Post on 02-Dec-2020
Catabolism of Extracellular Protein
by Pancreatic Cancer Cells
Michel Ibrahim Nofal
A Dissertation
Presented to the Faculty
of Princeton University
in Candidacy for the Degree
of Doctor of Philosophy
Recommended for Acceptance
by the Department of
Quantitative and Computational Biology
Adviser: Professor Joshua D. Rabinowitz
November 2018
c© Copyright by Michel Ibrahim Nofal, 2018.
All rights reserved.
Abstract
All cells require amino acids to support protein synthesis and cell growth. Until
recently, mammalian cells were thought to depend on monomeric amino acids in the
environment. I showed that pancreatic tumor cells can use extracellular protein as
a source of amino acids. These cells take up intact protein via macropinocytosis
and catabolize it in lysosomes. This process – “protein eating” – enables cultured
pancreatic cancer cells to grow in amino acid-deficient environments. In this thesis, I
present my work on protein eating.
To show that protein eating is capable of fueling growth, I cultured murine pan-
creatic cancer cells in medium lacking leucine (an essential amino acid) and supple-
mented with a physiological concentration of serum albumin. Many cells cells died in
this medium, but some survived and grew to confluence. I passaged these survivors
for months, and they gradually adapted to growth fueled by protein eating. This
proved that protein eating is a viable form of amino acid uptake.
I developed isotope tracer-based methods to quantitatively measure protein eating.
Cells are grown in medium with stable isotope-labeled amino acids and unlabeled
serum protein. Mass spectrometry enables distinction of amino acids taken up as
monomers (labeled) from amino acids taken up as intact protein and catabolized
(unlabeled).
I conducted genome-wide screens to systematically identify genes essential for
growth fueled by protein eating. The most essential gene was GCN2, which suppresses
translation initiation in cells starved for amino acids. I discovered that loss of GCN2
impairs catabolism in amino acid-deprived cells. I propose that GCN2 supports
catabolism by directing amino acids emerging from lysosomes into newly synthesized
proteins that increase the catabolic capacity of the cell – for example, the lysosomal
hydrolase cathepsin L.
iii
Advances in our understanding of protein eating may lead to the development of
better therapies for pancreatic cancer patients. The importance of protein eating as
an amino acid supply route for cells in healthy tissues remains unexplored.
iv
Acknowledgements
I am lucky to be where I am today. I have a lot of people to thank. Before I was
born, my parents immigrated to the Silicon Valley to work as hardware engineers.
They worked hard throughout their lives to make things easy for me. They taught
me to be curious and ambitious. I thank them for their continuous love and support.
Once I have a PhD, I will call more often. I also thank my sister, who had to put up
with me growing up – I was probably more competitive than I even realize now.
My parents spent a lot of money sending me to a private school, Pinewood School,
from kindergarten through twelfth grade. (Pinewood had three campuses: K-2, 3-6,
and 7-12.) Many of my classmates were the children of similarly enterprising parents
who had moved to the area from overseas to work in high tech. Our parents all pushed
us, and we pushed each other. It was easy to succeed in that environment. (Of the
57 students in my graduating class, two went on to Stanford, one to Harvard, one to
Brown, one to Duke, one to Bowdoin, one to Amherst, six to UC Berkeley, two to
UCLA, and so on.) I thank my friends and teachers at Pinewood for preparing me
so well, especially my calculus teacher, Mr. Green. I also thank my basketball coach,
Coach Slayton, and my piano teacher, Mrs. Wang, for toughening me up, preparing
me for grad school.
After Pinewood, I became an undergraduate at Berkeley. For the first two years,
I lived with a high school friend, Tim Wang, who was a year ahead of me in school.
We both majored in bioengineering and joined labs. When the time came for me to
decide what to do after college, I saw Tim applying to PhD programs, and it seemed
like something I would like, so I did it too. I thank Tim for his friendship and his
mentorship over the years.
With my undergraduate resume, I was fortunate to be admitted to exactly one
PhD program: Princeton Quantitative and Computational Biology. Soon after arriv-
ing in Princeton, I joined the Rabinowitz lab and started working on protein eating.
v
(The first experiment I ever did was a growth experiment in leucine-free medium to
see if cells could use extracellular protein to support growth.) Since then, I have
consumed more than my fair share of lab resources while constantly annoying others
peacefully going about their work. I thank everyone that I ever overlapped with in
the Rabinowitz lab for contributing to a great intellectual environment.
In particular, I thank Jurre Kamphorst and Jing Fan for getting me started;
Wenyun Lu and Lin Wang for spending their valuable time helping me maintain
a more-or-less personal mass spectrometer; Sean Hackett and David Robinson for
teaching me R; Tomer Shlomi and Vito Zanotelli for helping me learn metabolic
flux analysis; Ian Lewis, Greg Ducker, Jon Ghergurovich, and Juan Carlos Garcia
Canaveras for putting up with my incessant questions during working hours; Raphael
Morscher and Matt Sonnett for putting up with my incessant questions extremely
late at night; Jun Park for asking me as many questions as I ask him; Lukas Tanner
for a strong friendship in a lonely time; Xiaoyang Su and Rob Marmion for invaluable
cloning help; Sophia Li and Mark Esposito for experiencing graduate school with me
from start to finish; Cholsoon Jang and Gina Lee for sharing valuable papers and
reagents with me; and Lifeng Yang, who is one of my best friends and not a nao-
can. I thank my thesis committee members Yibin Kang, Eileen White, and Martin
Wuhr for providing me with thoughtful advice, as well as Fred Hughson and Alexei
Korennykh for rare high-level discussion of cell biology. I also thank Martin Wuhr
for his generosity with respect to our ongoing collaboration, which I hope does not
end soon. I thank Gary Laevsky and Tina DeCoste for imaging and flow cytometry
support, and the Devenport and Toettcher labs for imaging help. I thank the hard-
working undergraduate researchers I mentored, Kevin Zhang, Aiden Han, Sriram
Cyr, and Agustin Zavala. Most importantly, I thank Josh, who saw potential in me
when few others did. Josh trusted me enough to let me and my research program
drift further and further away from the main interests of his lab over the course of
vi
seven years. Josh taught me biochemistry, how to teach biochemistry, how to think
scientifically, how to write scientifically, and how to see the story through the data.
