
bwconncompO in Matlab documentation). This function uses a 4-connected
pixel neighborhood to assess connectivity and provides outputs including the
total number of bodies and a list of pixels included in each body. Bodies
comprising less than 15 urn2 were excluded from further analysis as too small,
and those for which the projected area to perimeter ratio was greater than 0.3
17


um were excluded from further analysis as too wide. These criteria were
determined empirically to remove bodies which were obviously detected
aberrantly and were used to prevent error in algorithm measurements due to
faulty image binarization.
Using mitochondrial bodies that were not rejected according to these
criteria, average mitochondrion size was then taken to be the total
mitochondrial network size divided by the number of mitochondrial bodies. The
average mitochondrion perimeter was likewise taken to be the total
mitochondrial network perimeter divided by the number of mitochondrial
bodies. Next, average mitochondrion size and perimeter were used to calculate
average mitochondrion length and width. Observing that most of the
mitochondria in the network are shaped like long circular tubes with rounded
ends, the geometric relationships depicted in Figure 2A and Equations 2.8-2.9
below were used to solve for average length and width. In these equations, P
is perimeter, A is area, W is width, and L is length.
Pr
mean
= n Wr
mean
+ 2 L
'mean
[2.8]
[2.9]
18


2.7 Correction of Raw Mitochondrial Morphology Parameters
The raw parameters computed using the algorithms described in the
previous section were subsequently corrected for total system error in
measurement and for more general field-of-view parameters, such as total
cytoplasm area and number of cells. Total system error comprises the
microscope point spread function (PSF), algorithm error in binarization of raw
mitochondrial images, and discretization error in computational measurement
of these binarized mitochondrial images. Rather than attempting to measure or
estimate each of these sources of error independently, the magnitude of the
total system error was directly measured as the difference between the known
400 nm diameter of fluorescent beads (CytoDiagnostics FBC-620-400) and
their measured diameter using the same image processing scripts as were
used to measure mitochondrial network parameters. For both microscopes
used, 400 nm beads were measured at concentrations spanning four orders of
magnitude. Bead images were binarized using identical methods to those used
to binarize mitochondrial images, and measured bead diameter was inferred
from total projected area and perimeter using similar methods to those outlined
in Equations 8-9, except that a spherical rather than tubular shape was
assumed. No significant differences were found between calculated bead
diameters when measured at different concentrations for either microscope
system, indicating that clustering of beads did not compromise this analysis
19


(Figure 3C). It was found that the high resolution setup on the Olympus FV1000
microscope overestimated bead diameter by 170 nm, while the intentionally de-
tuned setup on the Leica SP8 overestimated bead diameter by 360 nm. These
error values are similar to or smaller than pixel size of 206.8 nm and 360.4 nm
using the Olympus and Leica microscopes, respectively. The bulk
measurement error values inferred from bead calibration thus represent
statistical averages (i.e. either 0 or 1 superfluous pixel detected along each axis
and the reported measurement error represents probability of a single
superfluous pixel along each axis) and are applicable to whole-network but not
single-mitochondrion measurements. Raw length and width values were then
corrected by subtracting the bead diameter overestimation obtained using the
imaging system and setup corresponding to each mitochondrial image.
Total cytoplasm area was estimated from the total number of illumined
pixels in the binarized cytoplasm image multiplied by pixel area in pm2 The
total number of cells in the image was estimated from the number of bodies in
the binarized nucleus channel using the same built-in Matlab function as was
used to count mitochondrial bodies. The total projected area of the
mitochondrial network was divided by the total cytoplasm area to obtain an
estimate of the fraction of the cytoplasm filled by mitochondria as a normalized
metric of mitochondrial mass. Likewise, the total projected area of the
mitochondrial network and the total number of mitochondria were divided by
20


the number of cells in the image to obtain an estimate of total mitochondrial
mass per cell and total number of mitochondrial bodies per cell, respectively.
For most inter-group comparisons, corrected parameters were used rather than
raw parameters in order to reduce the influence of machine error, random
differences in cell size or confluence between fields of view, etc. on reported
end-points.
21


Solve for W,eau
and L
mean
^msosured ^ *
^true ~ ^measured 1
FIGURE 3 CALIBRATION USING FLUORESCENT BEADS OF KNOWN SIZE.
(A) Deterministic relationship between edge length, cross-sectional area, and
length and width. Assuming the typical mitochondrion to be shaped like an
elongated tube, the average length and width in the network can be calculated
from the total network area, edge length, and number of mitochondria. (B)
Representative images of fluorescent beads at 1000x and 100x minimum
concentration. (C) Measuring the diameter of fluorescent beads. Using the
same technique as previously described but modified to assume spherical,
rather than tubular bodies. At bead concentrations spanning four orders of
magnitude, calculated bead diameter remained constant at 570.1 nm using the
Olympus microscope and 760 nm using the Leica microscope, compared to a
true mean diameter of 400 nm. The total system bias of 170.1 nm and 360 nm
for the Olympus and Leica images, respectively, was then subtracted from
average mitochondrion length and width values to correct for the microscope
PSF and errors in image processing.
22


2.8 Post-Processing of Data
Following initial measurements and corrections, outliers in the data set
were automatically detected and removed by the software according to the
following scheme: any image for which any parameter measured was greater
than three standard deviations from the treatment group mean was removed
from the data set entirely. Under this scheme, images with unusually sparse
mitochondria would not be analyzed for mitochondrial length, images with an
unusual number of cells would not be analyzed for mitochondrial density, etc.
Remaining images were then used to calculate treatment group means and
SEMs for each experimental repetition and across all experimental repetitions.
In order to verify that this mode of data grooming was not limiting sample size,
the total number of images included and rejected in each analysis was
monitored. Less than 10% of images were removed from analysis in all image
sets analyzed.
2.9 Statistics and Software Packages
For comparisons involving three or more groups, a one-way ANOVA
was used to assess statistical significance. For comparisons involving two
groups only, an unpaired t-test was used to assess statistical significance. To
assess correlations and their statistical significance, Pearsons R and the
corresponding t-value were computed. Effects for which p<0.1 were considered
23


trending and those for which p<0.05 were considered significant. For
comparisons in which one-way ANOVA was employed, Tukeys post-test was
used to correct for the effects of multiple comparisons. All post-hoc statistical
analyses were performed in Prism GraphPad version 5.0. The mitochondrial
measurement algorithms and automated post-processing were coded into
Matlab 2013. Where available, built-in functions in these software packages
were used for statistical computations.
24


3. Internal Validation of Mitochondrial Measurement Software
3.1 Baseline Cell Line Characteristics
This section documents basic mitochondrial network characteristics of
GK vSMCs, Wistar vSMCs, HUVECs, and SH-SY5Ys with no special treatment
prior to fixation. Figure 4 summarizes differences between these four cell lines.
Consistent with visual inspection, HUVECs were found to have the longest,
thinnest, and least densely packed mitochondria, SH-SY5Ys the shortest and
most densely packed mitochondria, and Wistar vSMCs the widest
mitochondria. Values obtained for average (+/-SD) mitochondrion length were
11.35+/-2.41 urn, 6.80+/-0.86 urn, 5.09+/-0.77 urn, and 4.31+/-0.66 urn for
HUVECs, GK vSMCs, Wistar vSMCs, and SH-SY5Ys, respectively. Values
obtained for average mitochondrion width were 272+/-24 nm, 315+/-20 nm,
351+/-23 nm, and 320+/-14 nm for HUVECs, GK vSMCs, Wistar vSMCs, and
SH-SY5Ys, respectively. Values obtained for fraction of cytoplasm filled by
mitochondria were 9.6+/-2.0%, 16.0+/-1.8%, 15.8+/-1.7%, and 23.9+/-1.8% for
HUVECs, GK vSMCs, Wistar vSMCs, and SH-SY5Ys, respectively. In addition
to these characteristics of the mitochondrial networks, the fraction of each
mitochondrial network detected correctly was estimated by eliminating
detected
25


mitochondrial bodies that were excessively small or excessively wide. The
fraction of the detected mitochondrial network included in morphological
analysis was 85.7+/-6.8%, 85.1+/-5.7%, 64.6+/-15.7%, and 17.1+/-6.8% for
HUVECs, Wistar vSMCs, GK vSMCs, and SH-SY5Ys, respectively. These
values are also listed in Table 2 below.
TABLE 2: BASELINE CELL LINE CHARACTERISTICS
HUVEC GK vSMCs Wistar vSMCs SH-SY5Y
Mean Mitochondrion Length (mean+/-SD) 11.35+/-2.41 um 6.80+/-0.86 5.09+/-0.77 um 4.31+/-0.66 um
Mean Mitochondrion Width (mean+/-SD) 272+/- 24 nm 315+/-20 nm 351+/-23 nm 320+/-14 nm
Fraction of Cytoplasm Filled by Mitochondria 9.6+/-2.0% 16.0+/-1.8% 15.8+/-1.7% 23.9+/-1.8%
Fraction of Mitochondrial Mass Included in Analysis 85.7+/-6.8% 85.1+7-5.7% 64.6+7-15.7% 17.1+7-6.8%
26


FIGURE 4: BASELINE CELL LINE CHARACTERISTICS (A) Representative
microscope images of cell types used in this study. Mitochondria were stained
using a Tom20 antibody, cytoplasm using a nitrotyrosine antibody, and nuclei
using DAPI. (B) Average mitochondrion length by cell type. HUVECs had the
longest mitochondria at 11.5 urn while SH-SY5Ys had the shortest at 4.31 urn.
Each cell type displayed a highly significantly different (p<0.0001) average
mitochondrion length from every other cell type. (C) Average mitochondrion
width by cell type. Wistar vSMCs had the widest mitochondria at 351 nm while
HUVECs had the thinnest at 212 nm. All cell type differences were highly
significant (p<0.0001), except for GK vSMCs and SH-SY5Ys, which did not
differ significantly (p>0.05). (D) Mitochondrial mass by cell type. SH-SY5Ys had
the greatest fraction of the cytoplasm filled by mitochondria at 23.9% while
HUVECs had the least at 9.6%. All cell type differences were highly significant
(p<0.0001), except for GK vSMCs and Wistar vSMCs, which did not differ
significantly (p>0.05). (E) Fraction of mitochondrial network correctly detected
by cell type. HUVECs and GK vSMCs had the best coverage at 85.7%> and
85.1%o, respectively while SH-SY5Ys had the worst at 17.1 %>. All cell type
differences were highly significant (p<0.0001), except for Wistars and GK
vSMCs, which did not differ significantly (p>0.05).
27


3.2 Performance of Custom Algorithms and Simple Thresholding
This section documents the differences between results obtained using
the novel algorithms described in the methods section and results obtained
using simple thresholding. HUVEC and SH-SY5Y images were used in this
analysis in order to maximize the range of values obtained.
Figure 5 summarizes differences between results obtained from the
algorithm developed in this manuscript and application of simple thresholding.
Our scheme for simple thresholding involved applying the algorithm described
in the methods section directly to raw mitochondrial images rather than the pre-
processed algorithm-defined intensity prominence images (Figure 2A). Simple
thresholding is comparable to predicate methods using simple thresholding
such as the Mitochondrial Morphology plugin in ImageJ [27, 28, 29], except that
the threshold is automatically determined rather than user defined. A greater
dynamic range of values for mitochondrial length was recorded using the
custom algorithm than using simple thresholding (algorithm range: 3.12 um -
20.4 um, threshold range 2.47 um 9.14 um, significant reduction in variance
using simple thresholding, p<0.0001 using an F-test) and measured
mitochondrial lengths were significantly smaller when measured using simple
thresholding (p<0.0001 using unpaired T-test with Welchs correction for both
FlUVECs and SH-SY5Ys). In addition, the fraction of the mitochondrial network
detected adequately for inclusion in subsequent analyses was significantly
28


reduced with simple thresholding (p<0.0001 for both cell types). Noting that the
portion of the network automatically discarded from further analysis using
simple thresholding was primarily perinuclear, the measurements obtained
using simple thresholding were compared to measurements of the outermost
10% of mitochondrial mass using the custom algorithm. Under this comparison,
no differences in the mean or variance (p>0.05 for both) were found between
algorithm and thresholding groups for either cell type, consistent with a bias
towards shorter, less variable mitochondria due to exclusion of perinuclear
mitochondria from subsequent analysis.
29


FIGURE 5: PERFORMANCE OF CUSTOM ALGORITHM RELATIVE TO
THRESHOLDING. (A) Mean mitochondrion length as measured using the
custom algorithm vs simple thresholding. Dynamic range of lengths measured
is greater using the custom algorithm, and lengths measured using
thresholding are routinely less than those measured using the custom
algorithm. (B) Mean mitochondrion length in the outermost 10% of the
mitochondrial network as measured using the custom algorithm vs mean
mitochondrion length throughout the entire network as measured using
thresholding. Although the agreement is imprecise, dynamic range is similar
between the two metrics and no bias is observed. (C) Examples of images
binarized using the custom algorithm and simple thresholding. Portions of the
mitochondrial network that were not detected correctly are colored purple. A
much greater portion of the network is excluded from analysis using simple
thresholding and the portions of the network excluded are largely perinuclear.
(D) Fraction of the mitochondrial network detected correctly by cell type and
algorithm. Regardless of detection quality using the custom algorithm,
detection quality using simple thresholding is substantially lower (p<0.0001 for
both HUVECs and SH-SY5Ys)
30


3.3 Consistency of Measures Between Different Microscopes
This section documents the similarities and differences between results
obtained on different microscopes. HUVEC and SH-SY5Y samples were
imaged on both an Olympus FV-1000 and a Leica SP8. Images were obtained
at 1024x1024 pixels on the Olympus microscope and at 512x512 pixels on the
Leica microscope. In addition, exposure time was limited and lasers
intentionally de-tuned on the Leica microscope in order to obtain grainy, low-
resolution images for comparison with the clear, high-resolution images taken
on the Olympus microscope. Mean values for each experimental repetition
were compared between the microscope setups.
Figure 6 summarizes comparisons of mitochondrial network parameters
between the same samples measured on different microscopes.
Measurements of average mitochondrion length and average mitochondrion
width were significantly correlated between the two imaging systems (R=0.82
and R=0.73, length and width respectively, p<0.0001 for both). A bias towards
measurement of longer mitochondria was observed using the low-resolution
system, with mitochondrion length averaging 3.96 urn (~40%) longer when
measured using the low-resolution system (p<0.0001 using paired t-test),
compared to a ~100 nm pixel resolution difference. Given that the same method
of calibration was used for both imaging systems (see Correction of raw
mitochondrial morphology parameters above), it is likely that this bias was a
31


direct consequence of using lower resolution images. We observed that
fragmented mitochondria were often arranged collinearly and were very close
together, as shown in Figure 6E. Given that the gaps between mitochondria
were often only one or two pixels wide in the high-resolution images, it is likely
that collinear fragmented mitochondria were detected as single mitochondrial
bodies using the low-resolution images. No bias was observed in mitochondrial
width measurements, but the correlation between values obtained on different
imaging systems was less robust than for mitochondrial length measurements.
32


FIGURE 6: CONSISTENCY OF MEASURES USING DIFFERENT
MICROSCOPES (A) Representative image captured using high-resolution
settings on the Olympus microscope. (B) Representative image captured using
intentionally low-resolution settings on the Leica microscope. These images
were captured at 1/4th the pixel resolution and less than 1/10th the exposure
time of the high-resolution images. (C) Comparison of mitochondrion length
measurements between microscopes. The experimental repetition means of
mitochondrial length were closely correlated between the two microscopes
(R=0.82), but there was a bias towards longer measurements using the low
resolution system. Error bars represent SEM in measurements from each
microscope. (D) Comparison of mitochondrion width measurements between
microscopes. The experimental repetition means of mitochondrial width were
modestly correlated between the two microscopes (R=0.73). No bias was
observed in mitochondrial width measurements. (E) Representative high-
resolution microscope image showing colli near mitochondrial fragments. Such
fragments were likely indistinguishable using the low-resolution images, which
would account for the bias towards longer mitochondria with the low-resolution
system.
33


3.4 Consistency of Measures Between Experimental Repetitions
For all sample groups imaged, the first repetition of each experiment
was compared to the second repetition and the second was compared to the
third for each sample group metric. Images from the Olympus FV1000
microscope and the Leica SP8 microscope were analyzed separately to allow
assessment of experimental consistency within and between image sets
acquired using different imaging systems.
Figure 7 summarizes assessments of consistency between
experimental repetitions within treatment groups. Treatment group differences
were observed to be qualitatively consistent between experimental repetitions
(Figure 7A). Measured values of mitochondrial length for each group were
consistently similar between experimental repetitions, with R=0.96, p<0.0001
using the high-resolution image set and R=0.70, p=0.005 using the low
resolution image set (Figure 7B-C). Mean mitochondrion width was also
consistent between experimental repetitions, with R=0.77, p<0.0001 using the
high resolution image set and R=0.84, p=0.0002 using the low resolution image
set (Figure 7D-E). Although substantial variance was observed between
experimental repetitions, in most cases this variance was minor relative to
treatment group differences.
34


A)
Reps1&2 0im) Reps1&2(pm)
FIGURE 7: CONSISTENCY OF EXPERIMENTAL REPETITIONS. (A) Time
course of mean mitochondrion length in Wistar vSMCs exposed to 25 mM
glucose. Individual experimental repetitions (n=9 images) are shown in red,
while the overall average (n=27 images) is shown in black. (B) Agreement
between repetitions in mitochondrion length using high resolution microscope
settings. Very strong agreement is observed between successive experimental
repetitions (R=0.96). (C) Agreement between repetitions in mitochondrion
length using low resolution microscope settings. Modest agreement is
observed between successive experimental repetitions (R=0.70). (D)
Agreement between repetitions in mitochondrion width using high resolution
microscope settings. Strong agreement is observed between successive
experimental repetitions (R-0.77). (E) Agreement between repetitions in
mitochondrial width using low resolution microscope settings. Strong
agreement is observed between successive experimental repetitions (R=0.84).
35