Also, I recognize that my research program was not cheap.
A short aside: Once we knew protein eating could fuel growth, the two obvious
questions were (i) can we measure protein eating quantitatively, and (ii) can we
identify the genes essential for protein eating? Luckily, I was in exactly the right lab
to answer the first question. The second question, however, called for a genetic screen,
which I could not do in Princeton. At the time, two groups developed the ability to
conduct genome-wide screens in mammalian cells using CRISPR-Cas9 technology;
they published back-to-back papers in Science. One group was led by Feng Zhang,
who may win the Nobel Prize for the development of CRISPR-Cas9 technology. The
other group was led by Tim Wang, whose screening technology was better. The fact
that Tim and his advisors agreed to host me to do CRISPR screens was a stroke of
luck that lifted my career. I thank Tim and his advisors, David Sabatini and Eric
Lander, for their generosity in hosting me.
Many have supported me outside the lab during my tenure in graduate school. I
thank all my friends in Princeton, especially Mark Esposito, Rezma Shrestha, Sara
Forster, Matt Streeter, Jordan Maseng, Sophia Li, Fred Shipley, Benno Mirabelli, and
Jordan Ash. I also thank Max Wilson, who I lived with for two years and grew up
with scientifically. He managed to become a professor at UC Santa Barbara before I
finished graduate school, which qualified him to be my second thesis reader. Finally,
I thank my girlfriend, Gabriela Castro, for her love and patience, and her family, who
regularly welcomes me into their home as if I were one of their own.
vii
To my parents.
viii
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
1 Introduction 1
1.1 Foreward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Scientific Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Human Pancreatic Cancer Tumors Are Nutrient Poor and Tumor
Cells Actively Scavenge Extracellular Protein 13
2.1 Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Results - Metabolomic analysis of human PDAC tumors . . . . . . . 23
2.6 Results - Macropinocytosis in PDAC tumors . . . . . . . . . . . . . . 27
2.7 Results - Support of cultured tumor cell growth by albumin in the
absence of free amino acids . . . . . . . . . . . . . . . . . . . . . . . . 28
2.8 Results - Isotope tracing of serum protein catabolism . . . . . . . . . 30
2.9 Results - Amino acid patterns in cells fed by serum protein
macropinocytosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
ix
2.10 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3 mTOR Inhibition Restores Amino Acid Balance In Cells Dependent
on Catabolism of Extracellular Protein 38
3.1 Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 Results - Isotope-Tracer Method Measures Amino Acid Release Due
to Extracellular Protein Catabolism . . . . . . . . . . . . . . . . . . . 43
3.5 Results - Impact of Intracellular Protein Catabolism on Scavenging
Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.6 Results - Excessive mTOR Inhibition Slows Growth on Extracellular
Protein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7 Results - Amino Acid-Deficiency Induces Protein Scavenging Flux In-
dependently of mTOR . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.8 Results - mTOR Inhibition Induces Punctate DQ-BSA Fluorescence . 52
3.9 Results - mTOR Inhibition Restores Amino Acid Balance and Prevents
Cell Death in Amino Acid-Deprived Cells . . . . . . . . . . . . . . . . 56
3.10 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.11 Materials and Methods - Cell lines . . . . . . . . . . . . . . . . . . . 62
3.12 Materials and Methods - Measuring catabolism of extracellular protein 63
3.13 Materials and Methods - Other experimental methods . . . . . . . . . 69
4 A Genome-Wide Screen Identifies The Proteins Behind Protein Eat-
ing: GCN2 and cathepsin L 72
4.1 Proposed Manuscript Title . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
x
4.4 Results - Genome-wide screen systematically identifies genes required
for growth fueled by catabolized extracellular protein . . . . . . . . . 77
4.5 Results - The Three Major Categories of Selectively Essential Genes:
Uptake, Degradation, and Regulation of Translation . . . . . . . . . . 86
4.6 Results - Screen Validation and Proteomics . . . . . . . . . . . . . . . 96
5 Ribosomes on the Night Shift: The universal protein-making ma-
chine becomes a nutrient source between meals 107
6 Conclusion and Future Directions 113
Bibliography 118
xi
List of Figures
2.1 Cancer Research paper - Figure 1 . . . . . . . . . . . . . . . . . . . . 26
2.2 Cancer Research paper - Figure 2 . . . . . . . . . . . . . . . . . . . . 28
2.3 Cancer Research paper - Figure 3 . . . . . . . . . . . . . . . . . . . . 30
2.4 Cancer Research paper - Figure 4 . . . . . . . . . . . . . . . . . . . . 32
2.5 Cancer Research paper - Figure 5 . . . . . . . . . . . . . . . . . . . . 34
3.1 Molecular Cell paper - Graphical Abstract . . . . . . . . . . . . . . . 40
3.2 Molecular Cell paper - Figure 1 . . . . . . . . . . . . . . . . . . . . . 45
3.3 Molecular Cell paper - Figure 2 . . . . . . . . . . . . . . . . . . . . . 46
3.4 Molecular Cell paper - Figure 3 . . . . . . . . . . . . . . . . . . . . . 49
3.5 Molecular Cell paper - Figure 4 . . . . . . . . . . . . . . . . . . . . . 51
3.6 Molecular Cell paper - Figure 5 . . . . . . . . . . . . . . . . . . . . . 55
3.7 Molecular Cell paper - Figure 6 . . . . . . . . . . . . . . . . . . . . . 58
3.8 Molecular Cell paper - Figure 7 . . . . . . . . . . . . . . . . . . . . . 60
4.1 Genome-wide screen design and summary . . . . . . . . . . . . . . . . 79
4.2 Screen results for Gcn2 and Gcn1 . . . . . . . . . . . . . . . . . . . . 80
4.3 Selectively essential genes were highly expressed . . . . . . . . . . . . 81
4.4 K-Ras and the V-ATPase are essential in all growth conditions . . . . 82
4.5 Selective essentiality of actin capping protein isoforms . . . . . . . . . 83
4.6 Screen results for various protein complexes . . . . . . . . . . . . . . 85
xii
4.7 Selective essentiality of actin-related proteins . . . . . . . . . . . . . . 87
4.8 Selective essentiality of Rab proteins and Rabankyrin-5 . . . . . . . . 90
4.9 Comparison of the selective essentialities of translation regulators . . 95
4.10 Basic validation of selectively essential genes . . . . . . . . . . . . . . 97
4.11 GCN2 supports protein catabolism in amino acid-deprived cells . . . 99
4.12 The effect of GCN2 and amino acid depletion on protein levels . . . . 101
4.13 GCN2 is required to maintain cathepsin L levels in amino acid-deficient
conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.1 NUFIP1 Figure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
xiii
Chapter 1
Introduction
This introduction has two parts: an informal foreward and a formal scientific intro-
duction.
1.1 Foreward
In social settings when I am asked what I research, I find myself reluctant to answer.
I appreciate the curiosity, but I worry that I cannot possibly communicate what it is
that I work on in less than five or ten minutes, which is much longer than the person,
who probably just asked to be polite, is willing to listen. So I give short answers.
Just buzz words, no attempt to communicate real ideas.
“What do you study?”
“Pancreatic cancer.”
“What about pancreatic cancer?”
“I research the response of cultured pancreatic cancer cells to amino acid starva-
tion.”
“Oh, cool!”
Or sometimes: “Oh, sounds complicated...” Either way, I end up deeply unsatis-
fied, because I feel there has been a misunderstanding. To me, it seems the questions
1
I have spent years trying to answer are basic, fundamental, easy for anyone literate in
cell biology to understand. Indeed, the answers to these questions are complicated,
but the questions themselves have always been simple. Nevertheless, they remain in-
accessible to the average reasonable human being, who begins feeling uncomfortable
thirty seconds into any conversations involving words like “pancreatic” and “amino.”
The trouble I have in beginning to describe what I work on is that I dont work
on just one biological thing – a gene, a metabolic pathway, an organelle. I work
on relationships between apparently unrelated things – catabolism of extracellular
protein and regulation of translation, lysosomes and GCN2. Where to start?
Pancreatic tumors are bad news – if you find out you have one, you probably wont
live another year. They emerge from cells of the exocrine pancreas (as opposed to the
other part of the pancreas, the endocrine pancreas). The exocrine pancreas produces
digestive enzymes that are secreted in the form of “pancreatic juice.” These enzymes
travel through the bile duct and into the small intestine, where they digest food. The
endocrine pancreas produces insulin and glucagon, which control blood sugar levels.
Cells in both the exocrine pancreas and the endocrine pancreas have something in
common: their job is to synthesize and secrete proteins. I think it is fair to say that
protein synthesis and protein trafficking are the most important biological processes
in these cells.
Two things seem to be true of cells of the exocrine pancreas that turn into ma-
lignant tumors. First, these cells harbor a genetic mutation that activates the Ras
signaling pathway. Usually, the mutation results in a single amino acid substitution
in the K-Ras protein; commonly, the twelfth amino acid in K-Ras, originally glycine,
becomes aspartate, or valine, or cysteine. Second, the environment in which pancre-
atic tumor-initiating cells reside is inflamed. Inflammation activates other proteins,
including the transcription factor cellular Myc (c-Myc). A viral homolog of c-Myc,
v-Myc, is sufficient to induce tumorigenesis in cells infected by the Avian virus har-
2
boring this gene. In pancreatic cells, however, neither Ras nor Myc alone causes
malignancies; the two “cooperate” to form tumors.
The inflammatory environment of pancreatic tumors elicits responses from tu-
mor cells and non-tumor cells alike. Pancreatic stellate cells, which remain quiescent
(largely inactive) in healthy pancreatic tissue, are activated by inflammation. Acti-
vated stellate cells are induced to secrete extracellular matrix proteins – this is called
“fibrosis.” Fibrosis is noticeable macroscopically: pancreatic tissue is soft, but tu-
mors are hard, like scar tissue. Fibrosis limits the diffusion of circulating nutrients
into the tumor, so cells in the interior of the tumor are deprived of oxygen, glucose,
and other important metabolites. The metabolite most depleted in human pancreatic
tumors, as we report in the first paper included in this thesis, is glutamine. Thus,
pancreatic tumor cells have a problem: they must survive and even grow in an amino
acid-deficient environment. How do they solve this problem?