4. Cellular Perturbation Test Cases
4.1 Effects of Fixation Temperature in HUVEC Cells
To assess the impact of the fixation conditions on mitochondrial network
parameters, we tested the impacts initial reagent temperatures and incubation
temperature during fixation on mitochondrial morphology. HUVEC cells were
cultured without experimental perturbation for 6-10 passages. Cells were
divided into one of four experimental groups and fixed according to the protocol
described above with specific adjustments as outlined below. Group 1 was
fixed using reagents pre-heated to 37 degrees Celsius and all incubation steps
during fixation were performed at 37 degrees Celsius; Group 2 was fixed using
room-temperature reagents and all incubation steps during fixation were
performed at 37 degrees Celsius; Group 3 was fixed using room-temperature
reagents and all incubation steps during fixation were performed at room
temperature; Group 4was fixed using reagents pre-heated to 37 degrees
Celsius and all incubation steps during fixation were performed at room
temperature.
Figure 8 summarizes the effects of fixation temperature on mitochondrial
morphology in HUVEC cells. Visual observation of the samples indicated a
more fragmented mitochondrial network in the group treated with room
36


temperature formaldehyde and incubated at room temperature during fixation,
and possibly a more peripheral distribution of mitochondrial mass in the group
treated with pre-warmed formaldehyde and incubated at 37 degrees Celsius
during fixation compared to other groups (Figure 8A). It was found that the
average mitochondrion length was significantly (p<0.0001) reduced in the
group treated with room temperature formaldehyde and incubated at room
temperature during fixation relative to every other group, and that no other
treatment group differences in mitochondrial length were statistically significant
(Figure 8B). Group means for mitochondrion length were 11.35+/-2.41 urn,
11.01+/-2.15 urn, 8.16+/-2.28 urn, and 12.58+/-3.73 urn for groups 1,2, 3, and
4, respectively. No group differences were found in average mitochondrion
width (p>0.05). Although visual analysis suggested perinuclear localization of
mitochondria in all groups treated with room temperature reagents or incubated
at room temperature at any stage during fixation, the fraction of the 10% of the
cytoplasm closest to the nucleus filled by mitochondria did not vary significantly
(p>0.05). Group means for mitochondrion width were 272+/-24 nm, 286+/-36
nm, 280+/-28 nm, and 289+/-32 nm for groups 1, 2, 3, and 4, respectively.
Finally, no group differences were found in total fraction of the cytoplasm filled
by mitochondria, as would be expected given that the time required for fixation
is considerably less than the time required for substantial mitochondrial
turnover. The means of reported endpoints are included in Table 3 below.
37


TABLE 3: EFFECTS OF TEMPERATURE DURING FIXATION ON
MITOCHONDRIAL MORPHOLOGY
Mean Mitochondrion Length (mean+/-SD) Mean Mitochondrion Width (mean+/-SD)
Group 1 11.35+/-2.41 um 272+1-24 nm
Group 2 11.01+/-2.15 um 286+/-36 nm
Group 3 8.16+/-2.28 um 280+/-28 nm
Group 4 12.58+7-3.73 um 289+7-32 nm
FIGURE 8: EFFECTS OF TEMPERATURE DURING FIXATION ON
MITOCHONDRIAL MORPHOLOGY. HUVEC cells were fixed using either
body-temperature (Group 1) or room-temperature (Group 2) reagents and
subsequently incubated at either body-temperature (Group 3) or room-
temperature (Group 4) for fifteen minutes during fixation. (A) Representative
images of mitochondrial networks following each treatment. Some
rearrangement of the mitochondrial network is apparent between any two
experimental cases. Cells fixed with room-temperature reagents and incubated
38


at room temperature (CC) show visible mitochondrial fragmentation. (B) Mean
mitochondrion length by treatment. Cells fixed with room-temperature reagents
and incubated at room temperature (CC) show a highly significant (p<0.0001)
reduction in mean mitochondrion length compared to other treatments, which
do not differ significantly. (C) Mean mitochondrion width by treatment. No
temperature effects on mitochondrial width were observed.
4.2 High Glucose Exposure in Primary Rat vSMCs
Primary smooth muscle cells from the aorta of GK or Wistar rats were
cultured in low glucose (5 mM) smooth muscle growth media until passages 8-
10. Prior to treatment, cells were switched to low glucose, serum free starvation
media for 48 hours. Serum starved cells were exposed to high glucose (25
mM), serum free media. For each cell type (GK and Wistar) in each repetition
of each experiment, 15 slides were prepared. Of these 15 slides, three were
merely rinsed with high glucose media before fixation and staining (0 minute
exposure), three were fixed and stained after 10 minutes in high glucose media,
three were fixed and stained after 30 minutes in high glucose media, three were
fixed and stained after 30 minutes in high glucose media, three were fixed and
stained after 60 minutes in high glucose media, and the remaining three were
fixed and stained after 240 minutes in high glucose media. These time points
were selected based on empirical observation that mitochondrial morphology
was highly dynamic in the first hour of high glucose exposure and then
underwent more gradual changes in the subsequent several hours.
39


Figure 9 summarizes the response of GK and Wistar vSMCs to
exposure to 25 mM glucose exposure. Dynamic changes in mitochondrial
morphology were not apparent to the naked eye, although the greater
mitochondrial length and lesser mitochondrial width among GK vSMCs relative
to Wistar vSMCs at baseline (Figure 4A) were apparent at all time-points.
Although subtle, dynamic changes detected using our algorithm were
consistent between experimental repetitions (Figure 7A). Wistar vSMCs
demonstrated a significant decrease in average mitochondrion length from 10
to 30 minutes (p<0.0001), followed by a significant (p<0.005) increase from 60
to 240 minutes (Figure 9A). No time-point changes in average mitochondrion
length were statistically significant among GK vSMCs, but trending (p<0.1)
decreases and increases were observed at the same time points as were
significant for the Wistar vSMCs. In addition to these differences in dynamic
changes in length, GK vSMCs had significantly (p<0.0001) longer mitochondria
than did Wistar vSMCs at all time-points recorded. Similarly, although average
mitochondrion width did not change significantly over time (p>0.05) for either
cell type, Wistar vSMCs had significantly (p<0.0001) wider mitochondria than
did GK vSMCs at all time-points recorded (Figure 9C) with overall mean values
of 309+/-26 nm for GK vSMCs and 347+/-22 nm for Wistar vSMCs.. Recorded
mean values for mitochondrion length were 6.80+/-0.86 urn at baseline, 6.78+/-
0.80 urn at 10 minutes, 6.35+/-0.79 urn at 30 minutes, 6.44+/-0.78 urn at 60
40


minutes, and 6.77+/-0.83 um at 240 minutes for GK vSMCs, and 5.09+/-0.77
urn at baseline, 5.28+/-0.59 um at 10 minutes, 4.65+/-0.54 um at 30 minutes,
4.64+/-0.52 um at 60 minutes, and 5.86+/-0.44 um at 240 minutes for Wistar
vSMCs.
In addition, the fraction of the 10% of the cytoplasm closest to the
nucleus filled by mitochondria was investigated (Figure 9B). Wistar vSMCs
showed a statistically significant increase (p<0.05) in the fraction of the
perinuclear space filled by mitochondria from 60 to 240 minutes. GK vSMCs
showed no dynamic response (p>0.1 for all comparisons) in this parameter,
and the fraction of the perinuclear space filled by mitochondria in GK vSMCs
was significantly greater (p<0.0001) than in Wistar vSMCs at all time-points
measured. Overall mean values for fraction of perinuclear space filled by
mitochondria were 11.8+/-1.9% at baseline, 10.7+/-1.7% at 10 minutes, 11.0+/-
2.0% at 30 minutes, 11.0+/-2.4% at 60 minutes, and 11.1+/-3.2% at 240
minutes for GK vSMCs, and 8.2+/-2.4% at baseline, 7.9+/-1.6% at 10 minutes,
7.9+/-1.8% at 30 minutes, 8.0+/-1.7% at 60 minutes, and 9.3+/-2.1% at 240
minutes for Wistar vSMCs.
41


FIGURE 9: MITOCHONDRIAL NETWORK RESPONSE TO INCUBATION IN
HIGH GLUCOSE (A) Mean mitochondrion length vs time in 25 mM glucose.
Mean mitochondrion length was much greater in GK vSMCs than in Wistar
VSMCs at all time-points measured (p<0.0001). Wistar vSMCs displayed a
highly dynamic response to glucose exposure including significant
fragmentation between 10 and 30 minutes (p<0.0001) and a significant
increase in networking between one hour and four hours (p<0.01). Although no
changes in average length were statistically significant in the GK vSMCs, there
was a trend towards qualitatively similar changes in network morphology at the
same time points (p<0.1 for both). (B) Fraction of the 10% of the cytoplasm
closest to the nucleus filled by mitochondria vs time in 25 mM glucose. No
significant changes in perinuclear mitochondrial mass were observed in GK
vSMCs. A significant increase in perinuclear mass between one and four hours
was observed in Wistar vSMCs, however (p<0.05). (C) Mean mitochondrion
width vs time in 25 mM glucose. No significant effects of high glucose on
mitochondrial width were observed. Wistar VSMC mitochondria were much
wider than GK vSMCs mitochondria at all time points measured (p<0.0001).
42


4.3 Effects of Hydrogen Peroxide Exposure with SH-SY5Y Neurons
Oxidative stress has been reported to perturb mitochondrial
morphology. These experiments examined the impact of H2O2 in a neuronal
cell line. SH-SY5Y neuronal cells were cultured without experimental
perturbation for a minimum of 5 passages. Cells were then changed to media
containing either 200 nM or 0 nM H2O2 for 2 hours, at least 24 hours after the
previous media change. Cells were then fixed and stained as described above.
Figure 10 summarizes the effects of hydrogen peroxide exposure on
mitochondrial morphology in SH-SY5Y cells. Mitochondria from cells exposed
to hydrogen peroxide were found to be visibly more fragmented than controls,
although the visible difference was subtle (Figure 10A). This difference was
reflected in algorithmic measurements, which indicated a significant (p<0.0001)
reduction in average mitochondrion length with hydrogen peroxide exposure
(Figure 10B). In addition, a trending increase (p<0.1) in average mitochondrion
width was recorded with hydrogen peroxide exposure, possibly indicating
mitochondrial swelling. Mean values for mitochondrion length were 4.31 +/-0.66
urn at baseline and 3.75+/-0.41 urn following hydrogen peroxide exposure,
while those for mitochondrion width were 320+/-14 nm at baseline and 327+/-
11 nm following hydrogen peroxide exposure. These values are also included
in Table 4 below.
43


TABLE 4: EFFECTS OF HYDROGEN PEROXIDE ON MITOCHONDRIAL
MORPHOLOGY
Mean Mitochondrion Length (mean+/-SD) Mean Mitochondrion Width (mean+/-SD)
Control 4.31+/-0.66 um 320+/-14 nm
200 nM H202 3.75+/-0.41 um 327+/-11 nm
A)
\
1 J
% V . V 'J V (
. c J ; . i V
*
N J ' O > /
BL
Average Length
FIGURE 10: EFFECTS OF HYDROGEN PEROXIDE ON MITOCHONDRIAL
MORPHOLOGY. (A) Representative images before and after peroxide
treatment. Mitochondria were visibly fragmented following treatment, but
changes were difficult to observe due to the very small size of the cells. (B)
Mean mitochondrion length before and after peroxide treatment. Hydrogen
peroxide induced a highly significant decrease in mean mitochondrion length
(p<0.0001). (C) Mean mitochondrion width before and after peroxide treatment.
Although the effect was very subtle, a trending increase in mitochondrial width
was observed following hydrogen peroxide treatment (p<0.1), possibly
indicating mitochondrial swelling.
44


5. Exporting this Technology for Use by Other Laboratories
5.1 Software Testing with Independently Produced Images
If this technology is to be exported to other laboratories, it is imperative
that investigators other than the PI of this study be able to produce
mitochondrial network images suitable for analysis with this software. The
staining protocol documented under 2.3 Fixation and staining was implemented
by another technician within our laboratory and applied to HUVEC cells either
in baseline conditions or following 1 or 4 hours of incubation in conditioned
media previously used to culture Wistar or GK vSMCs. Thus there were a total
of 5 groups: control, 1 hour exposure to GK media, 4 hour exposure to GK
media, 1 hour exposure to Wistar media, and 4 hour exposure to Wistar media.
All groups were imaged on the Leica microscope by a third independent
technician. The bead-calibration technique documented under 2.7 Correction
of raw mitochondrial morphology parameters was not employed in this
experiment, and so equivalent measures of HUVEC baseline parameters are
somewhat different between this experiment and those documented under 3.1
45


Baseline cell line characteristics and Effects of fixation temperature in HUVEC
cells. Note that although there are certain commonalities between this
experiment and the previously discussed glucose exposure time course (4.2
High Glucose Exposure in Primary Rat vSMCs), the data presentation in this
section has been altered to show time in hours rather than minutes (because
no time points less than 1 hour were employed) and to use a bar graph rather
than a line graph, reflecting the discrete rather than continuous nature of this
experiment.
The results of this experiment are summarized in Figure 11.
Representative images from each group are shown in Figure 11 A. All images
appeared to have been binarized successfully (Fig. 11B) and the fraction of the
network detected suitably for subsequent analysis was consistent with that in
previous analyses of FIUVEC cells (mean=0.96), indicating that the staining
protocols document under 2.3 Fixation and staining can be implemented
successfully by independent investigators. Furthermore, this experiment was
able to reveal that GK vSMC conditioned media triggered mitochondrial fusion
while Wistar vSMC conditioned media did not affect mitochondrial networking
(p<0.01, Fig. 11C), GK vSMC conditioned media triggered mitochondrial
swelling while Wistar vSMC conditioned media did not affect mitochondrial
diameter (p<0.05, Fig. 11D), and Wistar vSMC conditioned media triggered an
increase in the fraction of the perinuclear space filled by mitochondria while GK
46


vSMC conditioned media did not affect intracellular distribution of mitochondria
(p<0.01, Fig. 11E). All three of these group differences are consistent with
direct visual observations by the technicians who performed the staining and
imaging. The finding of mitochondrial swelling with GK vSMC conditioned
media was the first (and to date, the only) instance in which this software found
statistically significant differences in mitochondrial width between groups of the
same cell line. Mean values for mitochondrion length were 17.25+/-5.67 urn at
baseline, 20.41 +/-6.37 urn following 1 hr in GK-conditioned media, 21.82+/-
3.05 urn following 4 hrs in GK-conditioned media, 16.51+/-4.36 urn following 1
hr in Wistar-conditioned media, and 15.97+/-4.39 urn following 4 hrs in Wistar-
conditioned media. Mean values for mitochondrion width were 555+/-76 nm at
baseline, 544+/-88 nm following 1 hr in GK-conditioned media, 630+/-103 nm
following 4 hrs in GK-conditioned media, 561+/-97 nm following 1 hr in Wistar-
conditioned media, and 606+/-78 nm following 4 hrs in Wistar-conditioned
media. Mean values for fraction of perinuclear space filled by mitochondria
were 10.9+/-3.9% at baseline, 10.8+/-3.2% following 1 hr in GK-conditioned
media, 11.6+/-2.6% following 4 hrs in GK-conditioned media, 1 hr 10.1+/-2.3%
following 1 hr in Wistar-conditioned media, and 12.5+/-4.4% following 4 hrs in
Wistar-conditioned media. These values are also included in Table 5 below.
47


TABLE 5: EFFECTS OF VSMC CONDITIONED MEDIA ON
MITOCHONDRIAL MORPHOLOGY
Mean Mitochondrion Length (mean+/- SD) Mean Mitochondrion Width (mean+/- SD) Fraction of Perinuclear Space Filled by Mitochondria (mean+/-SD)
Control 17.25+/-5.67 um 555+/-76 nm 10.9+/-3.9%
GK 1 hr 20.41+/-6.37 um 544+/-88 nm 10.8+/-3.2%
GK 4 hrs 21.82+/-3.05 um 630+/-103 nm 11.6+/-2.6%
Wistar 1 hr 16.51+/-4.36 um 561+/-97 nm 10.1 +/-2.3%
Wistar 4 hrs 15.97+7-4.39 um 606+/-78 nm 12.5+7-4.4%
1 hour 4 hours
Average Length
Average Width
FIGURE 11: EFFECTS OF VSMC CONDITIONED MEDIA ON
MITOCHONDRIAL MORPHOLOGY. (A) Representative images from each
treatment group. Subtle differences were visible between treatment groups, but
heterogeneity within each group was dominant relative to group differences.
(B) Representative comparison of raw and binarized images. All images in this
independently produced data set were binarized successfully. (C) Mean
48


mitochondrion length by treatment group. GK media triggered mitochondrial
fusion relative to Wistar media (p<0.01). (D) Mean mitochondrion width by
treatment group. GK media triggered mitochondrial swelling (p<0.05). (E)
Fraction of the perinuclear space filled by mitochondria by treatment group.
Wistar media triggered perinuclear gathering of mitochondria (p<0.01).
5.2 Software Testing with an Uncontrolled Image Sample
As a litmus test of general applicability, the performance of the
algorithms reported in this manuscript was tested on an uncontrolled sample of
mitochondrial network images downloaded from the internet. The first ten
results of a Google Images search for mitochondrial fluorescence microscopy
as of 8-25-2015 were downloaded and the mitochondrial channel of each
image isolated. No attempt was made to determine the staining protocol, cell
type, microscope, or laser lines used for any of these images; rather, the
purpose of this image sample was to demonstrate that virtually any high-quality
microscope image could be used regardless of preparation. Because the
resolution of these images was not known, the pixel-wise correlation coefficient
(Pearsons R) between raw and binarized images (see Figure 1 A) was used as
a metric of binarization quality rather than attempting to infer the fraction of the
network validly detected, which requires knowledge of physical pixel
dimensions. This analysis thus assesses pre-processing of images, but not
post-processing and data extraction. Given that the latter two steps are
microscope-independent, we refer the reader to previous sections
49


documenting consistency and efficacy of software performance in various
cellular perturbation test cases.
Results of this analysis are illustrated in Figure 12. When applied to
baseline cell images (the same image set as was used for section 3.1 Baseline
cell line characteristics), the correlation coefficient between raw and binarized
images and fraction of the mitochondrial network validly detected were strongly
associated (R=0.99), indicating that fraction of network validly detected and
correlation strength between raw and binarized images function as equivalent
measures of binarization quality. When applied to the uncontrolled image
sample used for this litmus test, the mean correlation coefficient between raw
and binarized images was 0.75 +/- 0.05 (mean +/- sd), comparable to the
quality of detection in unperturbed HUVEC cells. The highest and lowest quality
binarized images are shown in Figure 12B-C.
50


FIGURE 12: ALGORITHM PERFORMANCE WITH AN UNCONTROLLED
IMAGE SAMPLE (A) Similarity between pixel intensity correlation and network
coverage metrics of binarization quality. Each metric can be almost entirely
predicted by the other, but network coverage provides a slightly higher dynamic
range. (B) Highest quality of binarization in the uncontrolled image sample;
correlation between raw and binary images R=0.83. (C) Lowest quality of
binarization in the uncontrolled image sample; correlation between raw and
binary images R=0.66.
51