While the levels of some amino acid monomers are low in pancreatic tumors, the
levels of serum proteins are high. In fact, due to leaky vasculature and deficient
lymphatic drainage, serum protein accumulates in pancreatic tumors. This serum
protein, as well as the extracellular matrix protein secreted by activated pancreatic
stellate cells, is a potential source of amino acids. Tumor cells can take up protein
from the extracellular space and degrade it in lysosomes to produce amino acids to
fuel growth – I refer to the totality of this process as “protein eating.”
The idea that tumor cells might use extracellular protein as a fuel source is rel-
atively new – there were no published papers about it when I started my graduate
career. The biochemical processes underlying protein eating remain unclear. Up-
take is achieved through a process called “macropinocytosis,” in which extracellular
material is taken up non-specifically into big vesicles called “macropinosomes.” The
protein internalized in macropinosomes somehow is delivered to “lysosomes,” where it
is degraded. Degradation yields amino acids, which can be used to synthesize protein
3
required for survival or growth. How exactly are uptake, trafficking, and degradation
achieved? Are any steps in the complicated process of eating extracellular protein
particularly important? Perhaps a pharmacological inhibitor of a protein critical to
this process would thwart pancreatic cancer growth.
These were the questions I started with, but in trying to answer them, another
question emerged: What do amino acid-deprived cells do with the limited amino
acids that they generate through protein eating? These are important decisions for
pancreatic tumor cells residing in the poorly perfused interior of a pancreatic tumor.
Cells that make the wrong decisions die; those that make the right decisions survive
and grow. If enough cells consistently make the right decision, they kill their host, the
patient. Unfortunately, some drugs (mTOR, PI3K inhibitors) that have been tested
in pancreatic cancer clinical trials help cells make the right decision. Big mistakes
have been made because we do not understand this question, and big opportunities
may present themselves once we do.
I have found it useful to approach this question logically. I assume that amino
acid-deprived cells will use the limited amino acids generated by protein eating wisely,
and I try to imagine what a wise cell would do. This is how I generate hypotheses.
A pancreatic tumor cell starved of amino acids would be wise to (i) degrade as much
protein as possible and (ii) use the resulting amino acids to synthesize proteins that
will increase the catabolic capacity of the cell. The cell would then degrade more
proteins, producing more amino acids, enabling further increases in catabolic capacity,
until the cell has enough amino acids. At that point, the amino acids could be used
for growth.
Based on this reasoning, I have had the idea that amino acid-deprived cells upreg-
ulate the synthesis of proteins that help them catabolize more protein. GCN2 is the
protein that directs amino acid-starved cells to use their amino acids wisely. GCN2 is
known to suppress translation initiation when amino acid pools are depleted. My pre-
4
diction is that GCN2 suppresses the translation of most transcripts (GCN2-sensitive
transcripts) to free up amino acids for the synthesis of proteins encoded by other tran-
scripts (GCN2-resistant transcripts). For the past six months, I have been searching
for the key proteins upregulated by GCN2 that induce higher catabolic rates.
Just a few weeks ago, upon analyzing proteomics data from GCN2 wild-type and
GCN2 knockout cells, I developed the following hypothesis. I think that the synthesis
of cathepsin L, a short-lived lysosomal hydrolase that seems to be especially impor-
tant for protein eating, is induced by GCN2. Thus, loss of GCN2 impairs cathepsin
L synthesis in amino acid-deprived cells, and as a result, these cells exhibit lower
catabolic rates. I think that inhibitors of GCN2 and cathepsin L could potentially
block pancreatic tumor growth by impairing the tumor cell response to amino acid
deprivation. I could be wrong.
This thesis contains an introduction, two published papers, a perspective, the
beginnings of a third paper, and a forward-looking conclusion. In the two published
papers, the language and figures are refined, but the findings are crude. (This is a
harsh assessment, but with the benefit of hindsight, I think it is fair.) The opposite is
true for the incomplete third paper and conclusion, which includes a short proposal
to study protein trafficking as a network of fluxes. In other words, the intellectual
contents of the final section have been honed over the course of many years, but
the language and figures are far from honed. I have been busy with experiments,
unwilling to divert time and energy away from the never-ending struggle to get results
that support my ideas. So it goes as a graduate student in biology.
1.2 Scientific Introduction
Pancreatic ductal adenocarcinoma (PDAC) is a devastating disease, with a 5-year
overall survival rate of 8% [103]. Existing therapies for PDAC patients have generally
5
been ineffective for the following reasons. First, these patients often do not learn that
they have pancreatic cancer until the tumor has progressed to an advanced stage,
infiltrating surrounding tissues and sometimes metastasizing to distal organs. Only
9% of newly diagnosed pancreatic cancer is localized [103]. Second, despite sustained
efforts over several decades, clinicians and researchers have been unable to identify
specific vulnerabilities of pancreatic tumor cells that can be targeted pharmacologi-
cally. Instead, patients are treated with various combinations of blunt chemothera-
peutic agents that are toxic to pancreatic tumor cells but also cause serious systemic
problems, like thrombocytopenia, anemia, and peripheral neuropathy [120]. Lastly,
characteristic fibrosis by cancer-associated pancreatic stellate cells limits perfusion.
As a result, delivery of therapeutic agents into tumors is a challenge [74].
Pancreatic tumors arise from cells of the exocrine pancreas [41]. (This is not true
for all pancreatic tumors; rarely, pancreatic tumors arise from pancreatic neuroen-
docrine cells. In this thesis, “pancreatic tumors” refer to ductal adenocarcinomas of
the pancreas, and “pancreatic cancer cells” refer to PDAC tumor cells.) The exocrine
pancreas produces and secretes digestive enzymes that are secreted through the pan-
creatic duct, then the bile duct, into the small intestine, where they break down food.
Thus, the cells of the exocrine pancreas exist to synthesize and transport protein.
(This is also true for cells of the endocrine pancreas, which regulates blood glucose
levels through the actions of alpha cells, which produce and export glucagon, and
beta cells, which produce and export insulin.) As a whole, the pancreas is a protein
synthesis machine [33].
There seem to be two preconditions for pancreatic tumorigenesis. The first is that
pancreatic exocrine cells harbor a mutation that activates the Ras signaling pathway.
The second is inflammation in the pancreas, which activates the transcription factor c-
Myc. Activation of only one of these two proteins is insufficient to drive tumor growth;
both must be activated [54]. As mentioned previously, the resulting tumors are fibrotic
6
and poorly perfused; as such, they have long been assumed to be nutrient-poor. In
the second chapter of this thesis (a paper published in Cancer Research in 2015),
we validate this assumption: we report measurements of 127 soluble metabolites in
human pancreatic tumors and paired adjacent pancreatic tissue. The levels of glucose
and glutamine – the most important carbon and nitrogen sources for cultured cells
– were lower in tumors than in healthy adjacent tissue. Thus, we confirmed that
pancreatic tumor cells are starved for basic nutrients [48].
How do pancreatic tumor cells, with limited amounts of these primary nutrients
at their disposal, support survival and growth? This is the central question of pan-
creatic tumor cell metabolism. The ultimate goal of the cancer metabolism research
community is to identify metabolic activities that tumor cells use to overcome the
nutrient scarcity. If we found a metabolic activity uniquely important to pancreatic
tumor cells, for example, it would be an attractive therapeutic target.
Early in my tenure as a graduate student, directed by my advisor Josh Rabinowitz,
I discovered such an activity. To overcome amino acid scarcity, pancreatic tumor cells
take up extracellular protein from the environment and catabolize it in lysosomes. (In
this thesis, I refer to the totality of this process as “protein eating.”) We were not the
first to discover this phenomenon, nor to publish about it; Dafna Bar Sagi and others
were. Over thirty years ago, Bar Sagi had published a paper showing that Ras activity
induces membrane ruffling [4]. For years, the function of membrane ruffling was
not understood, but at some point, it probably became clear to her that membrane
ruffling enabled the uptake (and subsequent degradation) of extracellular protein
via a process called macropinocytosis. Macropinocytosis internalizes the contents of
the extracellular medium non-specifically into large vesicles called macropinosomes,
which enter the large cellular vesicle trafficking network. These contents can then be
trafficked to the degradative compartment, where they are catabolized into nutrients
that can be used for bioenergetics and biosynthesis.
7
In 2013, Bar Sagi and others published a paper showing that macropinocytosis is
an amino acid supply route in Ras-mutant cells and that cultured cells with K-Ras
mutations survived longer in low concentrations of glutamine when cultured with
physiological concentrations of albumin. (Cells without K-Ras mutations did not.)
Imaging experiments showed that serum protein was taken up in macropinosomes
and degraded in lysosomes, establishing macropinocytosis as the primary mode
of uptake of extracellular protein to be degraded. Tumor xenograft experiments
compared xenografted cells from a K-Ras-mutant pancreatic cell line (MIA PaCa-
2) with xenografted cells from a K-Ras-wild-type pancreatic cell line (BxPC-3).
Xenografts were treated with 5-(N-Ethyl-N-isopropyl)-amiloride (EIPA), which in-
hibits macropinocytosis. EIPA targets a plasma membrane Na+-H+ antiporter [51]
and is too toxic to be administered systemically, so it was injected directly into the
tumors. MIA PaCa-2 xenografts were sensitive to EIPA; BxPC-3 xenografts were not.
The authors claim that these results “indicate that a reduction in macropinocytic
capacity may compromise tumor growth” [16]. Given the suspiciousness of both the
tumor models and the inhibitor, I am unconvinced by this particular experiment.
The paper as a whole was important, contributing the idea that cancerous epithelial
cells can use extracellular protein as a fuel source, but evidence that macropinocytosis
is a critical process in pancreatic tumor cells in vivo remained lacking.
In the second chapter of this thesis (the Cancer Research paper), we provide
evidence that freshly resected human pancreatic tumor cells do indeed engage in
macropinocytosis. This evidence came from Cosimo Commisso (a post-doctoral re-
searcher mentored by Bar Sagi at the time and the first author of the 2013 paper
discussed above). He showed that in slices of human pancreatic tumor tissue, tu-
mor cells, moreso than other cells present in the tumor, engage in macropinocytosis
[48]. These experiments provided valuable evidence from human pancreatic tumors,
but the evidence was indirect. They did not confirm that human pancreatic tumor
8
cells catabolize meaningful amounts of extracellular protein – amounts that can fuel
growth.