6. Conclusions and Future Directions
6.1 Discussion of Results
In light of the tremendous quantity of research in recent years assessing
mitochondrial morphology endpoints, it is important that quantitative, objective,
practical, and consistent methods of obtaining mitochondrial morphology
measurements are developed. The algorithms documented in this manuscript
show excellent consistency in measured endpoints between experimental
repetitions and additional consistency in measured endpoints between different
microscopes. Our results indicate slightly better performance in length
measurements than in width measurements. The observation of consistent
results from one repetition to the next and between different microscopes holds
promise for enhancing consistency of results between different laboratories. In
addition, the binarization algorithms described in this study demonstrated a
quantitative advantage of over a simple thresholding algorithm by increasing in
the fraction of the mitochondrial network detected suitably for analysis, allowing
simultaneous measurement of perinuclear and peripheral mitochondria. Given
that perinuclear and peripheral mitochondria are subject to different regulation
[31,32, 331, any method for assessing mitochondrial morphology that excludes
52


a particular portion of the network is likely to introduce bias in whole-cell
measurements. Finally, the differences between cell types and treatment
groups detected by this software uniformly agree with subjective visual
assessment of mitochondrial morphology from raw microscope images.
Collectively, our results indicate that these methods enable quantitative,
objective, and consistent measurements of mitochondrial morphology.
Further evidence for this methods utility can be found in comparisons of
our experimental results to published observations. A number of studies in a
variety of cell types have reported mitochondrial fragmentation upon acute
exposure to high concentrations of glucose [16, 34, 35]. Yu et al report results
of particular interest, showing a similar two-phase mitochondrial morphology
response in neuronal cells to that which we observed in primary vSMCs,
consisting of rapid mitochondrial fission followed by gradual mitochondrial
fusion [16], Finally, a number of studies in a variety of cell types report
mitochondrial fragmentation upon acute exposure to hydrogen peroxide [36,
37, 381, echoing our own results in SFI-SY5Y neurons. Our findings that an
unsupervised algorithm was able to recreate these results indicates that
subjective or user-guided assessment of mitochondrial morphology is no longer
necessary.
The quantitative values for mitochondrial morphological parameters
obtained in this study are realistic. Mitochondrial morphology studies reporting
53


direct measurement of mitochondrial length typically report values in the range
of 1-15 urn [39, 40, 41_, 42, 43, 44, 45, 46], consistent with the values obtained
in this study. Literature values for mitochondrial width/diameter generally fall in
the range of several hundred nanometers [47, 48, 49, 50, 51_], also consistent
with those obtained in this study. Moreover, in studies that measure dynamic
changes in both length and width, width has been reported to be far more stable
than length [47, 50], agreeing with our findings of many statistically significant
changes of average mitochondrion length but none in average mitochondrion
width. In fact, the only arguable exception to stability in mitochondrial width in
this study was a trending increase (p<0.1) in mitochondrial width in SH-SY5Ys
exposed to 200 nM hydrogen peroxide. Oxidative stress has previously been
shown to induce mitochondrial swelling [52, 53, 54, 55], hinting that the
algorithms in this paper might be able to detect mitochondrial swelling under
certain conditions.
The results obtained using our algorithms hold important methodological
implications for future studies of mitochondrial morphology. First, the highest
resolution microscope available should be used for measurements of
mitochondrial morphology. We observed collinear arrangement of fragmented
mitochondria in several cell types, and these small, closely packed
mitochondrial bodies were not successfully distinguished using the lower
resolution of our two microscope setups. Even with the higher resolution setup,
54


SH-SY5Y mitochondria were poorly detected, indicating that this algorithm may
not be as well suited to smaller cells. The correlation strength between high
and low resolution image results was similar to the inter-experimental
agreement of low resolution images but lower than that of high resolution
images (Figures 5-6), indicating that measurement consistency in the low
resolution image set was the limiting factor to agreement with measurements
in the high resolution image set. Similar dynamic changes in mitochondrial
morphology were observed with both imaging systems and mitochondrial
fragmentation upon hydrogen peroxide exposure was detectable, however, so
use of a relatively low-resolution system with these algorithms may be
adequate to detect differences between treatment groups in the same
experiment, but not for comparisons between studies or for absolute
measurements. Likewise, the use of fluorescent beads to calibrate the
measurement software is necessary to remove bias and enable comparisons
between different microscopes or papers, but not necessary for distinguishing
differences between treatment groups within the same study.
Another interesting methodological finding is that fixation at room
temperature produced significantly and substantially different mitochondrial
morphology than did fixation at 37 degrees Celsius. We tested this effect after
observing markedly different mitochondrial morphologies between samples
fixed with cold formaldehyde vs warm formaldehyde during preliminary
55


methods development. This study did not attempt to determine the mechanism
underlying these differences, but it is reasonable to propose that studies of
mitochondrial morphology in fixed cells should use 37 degrees Celsius as a
physiologically relevant temperature during fixation. Given that many studies in
the literature use cooler or unspecified fixation temperatures [56, 57, 58, 59,
60, 611, it is possible that mitochondrial morphology endpoints in reports such
as these may have been compromised.
One advantage of the algorithms developed in this study is their ability
to reliably detect differences in mitochondrial morphology that are too subtle to
see with the naked eye, as was the case for both Wistar and GK vSMCs
exposed to 25 mM glucose. With this advantage comes the potential for
aberrant detection of artificially significant results. We found that a total of 3
slides per treatment group or time point per experimental repetition with 3
experimental repetitions and 3 images per slide for a total of n=27 images and
9 slides per treatment group was adequate to achieve statistical significance in
these cases. This large sample size was necessary in part because of the great
heterogeneity of mitochondrial network morphologies observed in a single slide
and in part due to the subtlety of some of the effects recorded. It would not be
feasible to use sample sizes this large if measurements had been performed
by hand. Using our algorithms, little additional time is dedicated to analysis with
increasing sample size. In the event of a subtle effect which is detected by the
56


software but not by the naked eye, we highly recommend comparing the results
from individual experimental repetitions as in Figure 6A. For purposes of this
study, if an endpoint difference was not visible to the naked eye, it was included
only if successive experimental repetitions gave qualitatively similar results.
In the introduction of this paper we proposed that software
measurements of mitochondrial morphology should not require a specific
microscope, algorithm expertise should not be required to operate the software,
endpoints should be consistent between experimental repetitions, no specific
portion of the mitochondrial network should be excluded from analysis or
incorrectly analyzed due to algorithm performance, and the endpoints reported
to the user should hold intuitive and tangible meanings. Our experiments
demonstrate algorithm independence from a specific microscope by showing
that measurements of the same samples on different microscopes yielded
similar results. Because the software is fully automated, no computational
expertise is required to operate it. Endpoints were found to be quantitatively
similar between successive experimental repetitions, and the method of
binarization developed in this study proved superior to simple thresholding in
preventing bias in measurements of specific portions of the mitochondrial
network. Finally, the primary endpoints reported to the user are average length
and width of a single mitochondrial body, which can easily be replicated with
by-hand measurements.
57


6.2 Likely Barriers to Implementation
Although the data included in this manuscript indicates the efficacy and
practicability of our software, widespread adoption of this technology will
require a good deal of continued development. For the same reasons as much
of the mitochondrial morphology literature uses subjective assessments of
mitochondrial morphology, it is relatively unlikely that the laboratories for whom
this software is intended will be proficient in Matlab. Future work on this
software will therefore include translation from Matlab (which allowed for rapid
iteration and testing during development) to either a stand-alone executable or
an ImageJ plugin (better suited for use by the general public). This translation
will serve to further ensure that no computational image processing or coding
expertise is required for use of the software.
Another likely barrier to implementation stems directly from the primary
advantage of the software: with absolute, quantitative measures it may prove
to be more difficult to reproduce experiments than with subjective assessment.
This increase in difficulty of replication, although it results from a higher
standard of replication, may make this software less attractive to potential
users. As an example of this effect, consider that the baseline FIUVEC
mitochondria included in the experiments of Chapter 5 are significantly (p<0.05
using unpaired T test) longer than the baseline FIUVEC mitochondria in the
experiments of Chapter 4 despite a nominally identical sample preparation.
58


Judging from inspection of the raw images, it appears likely that this difference
stemmed from different technician preferences in which cells to image; the
HUVECs imaged in Chapter 4 tended to be large, isolated cells, whereas those
in Chapter 5 tended to be smaller cells within a cell cluster. That our software
was able to detect a significant difference between these two image sets while
returning consistent results between experimental replicates within each image
set indicates that even very subtle procedural differences might result in one
laboratory obtaining different results from another. Although this fact may make
the software less attractive to its end users, it is indicative of a more rigorous
and objective analysis, and thus no effort will be made to remove this barrier to
implementation.
Finally, application of this software to samples prepared with different
fluorophores or different statistical designs (i.e. different number of
experimental replicates and/or sample sizes) may prove to be challenging.
Although the software was able to achieve high quality binarization of an
uncontrolled image set (see Chapter 5), it remains unclear whether consistent
detection of very subtle effects such as those reported in section 4.1 will be
possible with alternative study designs. In addition, the bead calibration method
used to account for system error may prove technically challenging for less
experienced laboratories. These and many other minor details of software
implementation are likely to prevent simple plug and play adoption of this
59


technology, but they are unlikely to outweigh the benefits of objective and
quantitative endpoints. The investigators will gladly provide this software to
other research groups upon request and assist in adoption and implementation
of this technology.
6.3 Potential Future Applications
To date, this software has only been rigorously tested on microscope
images of mitochondrial networks in fixed cells acquired using a confocal
microscope in. In principle, the same technique would be expected to work
regardless of slice width (although the quantities measured would vary), thus
enabling use with widefield microscopes. Ongoing experiments will further
define the relationship between slice width and software functionality. Similarly,
use with live cells and/or videomicroscopy is a logical future step. Preliminary
experiments (data not shown) indicate that live vs fixed cells do not influence
the ability of the software to measure mitochondria, and future experiments will
be required to determine if this method is suitable for videomicroscopy. As a
final note, although this method was validated with and designed for
mitochondrial images, it is equally applicable in concept to microscope images
of the endoplasmic reticulum (ER) and golgi apparatus, which both form
reticular networks that are qualitatively similar to the mitochondrial network. In
addition to the specific experimental questions outlined above, the largest
60


portion of future work on this project will involve assisting outside laboratories
with adoption of this technology. It is this last task that will take precedence
over any of the other proposed future directions discussed here.
61


LITERATURE CITED
1. Kaftan, E. J., Xu, T., Abercrombie, R. F., & Hille, B. (2000). Mitochondria
shape hormonally induced cytoplasmic calcium oscillations and
modulate exocytosis. Journal of Biological Chemistry, 275(33), 25465-
25470.
2. Marhl, M., Schuster, S., & Brumen, M. (1998). Mitochondria as an
important factor in the maintenance of constant amplitudes of cytosolic
calcium oscillations. Biophysical chemistry, 71(2), 125-132.
3. Smaili, S. S., Hsu, Y. T., Carvalho, A. C. P., Rosenstock, T. R., Sharpe,
J. C., & Youle, R. J. (2003). Mitochondria, calcium and pro-apoptotic
proteins as mediators in cell death signaling. Brazilian Journal of medical
and biological research, 36(2), 183-190.
4. Berridge, M. J. (2008). Smooth muscle cell calcium activation
mechanisms. The Journal of physiology, 586(21), 5047-5061.
5. Ebashi, S., & Endo, M. (1968). Calcium and muscle contraction.
Progress in biophysics and molecular biology, 18, 123-183.
6. Launay, P., Cheng, H., Srivatsan, S., Penner, R., Fleig, A., & Kinet, J.
P. (2004). TRPM4 regulates calcium oscillations after T cell activation.
Science, 306(5700), 1374-1377.
62


7. Rizzuto, R., De Stefani, D., Raffaello, A., & Mammucari, C. (2012).
Mitochondria as sensors and regulators of calcium signalling. Nature
Reviews Molecular Cell Biology, 13(9), 566-578.
8. Nishikawa, T., Edelstein, D., Du, X. L., Yamagishi, S. I., Matsumura, T.,
Kaneda, Y., Yorek, M. & Brownlee, M. (2000). Normalizing mitochondrial
superoxide production blocks three pathways of hyperglycaemic
damage. Nature, 404(6779), 787-790.
9. Sena, L. A., Li, S., Jairaman, A., Prakriya, M., Ezponda, T., Hildeman,
D. A., Wang, C., & Chandel, N. S. (2013). Mitochondria are required for
antigen-specific T cell activation through reactive oxygen species
signaling. Immunity, 38(2), 225-236.
10. Brand, M. D., Affourtit, C., Esteves, T. C., Green, K., Lambert, A. J.,
Miwa, S., Pakay, J., & Parker, N. (2004). Mitochondrial superoxide:
production, biological effects, and activation of uncoupling proteins. Free
Radical Biology and Medicine, 37(6), 755-767.
11. Pletjushkina, O. Y., Lyamzaev, K. G., Popova, E. N., Nepryakhina, O.
K., Ivanova, O. Y., Domnina, L. V., Chernyak, B. V., & Skulachev, V. P.
(2006). Effect of oxidative stress on dynamics of mitochondrial reticulum.
Biochimica et Biophysica Acta (BBA)-Bioenergetics, 1757(5), 518-524.
12. Murata, T., Goshima, F., Daikoku, T., Inagaki-Ohara, K., Takakuwa, H.,
Kato, K., & Nishiyama, Y. (2000). Mitochondrial distribution and function
63


in herpes simplex virus-infected cells. Journal of General Virology,
81(2), 401-406.
13. de Boer, V. C., & Houten, S. M. (2014). A Mitochondrial Expatriate:
Nuclear Pyruvate Dehydrogenase. Cell, 158(1), 9-10.
14. Chen, H., Chomyn, A., & Chan, D. C. (2005). Disruption of fusion results
in mitochondrial heterogeneity and dysfunction. Journal of Biological
Chemistry, 280(28), 26185-26192.
15. Youle, R. J., & Narendra, D. P. (2011). Mechanisms of mitophagy.
Nature reviews Molecular cell biology, 12(1), 9-14.
16. Yu, T., Robotham, J. L., & Yoon, Y. (2006). Increased production of
reactive oxygen species in hyperglycemic conditions requires dynamic
change of mitochondrial morphology. Proceedings of the National
Academy of Sciences of the United States of America, 103(8), 2653-
2658.
17. Quintana, A., & Hoth, M. (2012). Mitochondrial dynamics and their
impact on Tcell function. Cell calcium, 52(1), 57-63.
18. Liesa, M., & Shirihai, O. S. (2013). Mitochondrial dynamics in the
regulation of nutrient utilization and energy expenditure. Cell
metabolism, 17(4), 491-506.
64


19. Karbowski, M., & Youle, R. J. (2003). Dynamics of mitochondrial
morphology in healthy cells and during apoptosis. Cell Death &
Differentiation, 10(8), 870-880.
20. Liesa, M., Palacfn, M., & Zorzano, A. (2009). Mitochondrial dynamics in
mammalian health and disease. Physiological reviews, 89(3), 799-845.
21. Alano, C. C., Beutner, G., Dirksen, R. T., Gross, R. A., & Sheu, S. S.
(2002). Mitochondrial permeability transition and calcium dynamics in
striatal neurons upon intense NMDA receptor activation. Journal of
neurochemistry, 80(3), 531-538.
22. Picard, M., Shirihai, O. S., Gentil, B. J., & Burelle, Y. (2013).
Mitochondrial morphology transitions and functions: implications for
retrograde signaling?. American Journal of Physiology-Regulatory,
Integrative and Comparative Physiology, 304(6), R393-R406.
23. Yoon, Y., Galloway, C. A., Jhun, B. S., & Yu, T. (2011). Mitochondrial
dynamics in diabetes. Antioxidants & redox signaling, 14(3), 439-457.
24. Dietrich, M. O., Liu, Z. W., & Horvath, T. L. (2013). Mitochondrial
dynamics controlled by mitofusins regulate Agrp neuronal activity and
diet-induced obesity. Cell, 155(1), 188-199.
25.Seo, A. Y., Joseph, A. M., Dutta, D., Hwang, J. C., Aris, J. P., &
Leeuwenburgh, C. (2010). New insights into the role of mitochondria in
65


aging: mitochondrial dynamics and more. Journal of cell science,
123(15), 2533-2542.
26.0ng, S. B., & Hausenloy, D. J. (2010). Mitochondrial morphology and
cardiovascular disease. Cardiovascular research, 88(1), 16-29.
27. Aboud, A. A., Tidball, A. M., Kumar, K. K., Neely, M. D., Han, B., Ess, K.
C., Hong, C. C., Erikson, K. M., Hedera, P., & Bowman, A. B. (2015).
PARK2 patient neuroprogenitors show increased mitochondrial
sensitivity to copper. Neurobiology of disease, 73, 204-212.
28. Li, L., Nadanaciva, S., Berger, Z., Shen, W., Paumier, K., Schwartz, J.,
Mou, K., & Hirst, W. D. (2013). Human A53T a-synuclein causes
reversible deficits in mitochondrial function and dynamics in primary
mouse cortical neurons. PloS one, 8(12), e85815.
29. Xie, W., & Chung, K. K. (2012). Alpha-synuclein impairs normal
dynamics of mitochondria in cell and animal models of Parkinsons
disease. Journal of neurochemistry, 122(2), 404-414.
30. Koopman, W. J., Visch, H. J., Smeitink, J. A., & Willems, P. H. (2006).
Simultaneous quantitative measurement and automated analysis of
mitochondrial morphology, mass, potential, and motility in living human
skin fibroblasts. Cytometry Part A, 69(1), 1-12.
31. Park, M. K., Ashby, M. C., Erdemli, G., Petersen, O. H., &Tepikin, A. V.
(2001). Perinuclear, perigranular and sub-plasmalemmal mitochondria
66


have distinct functions in the regulation of cellular calcium transport. The
EMBO Journal, 20(8), 1863-1874.
32. Bahamonde, M. I., & Valverde, M. A. (2003). Voltage-dependent anion
channel localises to the plasma membrane and peripheral but not
perinuclear mitochondria. European Journal of Physiology, 446(3), 309-
313.
33. Collins, T. J., Berridge, M. J., Lipp, P., & Bootman, M. D. (2002).
Mitochondria are morphologically and functionally heterogeneous within
cells. The EMBO Journal, 21(7), 1616-1627.
34. Trudeau, K., Molina, A. J., Guo, W., & Roy, S. (2010). High glucose
disrupts mitochondrial morphology in retinal endothelial cells:
implications for diabetic retinopathy. The American journal of pathology,
177(1), 447-455.
35. Yu, T., Jhun, B. S., & Yoon, Y. (2011). High-glucose stimulation
increases reactive oxygen species production through the calcium and
mitogen-activated protein kinase-mediated activation of mitochondrial
fission. Antioxidants & redox signaling, 14(3), 425-437.
36. Jendrach, M., Mai, S., Pohl, S., Voth, M., & Bereiter-Hahn, J. (2008).
Short-and long-term alterations of mitochondrial morphology, dynamics
and mtDNA after transient oxidative stress. Mitochondrion, 8(4), 293-
304.
67