To directly test whether protein eating can fuel the growth of pancreatic tumors,
I used cultured pancreatic cancer cells derived from autochthonous mouse models of
pancreatic ductal adenocarcinoma (PDAC) [60]. I cultured these cells in a medium
that resembled, in an over-simplified way, the metabolic environment of a pancreatic
tumor. Relative to human serum, typical cell culture medium has very high concen-
trations of amino acids and very low concentrations of serum protein. Instead, I used
medium that lacked leucine, an essential amino acid, to mimic the amino acid-deficient
environment of pancreatic tumors, and I supplemented this leucine-free medium with
a roughly physiological concentration of serum protein (50 g/L bovine serum albu-
min). Since leucine is essential, we reasoned that cells would only grow if protein
eating were a viable alternative to monomeric amino acid uptake. Indeed, while most
cells switched into this leucine-free medium died, some survived and eventually grew
to confluence. I cultured these surviving cells indefinitely, and after several months,
the cells grew robustly in leucine-free medium supplemented with albumin. These
adapted cells even grew in medium without any free amino acids when albumin was
added [48]. This proved that protein eating could provide enough amino acids to fuel
growth.
Also included in the second chapter (the Cancer Research paper) is the first it-
eration of a method that uses isotope tracers to measure the rate of amino acid
release due to catabolism of extracellular protein. Using the method, I showed that
catabolism of extracellular protein can contribute substantially to amino acid pools
[48]. This iteration had its flaws – differences in factors such as cell number and
growth rate were not accounted for – but to my knowledge, this was the first time
that stable isotope tracers were ever used to estimate the rate of a major catabolic
process.
9
Following the publication of the Cancer Research paper, Craig Thompson and
others – including former post-doctoral researcher Wilhelm Palm – discovered that
inhibition of mechanistic target of rapamycin complex 1 (mTORC1) enables robust
growth of cells dependent on protein eating for growth. (They too cultured cells in
leucine-free medium supplemented with serum albumin.) Using fluorescent imaging
experiments, they showed that inhibition of mTORC1 promotes catabolism. These
experiments relied on a fluorescent marker of protein degradation called DQ-BSA.
DQ-BSA is bovine serum albumin (BSA) heavily labeled with the fluorescent dye
BODIPY. BODIPY self-quenches at high concentrations, so intact DQ-BSA does
not fluoresce. Upon degradation, however, DQ-BSA de-quenches. When DQ-BSA
was administered to untreated cells and cells treated with an mTOR inhibitor, the
lysosomes of the mTOR-inhibited cells emitted far more fluorescence – in some ex-
periments, ten-fold more than lysosomes of untreated cells – suggesting that more
degradation is happening in these cells [78].
Degradation, however, is not the only potential cause of increased DQ-BSA fluo-
rescence. In amino acid-starved cells, small lysosomes fuse to create large lysosomes;
once the large lysosomes have degraded their contents, they redistribute into small
lysosomes again. Inhibition of mTORC1 is known to block the reformation of small
lysosomes from large ones [135]. Thus, increased DQ-BSA fluorescence could be the
result of slower fluorophore turnover in mTOR-inhibited cells. In general, imaging
experiments are imperfect substitutes for flux measurements. The authors likely real-
ized these limitations, but at the time, there was no quantitative method to measure
protein eating flux to confirm their DQ-BSA findings.
In the third chapter of this thesis (a paper published in Molecular Cell in 2017),
I present the second iteration of the isotope tracer method that measures the rate of
amino acid release due to catabolism of extracellular protein. This iteration of the
method accounted for differences in cell number and growth rate, producing catabolic
10
rate estimates in terms of amino acids released from lysosomes per µL cell volume per
hour. Using this method, we confirmed that mTOR inhibition induces an increase
in catabolism of extracellular protein. The magnitude of this increase, however, was
much smaller than the imaging experiments suggested. We found that, in amino acid-
replete medium, mTOR-inhibited cells catabolized 50-75% more extracellular protein
than untreated cells. Unexpectedly, in amino acid-deficient media, cells increased
their catabolic rates in mTORC1-independent fashion. (mTORC1 activity persisted
in these cells despite amino acid deprivation.) Moreover, mTOR inhibition did not
substantially increase the rates of catabolism of amino acid-deprived cells. Thus,
we reasoned that mTOR inhibition enhances growth fueled by protein eating by
promoting survival, not by enhancing catabolism [73]. How does mTOR inhibition
promote survival? The answers we provide in this paper – by “restoring amino acid
balance” – are not entirely satisfying.
In the fourth chapter of this thesis (a manuscript in progress), I present the results
of a genome-wide screen designed to systematically identify genes critical for growth
fueled by protein eating. This screen was conducted at MIT in collaboration with
David Sabatini, Eric Lander, and Tim Wang, a graduate student co-mentored by
Sabatini and Lander at the time. We found that the genes critical for growth on
extracellular protein encode proteins that vary widely in function. Some proteins are
required for macropinocytosis, others for degradation of extracellular protein. The
gene most essential for growth fueled by protein eating (and not essential in amino
acid-replete conditions) was GCN2, a regulator of protein synthesis with no known
connection to macropinocytosis or protein degradation. Like mTOR inhibition, GCN2
suppresses translation initiation. Using isotope tracers, we showed that GCN2 activity
was required to increase catabolic rates in amino acid-deprived cells. The biochemical
mechanism underlying this increase remains unknown. My current hypothesis is
the proposed title of the manuscript: GCN2 upregulates translation of cathepsin L-
11
encoding mRNAs, increasing the degradative capacity of amino acid-deprived cells.
This hypothesis remains to be proven. If true, GCN2 and cathepsin L provide two
appealing targets for pancreatic cancer therapy.
In the fifth chapter of this thesis (a perspective published in Science in 2018), I
review the finding, by David Sabatini and others, that NUFIP1 is a ribosome receptor
that delivers ribosomes to lysosomes in nutrient-starved cells [129, 72]. Whereas
protein eating delivers material to lysosomes non-specifically – in principal, anything
in the extracellular space can be taken up and degraded by this process – the NUFIP1
pathway is specific to ribosomes. It remains unclear how many distinct degradative
pathways exist.
To conclude, I summarize the findings in this thesis and discuss future directions.
12
Chapter 2
Human Pancreatic Cancer Tumors
Are Nutrient Poor and Tumor
Cells Actively Scavenge
Extracellular Protein
This chapter is a paper published in Cancer Research in 2015 [48]. Figure numbering
and citation numbers from the original paper are preserved. Supplementary figures
can be found at the Cancer Research website.
2.1 Authors
Jurre J. Kamphorst, Michel Nofal, Cosimo Commisso, Sean R. Hackett, Wenyun Lu,
Elda Grabocka, Matthew G. Vander Heiden, George Miller, Jeffrey A. Drebin, Dafna
Bar-Sagi, Craig B. Thompson, and Joshua D. Rabinowitz
13
2.2 Abstract
Glucose and amino acids are key nutrients supporting cell growth. Amino acids
are imported as monomers, but an alternative route induced by oncogenic KRAS in-
volves uptake of extracellular proteins via macropinocytosis and subsequent lysosomal
degradation of these proteins as a source of amino acids. In this study, we examined
the metabolism of pancreatic ductal adenocarcinoma (PDAC), a poorly vascularized
lethal KRAS-driven malignancy. Metabolomic comparisons of human PDAC and be-
nign adjacent tissue revealed that tumor tissue was low in glucose, upper glycolytic
intermediates, creatine phosphate, and the amino acids glutamine and serine, two
major metabolic substrates. Surprisingly, PDAC accumulated essential amino acids.
Such accumulation could arise from extracellular proteins being degraded through
macropinocytosis in quantities necessary to meet glutamine requirements, which in
turn produces excess of most other amino acids. Consistent with this hypothesis,
active macropinocytosis is observed in primary human PDAC specimens. Moreover,
in the presence of physiologic albumin, we found that cultured murine PDAC cells
grow indefinitely in media lacking single essential amino acids and replicate once in
the absence of free amino acids. Growth under these conditions was characterized by
simultaneous glutamine depletion and essential amino acid accumulation. Overall,
our findings argue that the scavenging of extracellular proteins is an important mode
of nutrient uptake in PDAC.
14
2.3 Introduction
One of the most lethal forms of cancer is pancreatic ductal adenocarcinoma (PDAC)
[39]. Almost all cases of PDAC involve activating KRAS mutations [8]. In addition to
driving growth, KRAS induces metabolic changes including enhanced glucose uptake,
glycolytic flux, and glucose flux into hexosamines and ribose-5-phosphate [132]. In
contrast to other driver oncogenes such as PI3K that broadly increase glucose flux
throughout metabolism [117], oncogenic RAS impairs flux of glucose through pyruvate
dehydrogenase into the tricarboxylic acid (TCA) cycle [26, 27]. RAS-driven cells
instead rely heavily on glutamine as a TCA carbon source, with glutamine catabolism
through the TCA cycle and malic enzyme essential in pancreatic cancer cells [106].
Thus, RAS-driven cancer cells are comparatively less dependent on glucose than other
cancer cells [136].
Generation of significant ATP from substrates other than glucose requires oxygen,
whose availability in tumors is classically limited because of poor perfusion. Indeed,
PDAC tumors, which are characterized by poor vascularization and high interstitial
pressure, are typically hypoxic [23, 53]. Given the high metabolic demands of tumor
growth, poor perfusion may lead to limitation not only for oxygen but also nutrients
including glucose and free amino acids. Given the particular importance of glutamine
as a source of both usable nitrogen and TCA cycle carbon, glutamine can potentially
be a limiting nutrient for tumor growth. Consistent with this, studies in murine
tumor models in the 1940s and 1950s found lower free glutamine in the tumor than
corresponding normal tissue [92, 91].
A potential alternative to traditional uptake of monomeric amino acids via mem-
brane transport proteins is macropinocytosis, a process activated by mutant KRAS
[4, 16]. Macropinocytosis involves bulk uptake of extracellular constituents, includ-
ing proteins that can be subsequently digested in lysosomes into free amino acids.
Intriguingly, in cell culture, feeding of albumin to RAS-driven cells enabled their
15
survival and proliferation in low glutamine, and such survival and proliferation was
dependent upon macropinocytosis [16]. Albumin has been reported to accumulate
in tumors, likely due to a combination of leaky vasculature and lymphatic deficiency
[107]. Thus, it is conceptually possible that plasma protein leakage from tumor vas-
culature provides a nutrient source for cancer cells. The extent to which this actually
occurs in human tumors, however, has not yet been explored. Nor has it been shown
whether such scavenging is sufficient to provide amino acids other than glutamine in
biologically significant quantities.
Here, we investigate protein scavenging in PDAC. Metabolomic analysis of freshly
isolated human PDAC tumor specimens (compared with benign adjacent tissue) re-
vealed that the tumors are low in glucose, upper glycolytic intermediates, glutamine,
and serine. PDAC tumors also accumulated amino acids that are useful primarily
for protein synthesis. Although uptake or synthesis of monomeric amino acids would
be expected to yield each amino acid in quantities balanced with total demand, pro-
tein catabolism instead produces amino acids in proportion to their abundance in
the catabolized protein. Those amino acids that are consumed by multiple anabolic
processes (such as glutamine) would accordingly become depleted relative to those
used solely or primarily for protein synthesis. Thus, the observed pattern of amino
acid depletion and accumulation in human PDAC suggests a reliance on protein scav-
enging. Consistent with this, we find that primary human PDAC specimens display
enhanced macropinocytosis. Moreover, we show that cultured pancreatic cancer cells
can obtain sufficient amino acids via protein scavenging to grow with albumin as the
sole amino acid source, and that this mode of growth is associated with glutamine
depletion and essential amino acid accumulation.