37. Jahani-Asl, A., Cheung, E. C., Neuspiel, M., MacLaurin, J. G., Fortin, A.,
Park, D. S., McBride, H. M., & Slack, R. S. (2007). Mitofusin 2 protects
cerebellar granule neurons against injury-induced cell death. Journal of
Biological Chemistry, 282(33), 23788-23798.
38. Fan, X., Flussien, R., & Brooks, G. A. (2010). H 2 0 2-induced
mitochondrial fragmentation in C2C 12 myocytes. Free Radical Biology
and Medicine, 49(11), 1646-1654.
39. Rintoul, G. L., Filiano, A. J., Brocard, J. B., Kress, G. J., & Reynolds, I.
J. (2003). Glutamate decreases mitochondrial size and movement in
primary forebrain neurons. The Journal of neuroscience, 23(21), 7881-
7888.
40. Nozaki, T., Kagaya, Y., Ishide, N., Kitada, S., Miura, M., Nawata, J.,
Ohno, I., Watanabe, J., & Shirato, K. (2001). Interaction between
sarcomere and mitochondrial length in normoxic and hypoxic rat
ventricular papillary muscles. Cardiovascular Pathology, 10(3), 125-132.
41. Wang, D. B., Garden, G. A., Kinoshita, C., Wyles, C., Babazadeh, N.,
Sopher, B., Kinoshita, Y., & Morrison, R. S. (2013). Declines in Drp1 and
parkin expression underlie DNA damage-induced changes in
mitochondrial length and neuronal death. The Journal of Neuroscience,
33(4), 1357-1365.
68


42. Korobova, F., Ramabhadran, V., & Higgs, H. N. (2013). An actin-
dependent step in mitochondrial fission mediated by the ER-associated
formin INF2. Science, 339(6118), 464-467.
43. Choi, S., Yu, E., Kim, D. S., Sugimori, M., & Llinas, R. R. (2015). RNS60,
a charge-stabilized nanostructure saline alters Xenopus Laevis oocyte
biophysical membrane properties by enhancing mitochondrial ATP
production. Physiological reports, 3(3), e12261.
44. Cagalinec, M., Safiulina, D., Liiv, M., Liiv, J., Choubey, V., Wareski, P.,
Veksler, V., & Kaasik, A. (2013). Principles of the mitochondrial fusion
and fission cycle in neurons. Journal of cell science, 126(10), 2187-
2197.
45. Korobova, F., Gauvin, T. J., & Higgs, H. N. (2014). A role for myosin II
in mammalian mitochondrial fission. Current Biology, 24(4), 409-414.
46. Gan, X., Huang, S., Wu, L., Wang, Y., Hu, G., Li, G., Zhang, H., & Yan,
S. S. (2014). Inhibition of ERK-DLP1 signaling and mitochondrial
division alleviates mitochondrial dysfunction in Alzheimer's disease
cybrid cell. Biochimica et Biophysica Acta (BBA)-Molecular Basis of
Disease, 1842(2), 220-231.
47. Moran, M., Rivera, H., Sanchez-Arago, M., Blazquez, A., Merinero, B.,
Ugalde, C., Arenas, J., Cuezva, J. M., & Martin, M. A. (2010).
Mitochondrial bioenergetics and dynamics interplay in complex I-
69


deficient fibroblasts. Biochimica et Biophysica Acta (BBA)-Molecular
Basis of Disease, 1802(5), 443-453.
48. Finley, L. W., Lee, J., Souza, A., Desquiret-Dumas, V., Bullock, K.,
Rowe, G. C., Procaccio, V., Clish, C. B., Arany, Z., & Haigis, M. C.
(2012). Skeletal muscle transcriptional coactivator PGC-1a mediates
mitochondrial, but not metabolic, changes during calorie restriction.
Proceedings of the National Academy of Sciences, 109(8), 2931-2936.
49. Karam, S. M., Straiton, T., Hassan, W. M., & Leblond, C. P. (2003).
Defining epithelial cell progenitors in the human oxyntic mucosa. Stem
Cells, 21(3), 322-336.
50. Santamarfa, G., Martfnez-Diez, M., Fabregat, I., & Cuezva, J. M. (2006).
Efficient execution of cell death in non-glycolytic cells requires the
generation of ROS controlled by the activity of mitochondrial H+-ATP
synthase. Carcinogenesis, 27(5), 925-935.
51. Holloway, G. P., Gurd, B. J., Snook, L. A., Lally, J., & Bonen, A. (2010).
Compensatory increases in nuclear PGC1a protein are primarily
associated with subsarcolemmal mitochondrial adaptations in ZDF rats.
Diabetes, 59(4), 819-828.
52. Chattipakorn, S. C., Thummasorn, S., Sanit, J., & Chattipakorn, N.
(2014). Phosphodiesterase-3 inhibitor (cilostazol) attenuates oxidative
70


stress-induced mitochondrial dysfunction in the heart. Journal of
geriatric cardiology: JGC, 11(2), 151.
53. Lakroun, Z., Kebieche, M., Lahouel, A., Zama, D., Desor, F., &
Soulimani, R. (2015). Oxidative stress and brain mitochondria swelling
induced by endosulfan and protective role of quercetin in rat.
Environmental Science and Pollution Research, 22(10), 7776-7781.
54. Rodrfguez-Martfnez, E., Martinez, F., Espinosa-Garcfa, M. T.,
Maldonado, P., & Rivas-Arancibia, S. (2013). Mitochondrial dysfunction
in the hippocampus of rats caused by chronic oxidative stress.
Neuroscience, 252, 384-395.
55. Paneni, F., Capretti, G., Costantino, S., Chiandotto, S., Akhmedov, A.,
Di Stasio, E., Rocca, B., Luscher, T. F., Volpe. M., & Cosentino, F.
(2013). The lifespan determinant p66shc drives obesity-induced
oxidative stress, mitochondrial dysfunction and vascular inflammation.
European Fleart Journal, 34(suppl 1), 781.
56. Pyakurel, A., Savoia, C., Hess, D., & Scorrano, L. (2015). Extracellular
Regulated Kinase Phosphorylates Mitofusin 1 to Control Mitochondrial
Morphology and Apoptosis. Molecular cell, 58(2), 244-254.
57.0noue, K., Jofuku, A., Ban-lshihara, R., Ishihara, T., Maeda, M.,
Koshiba, T., Itoh, T., & Ishihara, N. (2013). Fis1 acts as a mitochondrial
71


recruitment factor for TBC1D15 that is involved in regulation of
mitochondrial morphology. Journal of cell science, 126(1), 176-185.
58. Zhu, M., Li, W., & Lu, C. (2012). Role of alpha-synuclein protein levels
in mitochondrial morphology and cell survival in cell lines. PloS one,
7(4), e36377.
59. Newman, L. E., Zhou, C. J., Mudigonda, S., Mattheyses, A. L., Paradies,
E., Marobbio, C. M. T., & Kahn, R. A. (2014). The ARL2 GTPase Is
Required for Mitochondrial Morphology, Motility, and Maintenance of
ATP Levels. PloS one, 9(6): e99270
60. Picard, M., White, K., & Turnbull, D. M. (2013). Mitochondrial
morphology, topology, and membrane interactions in skeletal muscle: a
quantitative three-dimensional electron microscopy study. Journal of
Applied Physiology, 114(2), 161-171.
61. Balan, I. S., Saladino, A. J., Aarabi, B., Castellani, R. J., Wade, C., Stein,
D. M., Eisenberg, H. M., Chen, H. H., & Fiskum, G. (2013). Cellular
alterations in human traumatic brain injury: changes in mitochondrial
morphology reflect regional levels of injury severity. Journal of
neurotrauma, 30(5), 367-381.
72


Full Text

PAGE 1

FULLY AUTOMATED SOFTWARE FOR QUANTITATIVE MEASUREMENTS OF MITOCHONDRIAL MORPHOLOGY by PENN MASON MCCLATCHEY JR B.S., Georgia Institute of Technology, 2013 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Master of Science Bioengineering 2015

PAGE 2

ii This thesis for the Master of Science degree by Penn Mason McClatchey, Jr. has been approved for the Bioengineering Program by Jane E.B. Reusch, Advisor Richard K.P. Benninger, Chair Kendall S. Hunter November 20, 2015

PAGE 3

iii McClatchey, Penn Mason (M.S. Bioengineering) Fully Automated Software for Quantitative Measurements of Mitochondrial Morphology Thesis directed by Professor Jane E.B. Reusch ABSTRACT Mitochondria undergo dynamic changes in morphology in order to adapt to changes in nutrient and oxygen availability, communicate with the nucleus, and modulate intracellular calcium dynamics. M any recent papers have been published assessing mitochondrial morphology endpoints. Alth ough these studies have yielded valuable insights, contemp orary assessment of mitochondrial morphology is typic ally subjective and qualitative, precluding direct comparison of outcomes between different studies and likely missing many subtle effects. In this paper, we describe a novel algor i thm for measuring the average length, average width, and spatial density of mitochondria in a fluorescent microscope image This method was applied to distinguish baseline characteristics of H uman Umbilical Vein Endothelial Cells (H UVECs ) primary G oto K akizaki rat aortic smoot h muscle cell s ( G K vSMC s ), primary Wistar rat

PAGE 4

iv aortic smooth muscle cells ( Wistar vSMC s ), and SH SY5Ys ( human neuroblastoma cell line). Consistent with direct observation, our algorithms found SH SY5Ys to have the greatest mitochondrial density and mitochondrial width while HUVECs were found to have the longest, thinnest mitochondria. M itochondrial morphology responses to temperature, nutrient, and oxidative stressors were characterized to test algorithm performance Large morphology changes recorde d by the software agreed with direct observation and subtle but consistent morphology changes were found that would not otherwise have been detected Endpoints were consistent between experimental repetitions with high quality microscope images (R=0.96 for length, R=0.77 for width), and maintained reasonable agreement even when the microscope used was intentionally de tuned to produce grainy, low resolution images (R=0.70 for l ength, R=0.84 for width). In addition, the correlation between high and low quality images was found to be similar to the self agreement of the low quality images (R=0.82 for length, R=0.72 for width), indicating that image quality rather than algorithm fu nction was the limiting factor in reproducibility. Finally, our algorithm was compared to a simple thresholding technique and was found to measure a far greater fraction of the mitochondrial network reliably (p<0.001), primaril y by enabling reliable simultaneous detection of perinuclear and peripheral mitochondria These results indicate that the automated

PAGE 5

v software described herein allow s quantitative and objective characterization of mitochondrial morphology from fluorescent microscope images. The f orm and content of this abstract are approved. I recommend its publication. Approved : Jane E.B. Reusch

PAGE 6

vi A CKNOWLEDGEMENT I would like to acknowledge each of my thesis committee members for their contributions to this project and for the mentorship they provided me during my pursuit of this degree. In addition, I would like to acknowledge the work done by Amy Keller, Ph.D., Leslie Knaub, M.S., Ron Bouchard, M.A., and Chelsea Connon, B.S., which enabled the investigation documented herein.

PAGE 7

vii TABLE OF CONTENTS Chapter 1. Introduction to M itochondrial M orphology 1.1 The Role of Mitochondria in ATP P roduction .. .1 1.2 The Role of M itochondria in Intracellular S ..... .3 1 .3 Structure Function Coupling in the Mitochondrial N etwork 1.4 Scientific Justification of the P roject 2. Methodological Design of Mitochondrial Measurement S oftware ... 2.1 Reagents 2.2 Cell C ulture 2.3 Fixation and S taining 10 2.4 Microscopy and Imaging T echniques 11 2.5 Image Binarization for Computational M easurements 12 2.6 Computation of Raw Mitochondrial Morphology P arameters .. 17 2.7 Correction of Raw Mitochondrial Morphology P arameters .. 19 2.8 Post Processing of D ata 22 2.9 Statistics and Software P ackages 22 3. Internal Validation of Mitochondrial Measurement S oftware 24 3.1 Baseline Cell Line C haracteristics 24

PAGE 8

viii 3.2 Performance of Custom Algorithms and Simple T hresholding .. 27 3.3 Consistency of Measure s B etween M icroscopes 30 3.4 Consistency of Measures Between Experimental R epetitions 32 4. Cellular Perturbation Measurement Test C ases 35 4.1 Effects of Fixation Temperature in HUVEC C ells .. 35 4.2 High Glucose E xposure in Primary R at vSMCs .. 38 4.3 Effects of Hydrogen Peroxide Exposure in SHSY 5Y N eurons 41 5. Exporting this Technology for Use by other L aboratories 5.1 Software Testing with Independently P roduced I mages 5.2 Software Testing with U ncontrolled Sample I mages 6. Conclusions and Future D irections 6.1 Discussion of R esults 6.2 Likely Barriers to I mplementation 6.3 Potential future applications Literature Cited

PAGE 9

ix LIST OF FIGURES Figure 1. Outline of Cellular Energetics and Mitochondrial R espiration .. .2 2. Method Used to Binarize Raw Microscope I mages .. 14 3. Calibration Using Fluorescent Beads of Known S ize .. 21 4. Baseline Cell Line C haracteristics 26 5. Performance of Custom Algorithm R elative t o T hresholding 6. Consistency of Measures Using Different Mi croscopes 7. Consistency of Experimental R epetitions 8. Effects o f Temperature During Fixation on Mitochondrial M orphology 9. Mitochondrial Network Response to Incubation in High G lucose 10. Effects of Hydrogen 11. Effects of vSMC Conditioned Media on Mitochondrial M orphology 12. Algorithm Performance with an Uncontrolled Image S ample

PAGE 10

x LIST OF TABLES Table 1. Microscope Settings Used in This 4. 5. Effects of vSMC Conditione

PAGE 11

1 1. Introduction to Mitochondrial Morphology 1.1 The Role of M itochondria in ATP P roduction The primary role of the mitochondrial network is to produce ATP for the cell. Raw macronutrients such as glucose and fatty acids undergo glycolysis and beta oxidation, respectively to prepare them for the citric acid cycle (TCA) The TCA serves to reduce NAD + to NADH, which acts as an electron donor in the electron transport chain (ETC), where a maj ority of intracellular ATP production occurs. An outline of these pathways is shown in Figure 1. Eukaryotic cells derive some energy from mitochondrial respiration (TCA and ETC) which generates 36 ATP molecules per glucose molecule and 131 ATP molecules p er fat molecule, and some from glycolysis which generates 2 ATP molecules per glucose molecule. Accordingly, when o xygen is not rate limiting to mitochondrial respiration (as is typically the case ), the TCA is the dominant source of energy for cellular processes. Highly metabolically active tissues such as the brain or skeletal muscle generally have high mitochondrial content to accommodate their unusually high energetic need s.

PAGE 12

2 FIGURE 1 : OUTLINE OF CELLULAR ENERGETICS AND MITOCHONDRIAL RESPIRATION Glycolysis and beta oxidation prepare glucose and fatty acids for mitochondrial respiration, respectively. Pyruvate produced through glycolysis is oxidized to acetyl CoA before entry into th e citric acid cycle (TCA). Glycolysis, beta oxidation, and the TCA are all involved in reduction of NAD+ and FAD to NADH and FADH 2 respectively, which serve as electron donors in the electron transport chain (ETC), where most ATP production occurs. Note t hat many intermediates are neglected in this figure, and the consumption of oxygen required for these processes is not illustrated here.

PAGE 13

3 1.2 The Role of Mitochondria in Intracellular S ignaling In addition to providing most of the ATP consumed in a typical Eukaryotic cell, mitochondria also serve as multi modal hubs of intracellular signaling. The mitochondrial network helps to buffer cytoplasmic calcium oscillations [ 1 2 ] and regulate increases in cytosolic calcium content during apoptosis [ 3 ] S ince calcium dynamics are critically involved in cellular processes ranging from muscle contraction [ 4 5 ] to immune cell activation [ 6 ], the role of mitochondria in regulating calcium dynamics is likely critical for normal cellular function. The role of mitochondria in regulation of cytosolic calcium signaling has been extensively reviewed elsewhere [ 7 ] Likewise, because the ETC is the primary site of oxidation/reduction reactions within the cell, mitochondria are also a major source of free radicals. Although mitochondrial free radical production can play a pathological role [ 8 ], mitochondrial free radicals are also involved in normal cellular signaling processes, typically in the context of a cellular stress response [ 9 ]. The role of mitochondrial superoxide (the prim ary free radical produced by mitochondria) in cellular signaling has been extensively reviewed elsewhere [ 10 ] An emerging area in mitochondrial signaling is possible commun ication between mitochondria and the nucleus It has been shown that certain stimuli can cause perinuclear gathering of mitochondria [ 11 12 ] and others can result in accumulation of mitochondrial proteins in the nucleus [ 13 ], but the meaning of

PAGE 14

4 these findings is not yet clear. A detailed review of mitochondrial involvement in intracellular signaling processes is beyond the scope of this introduction; for purposes of this manuscript it i s sufficient to recognize that visible changes in the structure of the mitochondrial network may indicate changes in intracellular signaling processes as well as changes in cellular energetics. 1.3 Structure Function Coupling in the Mitochondrial N etwork In order to perform its many functions in cellular signaling and energy homeostasis, the mitochondrial network must be highly dynamic. One important manifestation of this dynamism lies in dynamic changes in network morphology during certain physiologic responses. M itochondria operate more efficiently when networked [ 14 ], and this increase in efficiency is thought to be mediated by allowin g each individual organelle s flaws to be diluted in the larger network. Mitochondrial fission allows separation of damaged and intact mitochondrial fragments for subsequent autophagy or re integration into the mitochondrial network, respectively [ 15 ] In addition to preserving the quality and efficiency of the mitochondrial network, changes in mitochondrial morphology also modulate mitochondrial calcium signaling [ 7 ] and redox signa ling [ 16 ] Mitochondrial dynamics are altered during cellular activation and/or proliferation [ 17 ], and nutrient stressors tend to induce dynamic mitochondrial responses in cells such as pancreatic beta cells that work as

PAGE 15

5 metabolic sensors [ 18 ]. Consistent with a role for mitochondrial dynamics in these crucial cellular processes, mutations in genes regulating mitochondrial morphology result in disease and many d isease states involve changes in mitochondrial morphology [ 19 ]. 1.4 Scientific J us tification of the P roject The study of mitochondrial morphology has become a topic of intensive investi search returned 1331 results as of 4/14/2015, and many of these papers are published in high profile journals. This vas t body of research is unsurprising in light of recent reports indicating that mitochondrial dynamics are involved in regulating cellular respiration [ 14 ] calciu m dynamics [ 7 ] free radical production [ 16 ] substrate catabolism [ 18 ] cell survival [ 20 ] cell activation [ 21 ] and perhaps epigenetic s [ 22 ] A lterations in mitochondrial dynamics have been reported in a wide variety of disease states including diabetes [ 23 ] obesity [ 24 ] aging [ 25 ] vascular proliferative disorders [ 26 ] and several genetic conditions that directly influence regulation of mitochondrial morphology [ 19 ] Clearly, mitochondrial dynamics are critically involved in human physiology and their study is thus warranted.

PAGE 16

6 A great deal has been learn ed about mitochondrial morphology and dynamics in recent years, but the current state of the art for assessing mitochondrial morphology leaves much to be desired. Of the first 50 results on a PubMed search for mitochondrial morphology sorted by relevance (as of 4/14/2015) 31 are primary research articles assessing mitochondrial networking. Of these, 3 use d molecular or histologic markers as surrogates for mitochondrial morphology measurements, 18 report ed subjective assessment s of morphology, 4 perform ed morphological measurements by hand, and 6 use d some manner of software measurement. It is troubling that most common methods for assessing mitochondrial morphology involve no objective measurements of mitochondrial morpholog y S ubjective assessment is ill suited to comparison of findings between studies or between cell types, and molecular markers do not necessarily relate to actual morphology in an intuitive manner. B ecause these methods are relatively insensitive, the intervent ions employed to perturb mitochondrial morphology are often extreme relative to what a cell might encounter in vivo. T aking measurements by hand avoids these pitfalls, but it is very slow, thus limiting sample sizes and increas ing time per experimental ite ration. A quantitative, objective, and consistent software method of measuring mitochondrial morphology is attractive in that it allows comparison between studies or cell types, might detect more subtle changes

PAGE 17

7 in morphology than subjective assessment, and allows for efficient analysis of a large sample size Some investigators have begun to use software tools, such as the Mitochondrial Morphology plugin in ImageJ [ 27 28 29 ] This software allows introduction of investigator bias since the user s elects the thresholding parameters and the endpoints reported to the user are rather limited In addition, our results indicate that thresholding allows reliable measurement of only a portion of the mitochondrial network. T he Fourier based algorithm of Ko opman et al reliably detect s mitochondria [ 30 ] but implementation of this algorithm is technically sophisticated likely accounting for the limited use of this method in the nine years since its publication. We propose that a practical software solution for the measurement of mitochondrial morphology should meet six basic criteria: algorithm performance should not require a specific microscope algorithm expertise should not be required to operate the software, quantitative outcome s should be consistent between different microscopes when applied to the same samples, endpoints should be consistent between experimental repetitions, no s pecific portion of the mitochondrial network should be excluded from analysis or incorrectly analyzed due to algorithm performance, and the endpoints reported to the user should hold intuitive and tangible meanings. The final p o glance, but we

PAGE 18

8 contend that if the output does not align with visual impressions, it will be difficult to trust the results. These six criteria were used to guide our experimental design. Comparison of multiple cell types on two different microscopes yie lded consistent measurements within each cell type and between microscopes and definitively detected cell type differences in network morphology. A utomation of the analysis precludes the need for software expertise on the part of the end user, and the meth ods developed for removing aberrantly detected mitochondria successfully removed mitochondrial bodies which were obviously detected incorrectly. A variety of experimental perturbations were used to test the ability of this software to track dynamic changes in network morphology, and dynamic changes were in fact detected in with each perturbation

PAGE 19

9 2. Methodological Design of Mitochondrial Measurement Software 2.1 Reagents Dulbecco's Modified Eagles Medium (DMEM) 5 mM and 25 mM glucose and non essential amino obtained from Thermo Scientific Hyclone, and trypsin and trypsin inhibitor were purchased from Fisher Scientific. Fetal bovi ne ser um (FBS) was procured from Gemini Bioproducts. Hank's Balanced Salt Solution (HBSS) was purchased from Corning Life Sciences. Secondary detection an tibodies Alex Fluor 488 and 546 were purchased from Life Technologies. Antibodies to TOM20 (rabbit) and nitrotyrosine (mouse) were procured from Santa Cruz Biotechnology. 2.2 Cell C ulture Primary rat vascular smooth muscle cells ( vSMC s ) were cult ured in low glucose (5 mM) DMEM with 10% fetal bovine serum (FBS) 1% L glutamine, 1% non essential amino acid blend, and 1% Pen/Strep, all expressed as % by volume. For vSMC starvation media, the same proportions were maintained, except that FBS was reduced to 0.1%. Human umbilica l vein endothelial cells ( HUVECs ) were cultured in F 12k medium (Hyclone #SH30526.01) with 10%

PAGE 20

10 FBS, 1% Pen/Strep, 0.05 mg/mL endothelial cell growth supplement, and 0.1 mg/mL heparin. SH SY5Ys (ATTC neuronal cell line) were cultured in 45% F 12k medium, 4 5% low glucose DMEM, and 10% FBS. All cells were cultured at 37 degrees Celsius at 5 % CO 2 Media was changed at least every three days, and cells were split 2:1 at 80% confluence. Experiments were performed on cells at 50% 70% confluence. Unless otherwise specified, cell culture materials were obtained from Santa Cruz Biotechnology. 2.3 Fixation and S taining Prior to fixation, phosphate buffer solution ( PBS ) rinse and paraformaldehyde solutions were warmed to 37 degrees Celsius ( except as otherwise indicated ) Samples were washed 3x in warm PBS prior to incubation in 4% paraformaldehyde at 37 degrees Celsius for 15 minutes. Following fixation, samples were washed 3x in warm PBS and quenched in 50 mM NH 4 Cl for 15 minutes at room temperature. Samples were then rinsed 3x with room temperature PBS and stored immersed in PBS at 4 degrees Celsius until staining. All samples were stained within one week of fixation. Prior to staining, cells were permeablized by incubation in 0.1% TX 1000 at 37 degre es Celsius for 15 minutes. Following permeabilization, samples were rinsed 3x with room temperature PBS and then blocked by incubation in 10% fetal bovine serum ( FBS ) in PBS for 25 minutes and 37 degrees Celsius. Samples were

PAGE 21

11 subsequently incubated in a pr imary antibody mix in 5% FBS in PBS at 37 degrees Celsius for one hour. Following treatment with primary antibody, samples were washed 3x with 5% FBS in PBS and incubated in a secondary antibody mix in 5% FBS in PBS at room temperature for one hour. Sample s were then washed 3x in PBS, incubated in DAPI diluted 1:10 000 in PBS at room temperature for 10 minutes, washed 3x in PBS again, then washed once in distilled water. Coverslips were then mounted to microscope slides for imaging. The primary antibody mix consisted of a Tom20 antibody (Santa Cruz Biotechnology sc 11415) diluted 400x to stain mitochondria and a nitrotyrosine antibody (Santa Cruz Biotechnology sc 32757) diluted 100x to stain the cytoplasm. The secondary antibody mix consisted of Alexafluor 4 88 (Life Technologies A 11008) diluted 1000x to react with the Tom20 antibody and Alexafluor 546 (Life Technologies A 11030) diluted 500x to react with the nitrotyrosine antibody. Following staining, slides were stored protected from light at 4 degrees Cel sius. 2.4 Microscopy and Imaging T echniques The primary imaging system used in these experiments was an Olympus FV1000 confocal microscope provided by the UC Denver Advanced Light Microscopy Core. A 488 nm laser was used to excite AlexaFluor 488 and a 543 nm laser was used to excite AlexaFluor 546. Built in filter settings for these

PAGE 22

12 fluorophores were used to detect fluorescence. Slides were imaged using a 60x oil immersion objective at a resolution of 206.8 nm/pixel and an image size of 1024 pixels b y 1024 pixels. An exposure time of 15 ms per pixel was used and the ave rage of three frame and line acquisitions each was used as the raw microscope image. In addition to the Olympus FV1000, further imaging was performed on a Leica SP8 confocal microscop e at the Denver VA Medical Center A white light laser was tuned to excitation wavelengths of 488 nm and 546 nm for AlexaFluor 488 and 546, respectively. Similar to images acquired on the Olympus system, a 60x oil immersion objective was used on the Leica system. In order to test the effects of poor image quality on algorithm performance images acquired on the Leica SP8 system used no averaging to reduce noise, an exposure time of 4.9 us per pixel was employed, and images were acquired at a resolution of 360.4 nm/pixel and an image size of 512 pixels by 512 pixels. Microscope settings for both sets of imaging experiments are included in Table 1 below. TABLE 1: MICROSCOPE SETTINGS USED IN THIS STUDY Olympus FV1000 Leica SP8 Image S ize (pixels) 1024x1024 512x512 Pixel S ize (nm) 206.8 360.4 Exposure Time (u s) 15 4.9 Averaging 3 frames none Numerical Aperture 1.4 1.4 Special Considerations none intentionally poor image quality

PAGE 23

13 2.5 Image Binarization for Computational M easurements Images were imported into Matlab in 16 bit TIF format and binarized using custom built software. The methods used to binarize the Tom20 (mitochondria) and nitrotyrosine (cytoplasm) channels are outlined in Figure 2 4x4 binning was applied to raw mitochondrial images to reduce the influence of noise. The noise reduced images were then normalized so that the minimum pixel intensity value was 0 and the maximum pixel intensity value was 1. Noting that the spac es between mitochondria in areas of high mitochondrial density were often brighter than were mitochondria themselves in areas of lower mitochondrial density, a Gaussian blur was applied to noise reduced mitochondrial images to obtain an approximation of ba ckground intensity. Equation 2. 1 was then used to compare noise reduced and background intensity images to identify regions of locally increased fluorescence intensity. In this equation, M refers to the noise reduced image, S refers to the background inten sity image, and P refers to the resulting intensity prominence image. [2.1 ]

PAGE 24

14 A ) FIGURE 2: METHOD USED TO BINARIZE RAW MICROSCOPE IMAGES ( A ) Binarization of mitochondrion channel. Raw microscope images are imported into Matlab in 16 bit TIF format. 4x4 binning is used to reduce the influence of machine noise. The noise reduced image is then compared to a Gaussian blur B)

PAGE 25

15 of the same image to det ermine intensity prominence. A threshold is then applied to the intensity prominence to detect mitochondria. ( B ) Binarization of cytoplasm channel. Cytoplasm is stained using a monoclonal antibody for nitrotyrosine. The raw microscope image is then blurred twice using a Gaussian blur, and the difference between the once blurred and twice blurred images The intensity prominence image obtained using Equation 1 was then processed using the thresholding alg orithm ou tlined in Equations 2 .2 2. 7 to maximize agreement between intensity prominence and the resulting binary image. T hreshold values between 0 and 1 were tested with a resolution of 0.02 and the image was then binarized at the threshold value with the lowest error score such that values above threshold equal 1 and those below threshold equal 0. In these equations, T is the binary image obtained at a given threshold value pixel refers to a pixel during single pixel operations, thresh refers to the threshold value being tested, Err1 is the component of the error score indicating background pixels detected as mitochondria, Err2 is the component of the error score indicating mitochondrial pix els detected as background, and Error is the combined error score. [2 .2 ] [ 2. 3]

PAGE 26

16 [ 2. 4] [ 2. 5] [ 2. 6] [ 2. 7] The nitrotyrosine (cytoplasm) channel initially appears as punctate dots of fluorescence throughout the cytoplasm. The process for binarizing this channel is outlined in Figure 1B. Prior to processing, images were normalized such that the minimum pixel intensity was 0 and the maximum pixel intensity was 1. A Ga ussian blur was used to transform the punctate distribution of fluorescence into a more uniform distribution of fluorescence within the cytoplasm. It was observed that more peripheral regions of the cytoplasm are often more similar in pixel intensity to ba ckground than to pixel intensity in perinuclear regions of the cytoplasm. A second Gaussian blur was therefore used to obtain an approximation of background pixel intensity, similar to the method employed in binarization of the mitochondrial channel. The d ifference

PAGE 27

17 in pixel intensities between the once smoothed and twice smoothed images was then used to construct an intensity prominence image. Threshold values between 0 and 1 were tested with a resolution of 0.02. The threshold value which minimized the rat io of edge length to background area was then selected to binarize the image. Finally, the DAPI (nucleus) channel was binarized directly by thresholding using the same algorithm as was used on the intensity prominence image from the mitochondrial channel. 2.6 Computation of Raw Mitochondrial Morphology P arameters Raw mitochondrial and cytoplasm images were binarized as described in the previous section s The number of illuminated pixels in each binary mitochondrial image was then multiplied by the area of a single pixel in m 2 to obtain a metric of mitochondrial network size. The number of illuminated pixels bordering non illuminated pixels was multiplied by the pixel edge length in m to obtain a metric of total mitochondrial network perimeter. The numb er of mitochondrial bodies was detected using a built in function in M atlab ( see bwconncomp() in M atlab documentation ). This function uses a 4 connected pixel neighborhood to assess connectivity and provides outputs including the total number of bodies and a list of pixels included in each body. Bodies comprising less than 15 um 2 were excluded from further analysis as too s mall and those for which the projected area to p erimeter ratio was greater than 0.3

PAGE 28

18 um were excluded from further analysis as too wide. These criteria were determined empirically to remove bodies which were obviously detected aberrantly and were used to p revent error in algorithm measurements due to faulty image binarization Using mitochondrial bodies that were not rejected according to these criteria average mitochondrion size was then taken to be the total mitochondrial network size divided by the number of mitochondrial bodies. The average mitochondrion perimeter was likewise taken to be the total mitochondrial network perimeter divided by the number of mitochondrial bodies. Next, average mitochondrion size and perimeter were used to calculate aver age mitochondrion length and width. Observing that most of the mitochondria in the network are shaped like long circular tubes with rounded ends, the geometric relationships depicted in Figure 2 A and Equations 2. 8 2. 9 below were used to solve for average l ength and width. In these equations, P is perimeter, A is area, W is width, and L is length. [ 2. 8 ] [ 2. 9 ]

PAGE 29

19 2.7 Correction of Raw Mitochondrial Morphology P arameters The raw parameters computed using the algorithms described in the previous section were subsequently corrected for total system error in measurement and for more general field of view parameters, such as total cytoplasm area and number of cells. Total system error comprises the microscope point spread fu nction (PSF), algorithm error in binarization of raw mitochondrial images, and discretization error in computational measurement of the se binarized mitochondrial images. Rather than attempt ing to measure or estimate each of these sources of error independe ntly, the magnitude of the total system error was directly measured as the difference between the known 400 nm diameter of fluorescent beads (CytoDiagnostics FBC 620 400) and their measured diameter using the same image processing scripts as were used to m easure mitochondrial network parameters. For both microscopes used, 400 nm beads were measured at concentrations spanning four orders of magnitude. Bead images were binarized using identical methods to those used to binarize mitochondrial images, and measu red bead diameter was inferred from total projected area and perimeter using similar methods to those outlined in Equations 8 9 except that a spherical rather than tubular shape was assumed. No significant differences were found between calculated bead di ameters when measured at different concentrations for either microscope system, indicating that clustering of beads did not co mpromise this analysis

PAGE 30

20 (Figure 3C ). It was found that the high resolution setup on the Olympus FV1000 microscope overestimated bead diameter by 170 nm, while the intentionally de tuned setup on the Leica SP8 overestimated bead diameter by 360 nm. These error values are similar to or small er than pixel size of 206.8 nm and 360.4 nm using the Olympus and Leica microscopes, respectively. The bulk measurement error values inferred from bead calibration thus represent statistical averages (i.e. either 0 or 1 superfluous pixel detected along eac h axis and the reported measurement error represents probability of a single superfluous pixel along each axis ) and are applicable to whole network but not single mitochondrion measurements. Raw length and width values were then corrected by subtracting th e bead diameter overestimation obtained using the imaging system and setup corresponding to each mitochondrial image. T otal cytoplasm area was estimated from the total number of illumined pixels in the binarized cytoplasm image multiplied by pixel are a in m 2 The total number of cells in the image was estimated from the number of bodies in the binarized nucleus channel using the same built in M atlab function as was used to count mitochondrial bodies The total projected area of the mitochondrial network was divided by the total cytoplasm area to obtain an estimate of the fraction of the cytoplasm filled by mitochondria as a normalized metric of mitochondrial mass. Likewise, the total projected area of the mitoc hondrial network and the total number of mitochondria were divided by

PAGE 31

21 the number of cells in the image to obtain an estimate of total mitochondrial mass per cell and total number of mitochondria l bodies per cell, respectively. For most inter group comparis ons, corrected parameters were used rather than raw parameters in order to reduce the influence of machine error, random differences in cell size or confluence between fields of view, etc. on reported end points.

PAGE 32

22 FIGURE 3 CALIBRATION USING FLUORESCENT BEADS OF KNOWN SIZE ( A ) Deterministic relationship between edge length, cross sectional area, and length and width. Assuming the typical mitochondrion to be shaped like an elongated tube, the average length and width in the n etwork can be calculated from the total network area, edge length, and number of mitochondria. ( B ) Representative images of fluorescent beads at 1000x and 100x minimum concentration. ( C ) Measuring the diameter of fluorescent beads. Using the same technique as previously described but modified to assume spherical, rather than tubular bodies. At bead concentrations spanning four orders of magnitude, calculated bead diameter remained constant at 570.1 nm using the Olympus microscope and 760 nm using the Leica microscope, compared to a true mean diameter of 400 nm. The total system bias of 170.1 nm and 360 nm for the Olympus and Leica images, respectively, was then subtracted from average mitochondrion length and width values to correct for the microscope PSF an d errors in image processing. A ) B) C)

PAGE 33

23 2.8 Post Processing of D ata Following initial measurements and corrections, outliers in the data set were automatically detected and removed by the software according to the following scheme: a ny image for which a ny parameter measured was greater than three standard deviations from the treatment group mean was removed from the data set entirely. Under this scheme, images with unusually sparse mitochondria would not be analyzed for mitochondrial length, images with an unusual number of cells would not be analyzed for mitochondrial density, etc. Remaining images were then used to calculate treatment group means and SEMs for each experimental repetition and across all experimental repetitions. In order to verify that thi s mode of data grooming was not limiting sample size, the total number of images included and rejected in each analysis was monitored Less than 10% of images were removed from analysis in all image sets analyzed. 2.9 Statistics and Software P ackages For comparisons involving three or more groups, a one way ANOVA was used to assess statistical significance. For comparisons involving two groups only, an unpaired t test was used to assess statistical significance. To assess correlations and their statist corresponding t value were computed. Effects for which p<0.1 were considered

PAGE 34

24 trending and those for which p<0.05 were considered significant. For comparisons in which one way ANOVA was employed, test was used to correct for the effects of multiple comparisons. All post hoc statistical analyses were performed in Prism GraphPad version 5.0. The mitochondrial measurement algorithms and automated post processing were coded into Matlab 2013. Where available, bu ilt in functions in these software packages were used for statistical computations.

PAGE 35

25 3. Internal Validation of Mitochondrial Measurement Software 3.1 Baseline Cell Line C haracteristics This section documents basic mitochondrial network characteristics of GK vSMC s, Wistar vSMC s, HUVECs, and SH SY5Ys with no special treatment prior to fixation. Figure 4 summarizes differences between these four cell lines Consistent with visual inspection, HUVECs were found to have the longest, thinnest, and least densely packed mitochondria, SH most densely packed mitochondria and Wistar vSMCs the widest mitochondria Values obtained for average (+/ SD) mitochondrion length were 11.35 +/ 2.41 um, 6.80 +/ 0.86 um, 5.09 +/ 0.77 um, and 4.31 +/ 0.66 um for HUVECs, GK vSMCs Wistar vSMCs and SH SY5Ys, respectively. Values obtained for average mitochondrion width were 272 +/ 24 nm, 315 +/ 20 nm, 351 +/ 23 nm, and 320 +/ 14 nm for HUVECs, GK vSMCs Wistar vSMCs and SH SY5Ys, respectively. Values obtained for fraction of cytopla sm filled by mitochondria were 9.6 +/ 2.0 %, 16.0 +/ 1.8 %, 15.8 +/ 1.7 %, and 23.9 +/ 1.8 % for HUVECs, GK vSMCs Wistar vSMCs and SH SY5Ys, respectively. In addition to these characteristics of the mitochondrial networks, the fraction of each mitochondrial network detected correctly was es timated by eliminating detected

PAGE 36

26 m itochondrial bodies that were excessively small or excessively wide. The fraction of the detected mitochondrial network included i n morphological analysis was 85.7 +/ 6.8 %, 85.1 +/ 5.7 %, 64.6 +/ 15.7 %, and 17.1 +/ 6.8 % for HUVECs, Wistar vSMCs GK vSMCs and SH SY5Ys, respectively. These values are also listed in Table 2 below. TABLE 2 : BASELINE CELL LINE CHARACTERISTICS HUVEC GK vSMCs Wistar vSMCs SH SY5Y Mean Mitochondrion Length (mean+/ SD) 11.35+/ 2.41 um 6.80+/ 0.86 5.09+/ 0.77 um 4.31+/ 0.66 um Mean Mitochondrion Width (mean+/ SD) 272+/ 24 nm 315+/ 20 nm 351+/ 23 nm 320+/ 14 nm Fraction of Cytoplasm Filled by Mitochondria 9.6+/ 2.0% 16.0+/ 1.8% 15.8+/ 1.7% 23.9+/ 1.8% Fraction of Mitochondrial Mass Included in Analysis 85.7+/ 6.8% 85.1+/ 5.7% 64.6+/ 15.7% 17.1+/ 6.8%

PAGE 37

27 FIGURE 4: BASELINE CELL LINE CHARACTERISTICS ( A ) Representative microscope images of cell types used in this study. Mitochondria were stained using a Tom20 antibody, cytoplasm using a nitrotyrosine antibody, and nuclei using DAPI. ( B ) Average mitochondrion length by cell type. HUVECs had the longest mitochondria at 11. 5 um while SH SY5Ys had the shortest at 4.31 um. Each cell type displayed a highly significantly different (p<0.0001) average mitochondrion length from every other cell type. ( C ) Average mitochondrion width by cell type. Wistar vSMCs had the widest mitocho ndria at 351 nm while HUVECs had the thinnest at 272 nm. All cell type differences were highly significant (p<0.0001), except for GK vSMCs and SH SY5Ys, which did not differ significantly (p>0.05). ( D ) Mitochondrial mass by cell type. SH SY5Ys had the grea test fraction of the cytoplasm filled by mitochondria at 23.9% while HUVECs had the least at 9.6%. All cell type differences were highly significant (p<0.0001), except for GK vSMCs and Wistar vSMCs, which did not differ significantly (p>0.05). ( E ) Fraction of mitochondrial network correctly detected by cell type. HUVECs and GK vSMCs had the best coverage at 85.7% and 85.1%, respectively while SH SY5Ys had the worst at 17.1%. All cell type differences were highly significant (p<0.0001), except for Wistars an d GK vSMCs, which did not differ significantly (p>0.05). A ) B) C) D) E)

PAGE 38

28 3.2 P erformance of Custom A lgorithms and Simple T hresholding This section documents the differences between results obtained using the novel algorithm s described in the methods section and results obtained using simple thresholding. HUVEC and SH SY5Y images were used in this analysis in order to maximize the range of values obtained Figure 5 summarizes differences between results obtained from the algorithm developed in this manuscript and application of simple thresholding. Our scheme for simple thresholding involved applying the algorithm described in the methods section directly to raw mit ochondrial images rather than the pre processed algorithm defined intensity prominence images (Figure 2A) S imple thresholding is comparable to predicate methods using simple thresholding such as the Mitochondrial Morphology plugin in ImageJ [ 27 28 29 ] exce pt that the threshold is automatically determined rather than user defined. A greater dynamic range of values for mitochondrial length was recorded using the custom algorithm than using simple thresholding (algorithm range: 3.12 um 20.4 um, threshold range 2.47 um 9.14 um, significant reduction in variance using simple thresholding, p<0.0001 using an F test) and measured mitochondrial lengths were significantl y smaller when measured using simple thresholding (p<0.0001 using unpaired T test with Welch HUVECs and SH SY5Ys). In addition, the fraction of the mitochondrial network detected adequately for inclusio n in subsequent analyses was significantly

PAGE 39

29 reduced with simple thresholding (p<0.0001 for both cell types ). Noting that the portion of the network automatically discarded from further analysis using simple thresholding was primarily perinuclear, the measurements obtained using simple thresholding were compared to measurements of the outermost 10% of mitochondrial mass using the custom algorithm. Under this comparison, no differences in the mean or variance (p>0.05 for both ) were found between algorithm and threshold ing groups for either cell type, consistent with a bias towards shorter, less variable mitochondria due to exclusion of perinuclear mitochondria from subsequent analysis.

PAGE 40

30 FIGURE 5: PERFORMANCE OF CUSTOM ALGORITHM RELATIVE TO T HRES HOLDING ( A ) Mean mitochondrion length as measured using the custom algorithm vs simple thresholding. Dynamic range of lengths measured is greater using the custom algorithm, and lengths measured using thresholding are routinely less than those measured using the custom algorithm. ( B ) Mean mitocho ndrion length in the outermost 10% of the mitochondrial network as measured using the custom algorithm vs mean mitochondrion length throughout the entire network as measured using thresholding. Although the agreement is imprecise, dynamic range is similar between the two metrics and no bias is observed. ( C ) Examples of images binarized using the custom algorithm and simple thresholding. Portions of the mitochondrial network that were not detected correctly are colored purple. A much greater portion of the n etwork is excluded from analysis using simple thresholding and the portions of the network excluded are largely perinuclear. ( D ) Fraction of the mitochondrial network detected correctly by cell type and algorithm. Regardless of detection quality using the custom algorithm, detection quality using simple thresholding is substantially lower (p<0.0001 for both HUVECs and SH SY5Ys) A ) B) C) D)

PAGE 41

31 3.3 C onsistency of M easure s Between Different M icroscope s This section documents the similarities and differences between results obtained on different microscopes. HUVEC and SH SY5Y samples were imaged on both an Olympus FV 1000 and a Leica SP8. Images were obtained at 1024x1024 pixels on the Olympus microscope and at 512x512 pixels on the Leica microscope. In addition, expo sure time was limited and lasers intentionally de tuned on the Leica microscope in order to obtain grainy, low resolution images for comparison with the clear, high resolution images taken on the Olympus microscope. Mean values for each experimental repeti tion were compared between the microscope setups. Figure 6 summarizes comparison s of mitochondrial network parameters between the same samples measured on different microscopes Measurements of average mitochondrion length and average mitochondrion width were significantly correlated between the two imaging systems (R=0.82 and R=0.73, length and width respectively, p<0.0001 for both). A bias towards measurement of longer mitochondria was observed using the low resolution system, with mitochondrion length a veraging 3.96 um (~40%) longer when measured using the low resolution system (p<0.0001 using paired t test) compared to a ~100 nm pixel resolution difference Given that the same method of calibration was used for both imaging systems (see Correction of r aw mitochondrial morphology parameters above) it is likely that this bias was a

PAGE 42

32 direct consequence of using lower resolution images. We observed that fragmented mitochondria were often arranged collinearly and were very clos e together, as shown in Figure 6 E. Given that the gaps between mitochondria were often only one or two pixels wide in the high resolution images, it is likely that collinear fragmented mitochondria were detected as single mitochondrial bodies using the low resolution images. No bias was observed in mitochondrial width measurements, but the correlation between values obtained on different imaging systems was less robust than for mitochondrial length measurements.

PAGE 43

33 FIGURE 6: CONSISTENCY OF MEASURES USING DIFFERENT MICROSCOPES ( A ) Representative image captured using high resolution settings on the Olympus microscope. ( B ) Representative image captured using intentionally low resolution settings on the Leica microscope. These images were captured at 1/ 4 th the pixel resolution and less than 1/10 th the exposure time of the high resolution images. ( C ) Comparison of mitochondrion length measurements between microscopes. The experimental repetition means of mitochondrial length were closely correlated betwee n the two microscopes (R=0.82), but there was a bias towards longer measurements using the low resolution system. Error bars represent SEM in measurements from each microscope. ( D ) Comparison of mitochondrion width measurements between microscopes. The exp erimental repetition means of mitochondrial width were modestly correlated between the two microscopes (R=0.73). No bias was observed in mitochondrial width measurements. ( E ) Representative high resolution microscope image showing collinear mitochondrial f ragments. Such fragments were likely indistinguishable using the low resolution images, which would account for the bias towards longer mitochondria with the low resolution system. A ) B) C) D) E)

PAGE 44

34 3.4 Consistency of M easure s B et ween Experimental R epetitions For all sample groups imaged, the first repetition of each experiment was compared to the second repetition and the second was compared to the third for each sample group metric Images from the Olympus FV1000 microscope and the Leica SP8 microscope were analyzed separately to allow assessment of experimental consistency within and between image sets acquired using different imaging systems. Figure 7 summarizes assessments of consistency between experimental repetitions within treatment groups Treatment g roup differences were observed to be qualitatively consistent between ex perimental repetitions (Figure 7 A) M easured values of mitochondrial length for each group were consistently similar between experimental repetitions, with R=0.96, p<0.0001 using the high r esolution image set and R=0.70, p= 0.005 using the low resolution image set (Figure 7 B C). Mean mitochondrion width was also consistent between exper imental repetitions, with R=0.77, p<0.0001 using the high resolution image set and R=0.84 p =0.0002 using th e low resolution image set (Figure 7 D E). Although substantial variance was observed between experimental repetitions, in most cases this variance was minor relative to treatment group differences.

PAGE 45

35 FIGURE 7: CONSISTENCY OF EXPERIMENTAL REPETITIONS ( A ) Time course of mean mitochondrion length in Wistar vSMCs exposed to 25 mM glucose. Individual experimental repetitions (n=9 images) are shown in red, while the overall average (n=27 images) is shown in black. ( B ) Agreement between repetitions in mitochondrion length using high resolution microscope settings. Very strong agreement is observed between successive experimental repetitions (R=0.96). ( C ) Agreement between repetitions in mitochondrion length using low resolution mi croscope settings. Modest agreement is observed between successive experimental repetitions (R=0.70). ( D ) Agreement between repetitions in mitochondrion width using high resolution microscope settings. Strong agreement is observed between successive experi mental repetitions (R=0.77). ( E ) Agreement between repetitions in mitochondrial width using low resolution microscope settings. Strong agreement is observed between successive experimental repetitions (R=0.84). A ) B) D) C) E)

PAGE 46

36 4. Cellular Perturbation Test Cases 4.1 Effects of Fixation Temperature in HUVEC C ells To assess the impact of the fixation conditions on mitochondrial network parameters, we tested the impact s initial reagent temperature s and incubation temperature during fixation on mitochondrial morphology HUVEC cells were cultured without experimental perturbation for 6 10 passages. C ells were divided into one of four experimental groups and fixed according to the protocol described above with specific adjustments as outlined be low Group 1 was fixed using reagents pre heated to 37 degrees Celsius and all incubation steps during fixation were performed at 37 degrees Celsius ; Group 2 was fixed using room temperature reagents and all incubation steps during fixation were performed at 37 degrees Celsius ; Group 3 was fixed using room temperature reagents and all incubation steps during fixation were performed at room temperature ; Group 4 was fixed using reagents pre heated to 37 degrees Celsius and all incubation steps during fixation were performed at room temperature. Figure 8 summarizes the effects of fixation temperature on mitochondrial morphology in HUVEC cells Visual observation of the samples indicated a more fragmented mitochondrial network in the group treated with room

PAGE 47

37 temperature formaldehyde and incubated at room temperature during fixation and possibly a more peripheral distribution of mitochondrial mas s in the group treated with pre warmed formaldehyde and incubated at 37 degrees Celsius during fixation co mpared to other groups (Figure 8 A). It was found that the average mitochondrion length was significantly (p<0.0001) reduced in the group treated with room temperature formaldehyde and incubated at room temperature during fixation relative to every other group, and that no other treatment group differences in mitochondrial length were statistically sig nificant (Figure 8 B). Group means for mitochondrion l ength were 11.35+/ 2.41 um, 11.01+/ 2.15 um, 8.16+/ 2.28 um, and 12.58+/ 3.73 um for groups 1, 2, 3, and 4, respectively. No group differences were found in avera ge mitochondrion width (p>0.05) Although visual analysis suggested perinuclear localization o f mitochondria in all groups treated with room temperature reagents or incubated at room temperature at any stage during fixation the fraction of the 10% of the cytoplasm closest to the nucleus filled by mitochondria did not vary significantly (p>0.05). Group means for mitochondrion width were 272+/ 24 nm, 286+/ 36 nm, 280+/ 28 nm, and 289+/ 32 nm for groups 1, 2, 3, and 4, respectively. Finally, no group differences were found in total fraction of the cytoplasm filled by mitochondria, as would be expecte d given that the time required for fixation is considerably less than the time required for substantial mitochondrial turnover. The means of reported endpoints are included in Table 3 below.

PAGE 48

38 TABLE 3 : EFFECTS OF TEMPERATURE DURING FIXATION ON MITOCHONDRIAL MORPHOLOGY Mean Mitochondrion Length (mean+/ SD) Mean Mitochondrion Width (mean+/ SD) Group 1 11.35+/ 2.41 um 272+/ 24 nm Group 2 11.01+/ 2.15 um 286+/ 36 nm Group 3 8.16+/ 2.28 um 280+/ 28 nm Group 4 12.58+/ 3.73 um 289+/ 32 nm FIGURE 8: EFFECTS OF TEMPERATURE DURING FIXATION ON MITOCHONDRIAL MORPHOLOGY HUVEC cells were fixed using either body temperature (Group 1) or room temperature (Group 2) reagents and subsequently incubated at either body temperature (Group 3) or room temp erature (Group 4) for fifteen minutes during fixation. ( A ) Representative images of mitochondrial networks following each treatment. Some rearrangement of the mitochondrial network is apparent between any two experimental cases. Cells fixed with room tempe rature reagents and incubated A ) B) C)

PAGE 49

39 at room temperature (CC) show visible mitochondrial fragmentation. ( B ) Mean mitochondrion length by treatment. Cells fixed with room temperature reagents and incubated at room temperature (CC) show a highly significant (p<0.00 01) reduction in mean mitochondrion length compared to other treatments, which do not differ significantly. ( C ) Mean mitochondrion width by treatment. No temperature effects on mitochondrial width were observed. 4.2 High Glucose E xposure in Primary R at vSMC s Primary smooth muscle cells from the aorta of GK or Wistar rats were cultured in low glucose (5 mM) smooth muscle growth media until passages 8 10. Prior to treatment cells were switched to low glucose, serum free starvation media for 48 hours. Ser u m starved cells were exposed to high glucose (25 mM), serum free media. For each cell type (GK and Wistar) in each repetition of each experiment, 15 slides were prepared. Of these 15 slides, three were merely rinsed with high glucose media before fixation and staining (0 minute exposure), three were fixed and stained after 10 minutes in high glucose media, three were fixed and stained after 30 minutes in high glucose media, three were fixed and stained after 30 minutes in high glucose media, three were fix ed and stained after 60 minutes in high glucose media, and the remaining three were fixed and stained after 240 minutes in high glucose media. These time points were selected based on empirical observation that mitochondrial morphology was highly dynamic i n the first hour of high glucose exposure and then underwent more gradual changes in the subsequent several hours.

PAGE 50

40 Figure 9 summarizes the response of GK and Wistar vSMC s to exposure to 25 mM glucose exposure. Dynamic changes in mitochondrial morphology were not apparent to the naked eye, although the greater mitochondrial length and lesser mitochondrial width among GK vSMC s relative to Wistar vSMC s at baseline (Figure 4 A) were apparent at all time points Altho ugh subtle, d ynamic changes detected using our algorithm were consistent between ex perimental repetitions (Figure 7 A). Wistar vSMC s demonstrated a significant decrease in average mitochondrion length from 10 to 30 minutes (p<0.0001) followed by a signific ant (p<0.005) increase from 60 to 240 minutes (Figure 9 A) No time point changes in average mitochondrion length were statistically significant among GK vSMC s, but trending (p< 0.1) decreases and increases were observed at the same time points as were significant for the Wistar vSMCs In addition to these differences in dynamic changes in length, GK vSMC s had significantly (p<0.0001) longer mitochondria than did Wistar vSMC s at a ll time points recorded. Similarly, although average mitochondrion width did not change significantly over time (p>0.05) for either cell type, Wistar vSMC s had significantly (p<0.0001) wider mitochondria than did GK vSMC s at all time points recorded (Figur e 9 C) with overall mean values of 309+/ 26 nm for GK vSMCs and 347+/ 22 nm for Wistar vSMCs. Recorded mean values for mitochondrion length were 6.80+/ 0.86 um at baseline, 6.78+/ 0.80 um at 10 minutes, 6.35+/ 0.79 um at 30 minutes, 6.44+/ 0.78 um at 60

PAGE 51

41 minutes, and 6.77+/ 0.83 um at 240 minutes for GK vSMCs, and 5.09+/ 0.77 um at baseline, 5.28+/ 0.59 um at 10 minutes, 4.65+/ 0.54 um at 30 minutes, 4.64+/ 0.52 um at 60 minutes, and 5.86+/ 0.44 um at 240 minutes for Wistar vSMCs. I n addition, the fractio n of the 10% of the cytoplasm closest to the nucleus filled by mitocho ndria was investigated (Figure 9 B). Wistar vSMCs showed a statistically significant increase (p<0.05) in the fraction of the perinuclear space filled by mitochondria from 60 to 240 minut es. GK v SMC s showed no dynamic response (p>0.1 for all comparisons) in this parameter, and the fraction of the perinuclear space filled by mitochondria in GK v SMC s was significantly greater (p<0.0001) than in Wistar v SMC s at all time points measured. Overall mean values for fraction of perinuclear space filled by mitochondria were 11.8+/ 1.9% at baseline, 10.7+/ 1.7% at 10 minutes, 11.0+/ 2.0% at 30 minutes, 11.0+/ 2.4% at 60 minutes, and 11.1+/ 3.2% at 240 minutes for GK vSMCs, and 8.2+/ 2.4% at base line, 7.9+/ 1.6% at 10 minutes, 7.9+/ 1.8% at 30 minutes, 8.0+/ 1.7% at 60 minutes, and 9.3+/ 2.1% at 240 minutes for Wistar vSMCs.

PAGE 52

42 FIGURE 9: MITOCHONDRIAL NETWORK RESPONSE TO INCUBATION IN HIGH GLUCOSE ( A ) Mean mitochondrion length vs time in 25 mM glucose. Mean mitochondrion length was much greater in GK vSMCs than in Wistar VSMCs at all time points measured (p<0.0001). Wistar vSMCs displayed a highly dynamic response to glucose exposure including significant fragmentation between 10 a nd 30 minutes (p<0.0001) and a significant increase in networking between one hour and four hours (p<0.01). Although no changes in average length were statistically significant in the GK vSMCs, there was a trend towards qualitatively similar changes in net work morphology at the same time points (p<0.1 for both). ( B ) Fraction of the 10% of the cytoplasm closest to the nucleus filled by mitochondria vs time in 25 mM glucose. No significant changes in perinuclear mitochondrial mass were observed in GK vSMCs. A significant increase in perinuclear mass between one and four hours was observed in Wistar vSMCs, however (p<0.05). ( C ) Mean mitochondrion width vs time in 25 mM glucose. No significant effects of high glucose on mitochondrial width were observed. Wistar VSMC mitochondria were much wider than GK vSMCs mitochondria at all time points measured (p<0.0001). B) A ) C)

PAGE 53

43 4.3 Effects of Hydrogen Peroxide E xposure with SH SY5Y N eurons Oxidative stress has been reported to perturb mitochondrial morphology. These experiments examined the impact of H 2 O 2 in a neuronal cell line. SH SY5Y neuronal cells were cultured without experimental perturbation for a minimum of 5 passages. C ells were then changed to media containing either 200 nM or 0 nM H 2 O 2 for 2 hours, at least 24 hours after the previous media change. Cells were then fixed and stained as described above. Figure 10 summarizes the effects of hydrogen peroxide exposure on mitochondrial morphology in SH SY5Y cells. Mitochondria from c ells exposed to hydrogen peroxide were found to be visibly more fragmented than control s although the visible difference was subtle (Figure 10 A). This difference was reflected in algorithm ic m easurements, which indicated a significant (p<0.0001) reduction in average mitochondrion length with hydr ogen peroxide exposure (Figure 10 B). In addition, a trending increase (p<0.1) in average mitochondrion width was recorded with hydrogen peroxide exposure, possibly indicati ng mitochondrial swelling. Mean values for mitochondri on length were 4.31+/ 0.66 um at baseline and 3.75+/ 0.41 um following hydrogen peroxide exposure, while those for mitochondrion width were 320+/ 14 nm at baseline and 327+/ 11 nm following hydrogen peroxide exposure. These values are also included in Tabl e 4 below.

PAGE 54

44 TABLE 4 : EFFECTS OF HYDROGEN PEROXIDE ON MITOCHONDRIAL MORPHOLOGY Mean Mitochondrion Length (mean+/ SD) Mean Mitochondrion Width (mean+/ SD) Control 4.31+/ 0.66 um 320+/ 14 nm 200 nM H 2 O 2 3.75+/ 0.41 um 327+/ 11 nm FIGURE 10: EFFECTS OF HYDROGEN PEROXIDE ON MITOCHONDRIAL MORPHOLOGY ( A ) Representative images before and after peroxide treatment. Mitochondria were visibly fragmented following treatment, but changes were difficult to observe due to the very small size of the cells. ( B ) Mean mitochondrion length before and after peroxide treatment. Hydrogen peroxide induced a highly significant decrease in mean mitochondrion length (p<0.0001). ( C ) Mean mitochondrion width before and after peroxide treatment. Although the effect was very subtle, a trending increase in mitochondrial width was observed following hydrogen peroxide treatment (p<0.1), possibly indicating mitochondrial swelling. A ) B) C)

PAGE 55

45 5. Exporting this Technology for Use by Other Laboratories 5.1 Softwar e Testing with Independently P roduced I mages If this technology is to be exported to other laboratories, it is imperative that investigators other than the PI of this study be able to produce mitochondrial network images suitable for analysis w ith this software. The staining protocol documented under 2.3 Fixation and staining was implemented by another technician within our laboratory and applied to HUVEC cells either in baseline conditions or following 1 or 4 hours of incubation in conditioned media previously used to culture Wistar or GK vSMCs. Thus there were a total of 5 groups: control, 1 hour exposure to GK media, 4 hour exposure to GK media, 1 hour exposure to Wistar media, and 4 hour exposure to Wistar media. All groups were imaged on the Leica microscope by a third independent technician. The bead calibration technique documented under 2.7 Correction of raw mitochondrial morphology parameters was not employed in this experiment, and so equivalent measures of HUVEC baseline parameter s are somewhat different between this experiment and those documented under 3.1

PAGE 56

46 Baseline cell line characteristics and Effects of fixation temperature in HUVEC cells. Note that although there are certain commonalities between this experiment and the previ ously discussed glucose exposure time course ( 4.2 High Glucose E xposure in Primary R at vSMC s ), the data presentation in this section has been altered to show time in hours rather than minutes (because no time points less than 1 hour were employed) and to u se a bar graph rather than a line graph, reflecting the discrete rather than continuous nature of this experiment. The results of this experi ment are summarized in Figure 11 Representative images from e ach group are shown in Figure 11 A. All images appeare d to have been binarized s uccessfully (Fig. 11 B ) and the fraction of the network detected suitably for subsequent analysis was consistent with that in previous analyses of HUVEC cells (mean=0.96 ), indicating that the staining protocols document under 2.3 F ixation and staining can be implemented successfully by independent investigators. Furthermore, this experiment was able to reveal that GK vSMC conditioned media triggered mitochondrial fusion while Wistar vSMC conditioned media did not affect mitochond rial networking (p<0.01, Fig. 11 C), GK vSMC conditioned media triggered mitochondrial swelling while Wistar vSMC conditioned media did not affect mitocho ndrial diameter (p<0.05, Fig. 11 D), and Wistar vSMC conditioned media triggered an increase in the frac tion of the perinuclear space filled by mitochondria while GK

PAGE 57

47 vSMC conditioned media did not affect intracellular distribution of mitochondria (p<0.01, Fig. 11 E). All three of these group differences are consistent with direct visual observations by the te chnician s who performed the staining and imaging. The finding of mitochondrial swelling with GK vSMC conditioned media was the first (and to date, the only) instance in which this software found statistically significant differences in mitochondrial width between groups of the same cell line. Mean values for mitochondrion length were 17.25+/ 5.67 um at baseline, 20.41+/ 6.37 um following 1 hr in GK conditioned media, 21.82+/ 3.05 um following 4 hrs in GK conditioned media, 16.51+/ 4.36 um following 1 hr in Wistar conditioned media, and 15.97+/ 4.39 um following 4 hrs in Wistar conditioned media. Mean values for mitochondrion width were 555+/ 76 nm at baseline, 544+/ 88 nm following 1 hr in GK conditioned media, 630+/ 103 nm following 4 hrs in GK conditioned media, 561+/ 97 nm following 1 hr in Wistar conditioned media, and 606+/ 78 nm following 4 hrs in Wistar conditioned media. Mean values for fraction of perinuclear space filled by mitochondria were 10.9+/ 3.9% at baseline, 10.8+/ 3.2% following 1 hr in GK conditioned media, 11.6+/ 2.6% following 4 hrs in GK conditioned media, 1 hr 10.1+/ 2.3% following 1 hr in Wistar conditioned media, and 12.5+/ 4.4% following 4 hrs in Wistar conditioned media. These values are also included in Table 5 below.

PAGE 58

48 TABLE 5 : EFF ECTS OF VSMC CONDITIONED MEDIA ON MITOCHONDRIAL MORPHOLOGY Mean Mitochondrion Length (mean+/ SD) Mean Mitochondrion Width (mean+/ SD) Fraction of Perinuclear Space Filled by Mitochondria (mean+/ SD) Control 17.25+/ 5.67 um 555+/ 76 nm 10.9+/ 3.9% GK 1 hr 20.41+/ 6.37 um 544+/ 88 nm 10.8+/ 3.2% GK 4 hrs 21.82+/ 3.05 um 630+/ 103 nm 11.6+/ 2.6% Wistar 1 hr 16.51+/ 4.36 um 561+/ 97 nm 10.1+/ 2.3% Wistar 4 hrs 15.97+/ 4.39 um 606+/ 78 nm 12.5+/ 4.4% FIGURE 11: EFFECTS OF VSMC CONDITIONED MEDIA ON MITOCHONDRIAL MORPHOLOGY ( A ) Representative images from each treatment group. Subtle differences were visible between treatment groups, but heterogeneity within each group was dominant relative to group differences. ( B ) Represen tative comparison of raw and binarized images. All images in this independently produced data set were binarized successfully. ( C ) Mean A ) B) C) D) E)

PAGE 59

49 mitochondrion length by treatment group. GK media triggered mitochondrial fusion relative to Wistar media (p<0.01). ( D ) Mean mitochondrion width by treatment group. GK media triggered mitochondrial swelling (p<0.05). ( E ) Fraction of the perinuclear space filled by mitochondria by treatment group. Wistar media triggered perinuclear gathering of mitochondria (p<0.01). 5.2 So ftware Testing with an Uncontrolled Image S ample As a litmus test of general applicability, the performance of the algorithms reported in this manuscript was tested on an uncontrolled sample of mitochondrial network images downloaded from the internet. The first ten results of a Google Images search for as of 8 25 2015 were downloaded and the mitochondrial channel of each image isolated. No attempt was made to determine the staining protocol, cell type, microscope, or laser lines used for any of these images; rather the purpose of this image sample was to demonstrate that virtually any high quality microscope image could be used regardless of preparation. Because the resolution of these images was not known, the pixel wise correlation coefficient en raw and binarized images (see Figure 1A) was used as a metric of binarization quality rather than attempting to infer the fraction of the network validly detected, which requires knowledge of physical pixel dimensions. This analysis thus assesses pre pr ocessing of images, but not post processing and data extraction. Given that the latter two steps are microscope independent, we refer the reader to previous sections

PAGE 60

50 documenting consistency and efficacy of software performance in various cellular perturbat ion test cases. Results of this analy sis are illustrated in Figure 12 When applied to baseline cell images (the same image set as was used for section 3.1 Baseline cell line characteristics ), the correlation coefficient between raw and binarized images an d fraction of the mitochondrial network validly detected were strongly associated (R=0.99), indicating that fraction of network validly detected and correlation strength between raw and binarized images function as equivalent measures of binarization quali ty. When applied to the uncontrolled image sample used for this litmus test, the mean correlation coefficient between raw and binarized images was 0.75 +/ 0.05 (mean +/ sd), comparable to the quality of detection in unperturbed HUVEC cells. The highest a nd lowest quality binariz ed images are shown in Figure 12B C

PAGE 61

51 FIGURE 12: ALGORITHM PERFORMANCE WITH AN UNCONTROLLED IMAGE SAMPLE ( A ) Similarity between pixel intensity correlation and network coverage metrics of binarization quality. Each metric can be almost entirely predicted by the other, but network coverage provides a slightly higher dynamic range. ( B ) Highest quality of binarization in the uncontrolled image sample; correlation between raw and binary images R=0.83. ( C ) Lowest qua lity of binarization in the uncontrolled image sample; correlation between raw and binary images R=0.66. A ) B) C)

PAGE 62

52 6. Conclusions and Future Directions 6.1 Discussion of R esults In light of the tremendous quantity of research in recent years assessing mitochondrial morphology endpoints, it is imp ortant that quantitative, objective, practical and consistent methods of obtaining mitochondrial morphology measurements are developed. The algorithms documented in this manuscript show excellent consistency in measured endpoints between experimental repetitions and additional consistency in measured endpoints between different microscopes Our results indicate slightly better performance in length measurements than in width measure ments. The observation of consistent results from one repetition to the next and between different microscopes hold s promise for enhanc ing consistency of results between different laboratories In addition, t he binarization algorithms described in this study demonstrated a quantitative advantage of over a simple thresholding algorithm by increasing in the fraction of the mitocho ndrial network detected suitably for analysis allowing simultaneous measurement of perinuclear and peripheral mitochondria Giv en that perinuclear and peripheral mitochondria are subject to different regulation [ 31 32 33 ] any method for assessing mitochondrial morphology that excludes

PAGE 63

53 a particular portion of the network is likely to introduce bias in whole cell measurements Finally, the differences between cell types and treatment groups detected by this softwar e uniformly agree with subjective visual assessment of mitochondrial morphology from raw m icroscope images. Collectively our results indicate that the se methods enable quantitative, objective, and consistent measurements of mitochondrial morphology. Further evidence for this method s utilit y can be found in comparisons of our experimental results to published observations A number of studies in a variety of cell types have reported mitochondrial fragmentation upon acute exposure to high concentration s of glucose [ 16 34 35 ] Yu et al report results of particular interest, showing a similar two phase mitochondrial morphology response in neuronal cells to that which we observed in primary vSMCs consisting of rapid mitochondrial fission followed by gradual mitochondrial fusion [ 16 ] Finally, a number of studies in a variety of cell types report mitochondrial fragmentation upon acute exposure to hydrogen peroxide [ 36 37 38 ] echoing our own results in SH SY5Y neurons. Our findings t hat an unsupervised algorithm was able to recreat e these results indicates that subjective or user guided assessment of mitochondrial morphology is no longer necessary. T he quantitative values for mitochondrial morphological parameters obtained in this study are realistic. M itochondrial morphology studies reporting

PAGE 64

54 direct measurement of mitochondrial length typically re port values in the range of 1 15 um [ 39 40 41 42 43 44 45 46 ] consistent with th e values obtained in this study. L iterature values for mitochondrial wid th /diameter generally fa ll in the range of several hundred nanometers [ 47 48 49 50 51 ] also consistent with those obtained in this study. Moreover, in studies that measure dynamic changes in both length and width, width has been reported to be far more stable than length [ 47 50 ] agreeing with our finding s of many statistically significant changes of average mitochondrion length but none in average mitochondrion width. In fact, the only arguable exception to stability in mitochondrial width in this study was a trending increase (p<0.1) in mitochondrial width in SH SY5Ys expo sed to 200 nM hydrogen peroxide. O xidative stress has previously been shown to induce mitochondrial swelling [ 52 53 54 55 ] hin ting that the algorithms in this paper might be able to detect mitochondrial swelling under certain conditions T he results obtained using our algorithms hold important methodological implications for future studies of mitochondrial morphology First, the highest resolution microscope available should be used for measurements of mitochondrial morphology. We observed collinear arrangement of fragmented mitochondria in several cell types, and these small, closely packed mitochondrial bodies were not successfully distinguished using the lower resolution of our two microscope setups E ven with the higher resolution setup,

PAGE 65

55 SH SY5Y mitochondria were poorly detected indicating that this algorithm may not be as well suited to smaller cells The correla tion strength between high and low resolution image results was similar to the inter experimental agreement of low resolution images but lower than that of high resolution images (Figures 5 6), indicating that measurement consistency in the low resolution image set was the limiting factor to agreement with measurements in the high resolution image set. Similar dynamic changes in mitochondrial morphology were observed with both imaging systems and mitochondrial fragmentation upon hydrogen peroxide exposure w as detectable, however, so use of a relatively low resolution system with these algorithms may be adequate to detect differences between treatment groups in the same experiment, but not for comparisons between studies or for absolute measurements. Likewise the use of fluorescent beads to calibrate the measurement software is necessary to remove bias and enable comparisons between different microscopes or papers but not necessary for distinguishing differences between treatment groups within the same study Another interesting methodological finding is that fixation at room temperature produced significantly and substantially different mitochondrial morphology than did fixation at 37 degrees Celsius. We tested this effect after observing markedly different mitochondrial morphologies between samples fixed with cold formaldehyde vs warm formaldehyde during preliminary

PAGE 66

56 methods development T his study did not attempt to determine the mechanism underlying these differences, bu t it is reasonable to propose that s tudies of mitochondrial morphology in fixed cells should use 37 degrees Celsius as a physiologically relevant temperature during fixation. Given that many studies in the literature use cooler or unspecified fixation temperatures [ 56 57 58 59 60 61 ] it is possible that mitocho ndrial morphology endpoints in report s such as these may have been compromised. One advantage of the algorithms developed in this study is their ability to reliably detect differences in mitochondrial morphology that are too subtle to see with the naked e ye, as was the case for both Wistar and GK v SMC s exposed to 25 mM glucose. With this advantage comes the potential for aberrant detection of artificially significant results. We found that a total of 3 slide s per treatment group or time point per experimental repetition with 3 experimental repetitions and 3 images per slide for a total of n=27 images and 9 slide s per treatment group was adequate to achieve statistical significance in these cases. This lar ge s ample size was necessary in part because of the great heterogeneity of mitochondrial network morphologies observed in a single slide and in part due to the subtlety of some of the effects recorded. It would not be feasible to use sample sizes this large if measurements had been performed by hand Using our algorithms, little additional time is dedicated to analysis with increasing sample size In the event of a subtle effect which is detected by the

PAGE 67

57 software but not by the naked eye, we highly recommend co mparing the results from individual experimental repetitions as in Figure 6A. For purposes of this study, if an endpoint difference was not visible to the naked eye, it was included only if successive experimental repetitions gave qualitatively similar res ults. In the introduction of this paper we proposed that software measurements of mitochondrial morphology should not require a specific microscope, algorithm expertise should not be required to operate the software, endpoints should be consistent between experimental repetitions, no specific portion of the mitochondrial network should be excluded from analysis or incorrectly analyzed due to algorithm performance, and the endpoints reported to the user should hold intuitive and tangible meanings Our exper iments demonstrate algorithm independence from a specific microscope by showing that measurements of the same samples on different microscopes yielded similar results. Because the software is fully automated, no computational expertise is required to opera te it. Endpoints were found to be quantitatively similar between successive experimental repetitions, and the method of binarization developed in this study proved superior to simple thresholding in preventing bias in measurements of specific portions of t he mitochondrial network. Finally, the primary endpoints reported to the user are average length and width of a single mitochondrial body, which can easily be replicated with by hand measurements.

PAGE 68

58 6.2 Likely Barriers to I mplementation Although the data inc luded in this manuscript indicates the efficacy and practicability of our software, widespread adoption of this technology will require a good deal of continued development. For the same reasons as much of the mitochondrial morphology literature uses subje ctive assessments of mitochondrial morphology, it is relatively unlikely that the laboratories for whom this software is intended will be proficient in Matlab. Future work on this software wil l therefore include translation from Matlab (which allowed for r apid iteration and testing during development) to either a stand alone executable or an ImageJ plugin (better suited for use by the general public). This translation will serve to further ensure that no computational image processing or coding expertise is required for use of the software. Another likely barrier to implementation stems directly from the primary advantage of the software: with absolute, quantitative measures it may prove to be more difficult to reproduce experiments t han with subjective asse ssment. This increase in difficulty of replication, although it results from a higher standard of replication, may make this software less attractive to potential users. As an example of this effect, consider that the baseline HUVEC mitochondria included i n the experiments of Chapter 5 are significantly (p<0.05 using unpaired T test) longer than the baseline HUVEC mitochondria in the experiments of Chapter 4 despite a nominally identical sample preparation.

PAGE 69

59 Judging from inspection of the raw images, it appe ars likely that this difference stemmed from different technician preferences in which cells to image; the HUVECs imaged in Chapter 4 tended to be large, isolated cells whereas those in Chapter 5 tended to be smaller cells within a cell cluster That our software was able to detect a significant difference between these two image sets while returning consistent results between experimental replicates within each image set indicates that even very subtle procedural differences might result in one laboratory obtaining different results from another. Although this fact may make the software less attractive to its end users, it is indicative of a more rigorous and objective analysis, and thus no effort will be made to remove this barrier to implementation. Fina lly, application of this software to samples prepared with different fluorophores or different statistical designs (i.e. different number of experimental replicates and /or sample size s ) may prove to be challenging. Although the software was able to achieve high quality binarization of an uncontrolled image set (see Chapter 5 ), it remains unclear whether consistent detection of very subtle effects such as those reported in section 4.1 will be possible with alternative study designs. In addition, the bead cal ibration method used to account for system error may prove technically challenging for less experience d laboratories. These and many other minor details of software

PAGE 70

60 technology, bu t they are unlikely to outweigh the benefits of objective and quantitative endpoints. The investigators will gladly provide this software to other research groups upon request and assist in adoption and implementation of this technology. 6.3 Potential Future A pplications To date, this software has only been rigorously tested on microscope images of mitochondrial networks in fixed cells acquired using a confocal microscope in. In principle, the same technique would be expected to work regardless of slice width (although the quantities measured would vary), thus enabling use with widefield microscopes. Ongoing experiments will further define the relationship between slice width and software functionality. Similarly, use with live cells and/or videomicroscopy is a logical futu re step. Preliminary experiments (data not shown) indicate that live vs fixed cells do not influence the ability of the software to measure mitochondria, and future experiments will be required to determine if this method is suitable for videomicroscopy. A s a final note, although this method was validated with and designed for mitochondrial images, it is equally applicable in concept to microscope images of the endoplasmic reticulum (ER) and golgi apparatus, which both form reticular networks that are quali tatively similar to the mitochondrial network In addition to the specific experimental questions outlined above, the largest

PAGE 71

61 portion of future work on this project will involve assisting outside laboratories with adoption of this technology. It is this la st task that will take precedence over any of the other proposed future directions discussed here.

PAGE 72

62 LITERATURE CITED 1. Kaftan, E. J., Xu, T., Abercrombie, R. F., & Hille, B. (2000). Mitochondria shape hormonally induced cytoplasmic calcium oscillations and modulate exocytosis Journal of Biological Chemistry, 275(33), 25465 25470. 2. Marhl, M., Schuster, S., & Brumen, M. (1998). Mitochondria as an important factor in the maintenance of constant amplitudes of cytosolic calcium oscillations Bio physical chemistry, 71(2), 125 132. 3. Smaili, S. S., Hsu, Y. T., Carvalho, A. C. P., Rosenstock, T. R., Sharpe, J. C., & Youle, R. J. (2003). Mitochondria, calcium and pro apoptotic proteins as mediators in cell death signaling Brazilian Journal of medical and biological research, 36(2), 183 190. 4. Berridge, M. J. (2008). Smooth muscle cell calcium activation mechanisms The Journal of physiology, 586(21), 5047 5061. 5. Ebashi, S., & Endo, M. (1968). Calcium and muscle contraction. Progress in biophysics and mole cular biology, 18, 123 183. 6. Launay, P., Cheng, H., Srivatsan, S., Penner, R., Fleig, A., & Kinet, J. P. (2004). TRPM4 regulates calcium oscillations after T cell activation Science, 306(5700), 1374 1377.

PAGE 73

63 7. Rizzuto, R., De Stefani, D., Raffaell o, A., & Mammucari, C. (2012). Mitochondria as sensors and regulators of calcium signalling Nature Reviews Molecular Cell Biology, 13(9), 566 578. 8. Nishikawa, T., Edelstein, D., Du, X. L., Yamagishi, S. I., Matsumura, T., Kaneda, Y., Yorek, M. & Brownlee, M. (2000). Normalizing mitochondrial superoxide production blocks three pathways of hyperglycaemic damage Nature, 404(6779), 787 790. 9. Sena, L. A., Li, S., Jairaman, A., Prakriya, M., Ezponda, T., Hildeman, D. A., Wang, C., & Chandel, N. S. (2013). Mitoch ondria are required for antigen specific T cell activation through reactive oxygen species signaling Immunity, 38(2), 225 236. 10. Brand, M. D., Affourtit, C., Esteves, T. C., Green, K., Lambert, A. J., Miwa, S., Pakay, J., & Parker, N. (2004). Mitochondrial superoxide: production, biological effects, and activation of uncoupling proteins Free Radical Biology and Medicine, 37(6), 755 767. 11. Pletjushkina, O. Y., Lyamzaev, K. G., Popova, E. N., Nepryakhina, O. K., Ivanova, O. Y., Domnina, L. V., Chernyak, B. V., & Skulachev, V. P. (2006). Effect of oxidative stress on dynamics of mitochondrial reticulum Biochimica et Biophysica Acta (BBA) Bioenergetics, 1757(5), 518 524. 12. Murata, T., Goshima, F., Daikoku, T., Inagaki Ohara, K., Takakuwa, H., Kato, K., & Nishiyama, Y. (2000). Mitochondrial distribution and function

PAGE 74

64 in herpes simplex virus infected cells Journal of General Virology, 81(2), 401 406. 13. de Boer, V. C., & Houten, S. M. (2014). A Mitochondrial Expatriate: Nuclear Pyruvate Dehydrogenase Cell, 158(1), 9 10. 14. Chen, H., Chomyn, A., & Chan, D. C. (2005). Disruption of fusion results in mitochondrial heterogeneity and dysfunction Journal of Biological Chemistry, 280(28), 26185 26192. 15. Youle, R. J., & Narendra, D. P. (2011). Mechanisms of mitophagy Nature review s Molecular cell biology, 12(1), 9 14. 16. Yu, T., Robotham, J. L., & Yoon, Y. (2006). Increased production of reactive oxygen species in hyperglycemic conditions requires dynamic change of mitochondrial morphology Proceedings of the National Academy of Sciences of the United States of America, 103(8), 2653 2658. 17. Quintana, A., & Hoth, M. (2012). Mitochondrial dynamics and their impact on T cell function Cell calcium, 52(1), 57 63. 18. Liesa, M., & Shirihai, O. S. (2013). Mitochondrial dynamics in the regulation of nutrient utilization and energy expenditure. Cell metabolism 17(4), 491 506.

PAGE 75

65 19. Karbowski, M., & Youle, R. J. (2003). Dynamics of mitochondrial morphology in healthy cells and during apoptosis Cell Death & Differentiation, 10(8), 870 880. 20. Lie sa, M., Palacn, M., & Zorzano, A. (2009). Mitochondrial dynamics in mammalian health and disease Physiological reviews, 89(3), 799 845. 21. Alano, C. C., Beutner, G., Dirksen, R. T., Gross, R. A., & Sheu, S. S. (2002). Mitochondrial permeability transition a nd calcium dynamics in striatal neurons upon intense NMDA receptor activation Journal of neurochemistry, 80(3), 531 538. 22. Picard, M., Shirihai, O. S., Gentil, B. J., & Burelle, Y. (2013). Mitochondrial morphology transitions and functions: implications for retrograde signaling? American Journal of Physiology Regulatory, Integrative and Comparative Physiology, 304(6), R393 R406. 23. Yoon, Y., Galloway, C. A., Jhun, B. S., & Yu, T. (2011). Mitochondrial dynamics in diabetes Antioxidants & redox signaling, 14(3), 439 457. 24. Dietrich, M. O., Liu, Z. W., & Horvath, T. L. (2013). Mitochondrial dynamics controlled by mitofusins regulate Agrp neuronal activity and diet induced obesity Cell, 155(1), 188 199. 25. Seo, A. Y., Joseph, A. M., Dutta, D., Hwang, J. C., Aris, J. P., & Leeuwenburgh, C. (2010). New insights into the role of mitochondria in

PAGE 76

66 aging: mitochondrial dynamics and more Journal of cell science, 123(15), 2533 2542. 26. Ong, S. B., & Hausenloy, D. J. (2010). Mitochondrial morphology and cardiovascular disease Cardiovascular research, 88(1), 16 29. 27. Aboud, A. A., Tidball, A. M., Kumar, K. K., Neely, M. D., Han, B., Ess, K. C., Hong, C. C., Erikson, K. M., Hedera, P., & Bowman, A. B. (2015). PARK2 patient neuroprogenitors show increased mitochondrial sensitivity to copper Neurobiology of disease, 73, 204 212. 28. Li, L., Nadanaciva, S., Berger, Z., Shen, W., Paumier, K., Schwartz, J., Mou, K., & Hirst, W. D. (2013). synuclein causes reversible deficits in mitochondrial function and dynamics in primary mouse cortical neurons PloS one, 8(12), e85815. 29. Xie, W., & Chung, K. K. (2012). Alpha synuclein impairs normal disease Journal of neurochemistry, 122(2), 404 414. 30. Koopman, W. J., Visch, H. J., Smeitink, J. A., & Willems, P. H. (2006). Simultaneous quantitative measurement and automated analysis of mitochondrial morphology, mass, potential, and motility in living human skin fibroblasts Cytometry Part A, 69(1), 1 12. 31. Park, M. K., Ashby, M. C ., Erdemli, G., Petersen, O. H., & Tepikin, A. V. (2001). Perinuclear, perigranular and sub plasmalemmal mitochondria

PAGE 77

67 have distinct functions in the regulation of cellular calcium transport The EMBO Journal, 20(8), 1863 1874. 32. Bahamonde, M. I., & Valverde, M. A. (2003). Voltage dependent anion channel localises to the plasma membrane and peripheral but not perinuclear mitochondria European Journal of Physiology 446(3), 309 313. 33. Collins, T. J., Berridge, M. J., Lipp, P., & Bootman, M. D. (2002). Mitochondria are morphologically and functionally heterogeneous within cells The EMBO Journal, 21(7), 1616 1627. 34. Trudeau, K., Molina, A. J., Guo, W., & Roy, S. (2010). High glucose disrupts mitochondrial morphology in retinal endothelial cells: implicatio ns for diabetic retinopathy The American journal of pathology, 177(1), 447 455. 35. Yu, T., Jhun, B. S., & Yoon, Y. (2011). High glucose stimulation increases reactive oxygen species production through the calcium and mitogen activated protein kinase mediated activation of mitochondrial fission Antioxidants & redox signaling, 14(3), 425 437. 36. Jendrach, M., Mai, S., Pohl, S., Vth, M., & Bereiter Hahn, J. (2008). Short and long term alterations of mitochondrial morphology, dynamics and mtDNA after transient oxidative stress Mitochondrion, 8(4), 293 304.

PAGE 78

68 37. Jahani Asl, A., Cheung, E. C., Neuspiel, M., MacLaurin, J. G., Fortin, A., Park, D. S., McBride, H. M., & Slack, R. S. (2007). Mitofusin 2 protects cerebellar granule neurons against injury induced cell death Journal of Biological Chemistry, 282(33), 23788 23798. 38. Fan, X., Hussien, R., & Brooks, G. A. (2010). H 2 O 2 induced mitochondrial fragmentation in C 2 C 12 myocytes Free Radical Biology and Medicine, 49(11), 1646 1654. 39. Rintoul, G. L., Filiano, A. J., B rocard, J. B., Kress, G. J., & Reynolds, I. J. (2003). Glutamate decreases mitochondrial size and movement in primary forebrain neurons The Journal of neuroscience, 23(21), 7881 7888. 40. Nozaki, T., Kagaya, Y., Ishide, N., Kitada S., Miura, M., Nawata, J., Ohno, I., Watanabe, J., & Shirato, K. (2001). Interaction between sarcomere and mitochondrial length in normoxic and hypoxic rat ventricular papillary muscles Cardiovascular Pathology, 10(3), 125 132. 41. Wang, D. B., Garden, G. A., Kinoshita, C., Wyles, C., Babazadeh, N., Sopher, B., Kinoshita, Y., & Morrison, R. S. (2013). Declines in Drp1 and parkin expression underlie DNA damage induced changes in mitochondrial length and neuronal death The Journal of Neuroscience, 33(4), 1357 1365.

PAGE 79

69 42. Korobova, F., Ramabhadran, V., & Higgs, H. N. (2013). An actin dependent step in mitochondrial fission mediated by the ER associated formin INF2 Science, 339(6118), 464 467. 43. Choi, S., Yu, E., Kim, D. S., Sugimori, M., & Llins, R. R. (2015). RNS60, a char ge stabilized nanostructure saline alters Xenopus Laevis oocyte biophysical membrane properties by enhancing mitochondrial ATP production Physiological reports, 3(3), e12261. 44. Cagalinec, M., Safiulina, D., Liiv, M., Liiv, J., Choubey, V., Wareski, P., Veks ler, V., & Kaasik, A. (2013). Principles of the mitochondrial fusion and fission cycle in neurons Journal of cell science, 126(10), 2187 2197. 45. Korobova, F., Gauvin, T. J., & Higgs, H. N. (2014). A role for myosin II in mammalian mitochondrial fission Cur rent Biology, 24(4), 409 414. 46. Gan, X., Huang, S., Wu, L., Wang, Y., Hu, G., Li, G., Zhang, H., & Yan, S. S. (2014). Inhibition of ERK DLP1 signaling and mitochondrial division alleviates mitochondrial dysfunction in Alzheimer's disease cybrid cell Biochimica et Biophysica Acta (BBA) Molecular Basis of Disease, 1842(2), 220 231. 47. Moran, M., Rivera, H., Snchez Arag, M., Blazquez, A., Merinero, B., Ugalde, C., Arenas, J., Cuezva, J. M., & Martin, M. A. (2010). Mitochondrial bioenergetics and dynamics interplay in complex I

PAGE 80

70 deficient fibroblasts Biochimica et Biophysica Acta (BBA) Molecular Basis of Disease, 1802(5), 443 453. 48. Finley, L. W., Lee, J., Souza, A., Desquiret Dumas, V., Bullock, K., Rowe, G. C., Procaccio, V., Clish, C. B., Arany, Z., & Haig is, M. C. (2012). Skeletal muscle transcriptional coactivator PGC mitochondrial, but not metabolic, changes during calorie restriction Proceedings of the National Academy of Sciences, 109(8), 2931 2936. 49. Karam, S. M., Straiton, T., Hassan, W. M ., & Leblond, C. P. (2003). Defining epithelial cell progenitors in the human oxyntic mucosa Stem Cells, 21(3), 322 336. 50. Santamara, G., Martnez Diez, M., Fabregat, I., & Cuezva, J. M. (2006). Efficient execution of cell death in non glycolytic cells req uires the generation of ROS controlled by the activity of mitochondrial H+ ATP synthase Carcinogenesis, 27(5), 925 935. 51. Holloway, G. P., Gurd, B. J., Snook, L. A., Lally, J., & Bonen, A. (2010). y associated with subsarcolemmal mitochondrial adaptations in ZDF rats Diabetes, 59(4), 819 828. 52. Chattipakorn, S. C., Thummasorn, S., Sanit, J., & Chattipakorn, N. (2014). Phosphodiesterase 3 inhibitor (cilostazol) attenuates oxidative

PAGE 81

71 stress induced mitochondrial dysfunction in the heart Journal of geriatric cardiology: JGC, 11(2), 151. 53. Lakroun, Z., Kebieche, M., Lahouel, A., Zama, D., Desor, F., & Soulimani, R. (2015). Oxidative stress and brain mitochondria swelling induced by endosulfan and protec tive role of quercetin in rat Environmental Science and Pollution Research, 22(10), 7776 7781. 54. Rodrguez Martnez, E., Martnez, F., Espinosa Garca, M. T., Maldonado, P., & Rivas Arancibia, S. (2013). Mitochondrial dysfunction in the hippocampus of rats caused by chronic oxidative stress Neuroscience, 252, 384 395. 55. Paneni, F., Capretti, G., Costantino, S., Chiandotto, S., Akhmedov, A., Di Stasio, E., Rocca, B., Luscher, T. F., Volpe. M., & Cosentino, F. (2013). The lifespan determinant p66shc drives obes ity induced oxidative stress, mitochondrial dysfunction and vascular inflammation European Heart Journal, 34(suppl 1), 781. 56. Pyakurel, A., Savoia, C., Hess, D., & Scorrano, L. (2015). Extracellular Regulated Kinase Phosphorylates Mitofusin 1 to Control Mit ochondrial Morphology and Apoptosis Molecular cell, 58(2), 244 254. 57. Onoue, K., Jofuku, A., Ban Ishihara, R., Ishihara, T., Maeda, M., Koshiba, T., Itoh, T., & Ishihara, N. (2013). Fis1 acts as a mitochondrial

PAGE 82

72 recruitment factor for TBC1D15 that is involved in regulation of mitochondrial morphology Journal of cell science, 126(1), 176 185. 58. Zhu, M., Li, W., & Lu, C. (2012). Role of alpha synuclein protein levels in mitochondrial morphology and cell survival in cell lines PloS one, 7(4), e36377. 59. Newm an, L. E., Zhou, C. J., Mudigonda, S., Mattheyses, A. L., Paradies, E., Marobbio, C. M. T., & Kahn, R. A. (2014). The ARL2 GTPase Is Required for Mitochondrial Morphology, Motility, and Maintenance of ATP Levels PloS one, 9(6): e99270 60. Picard, M., White, K ., & Turnbull, D. M. (2013). Mitochondrial morphology, topology, and membrane interactions in skeletal muscle: a quantitative three dimensional electron microscopy study Journal of Applied Physiology, 114(2), 161 171. 61. Balan, I. S., Saladino, A. J., Aarabi B., Castellani, R. J., Wade, C., Stein, D. M., Eisenberg, H. M., Chen, H. H., & Fiskum, G. (2013). Cellular alterations in human traumatic brain injury: changes in mitochondrial morphology reflect regional levels of injury severity Journal of neurotraum a, 30(5), 367 381.