16
2.4 Materials and Methods
Cell culturing and amino acid dropout experiments KRPC cells were kindly
provided by S. Lowe (Memorial Sloan- Kettering Cancer Center, New York, NY)
[60]. These cells were harvested from a murine tumor following orthotopic injection
of murine pancreas progenitor cells with endogenous KRASG12D that were addition-
ally engineered to have MYC expression and silenced (shRNA-mediated) P53. Cell
lines were routinely passaged in DMEM (Mediatech) with 25 mmol/L glucose and
4 mmol/L glutamine and supplemented with 10% (v/v) fetal bovine serum (FBS;
HyClone), 25 IU/mL penicillin, and 25 mg/mL streptomycin (MP Biomedicals), and
split at 80% confluence. For amino acid dropout experiments, pyruvate-free DMEM
was prepared from powder (Cellgro, cat. no. 10-017-CV) by adding glucose (25
mmol/L), salts, vitamins, phenol red, and amino acids, except for the amino acid to
be omitted. For single amino acid dropout experiments, cells were plated in DMEM
at 10% confluence. After 24 hours, cells were switched to dropout DMEM supple-
mented with 5% dialyzed FBS (Thermo). This medium was further supplemented
with 0% or 5% cell culture-grade bovine serum albumin (BSA; Sigma), which was
not fatty acid free. For assaying growth in amino acid-free DMEM, cells passaged in
leucine-free DMEM supplemented with 5% BSA were plated at 20% confluence. After
24 hours, medium was changed to amino acid-free DMEM with 5% BSA. Medium
was replaced as needed.
Imaging and cell counting For imaging, cells were fixed in 10% TCA for 15 min-
utes, and images were obtained using a Nikon Eclipse TE2000-U microscope operated
by Q-Capture Pro software (QImaging). KRPC cells growing in leucine-free medium
supplemented with BSA were seeded at 10% confluence and switched to fresh leucine-
free medium after 24 hours. Cell proliferation was assessed by cell number determina-
17
tion using a Countess Automated Cell Counter (Invitrogen) or by determining total
packed cell volume (PCV) using PCV tubes (Techno Plastic Products).
Stable isotope tracing experiments Medium with fully 13C- and 15N-labeled glu-
cose and amino acids, otherwise equivalent to DMEM, was reconstituted from indi-
vidual components (13C- and 15N-DMEM). This medium contained unlabeled sodium
bicarbonate and vitamins and was supplemented with 5% dialyzed FBS and 1% v/v
penicillin streptomycin (MP Biomedicals). After five doublings in this medium, cells
were seeded at low cell density and switched to specified 13C- and 15N-labeled me-
dia. After 24 hours, metabolites were extracted and analyzed by liquid chromatogra-
phy/mass spectrometry (LC/MS). Growth of KRPC cells in amino acid-free medium
is dependent on pregrowth in leucine-free medium. Thus, cells were grown for five
doublings in 13C- and 15N-DMEM then switched to leucine-free 13C- and 15N-DMEM
with 5% BSA. After two to three doublings in this medium, cells were switched to
(i) complete 13C- and 15N-DMEM with 5% BSA and (ii) amino acid-free 13C- and
15N-DMEM with 5% BSA. Cells in complete medium were grown for 24 hours be-
fore metabolite extraction. Cells in amino acid-free medium were grown for 48 hours
before metabolite extraction, with medium replaced after 24 hours.
Metabolite extraction from cultured cells For analysis of intracellular amino
acids, medium was aspirated and plates were rinsed three times with room tem-
perature PBS. Metabolism was quenched and metabolites extracted in -80◦C 80:20
methanol:water extraction solution. After 15 minutes at -80◦C, plates were scraped
and cell extracts were transferred to 15-mL conical tubes. Cell suspensions were vor-
texed, centrifuged at 3,000 g for 5 minutes, supernatant was kept, and cellular debris
was reextracted with -80◦C 80:20 methanol:water extraction solution. The resulting
suspension was centrifuged and the supernatant was combined with the supernatant
from the first extraction. The resulting solution was dried under nitrogen flow and re-
18
suspended in high-performance liquid chromatography (HPLC)-grade water. Twenty
microliters was added to 80-μL methanol in addition to 10-μL triethylamine and 2-μL
benzyl chloroformate and incubated at room temperature for 30 minutes, to derivatize
and thereby enhance measurement sensitivity of amino acids.
Tissue collection and metabolite extraction procedure Patients undergoing
surgical resection of pancreatic tumors consented to collection and analysis of their
resected tissues. Following partial pancreas resectomy, approximately 100 mg seg-
ments of tumor and adjacent tissue were isolated and immediately frozen in liquid
nitrogen, and the diagnosis of PDAC was confirmed histologically. The samples were
shipped overnight on dry ice to Princeton University (Princeton, NJ) and stored in
liquid nitrogen until analysis.
Tumor and benign adjacent tissues of the same patient were prepared and analyzed
in parallel. The samples were weighed and then pulverized by agitation with stainless
steel balls at liquid nitrogen temperature (CryoMill, Retsch, 25 Hz for 3 minutes). The
pulverized tissue was mixed by vortexing with 2 mL of -80◦C 80:20 methanol:water,
and split into two 1-mL replicate samples, which were set aside to extract for 5 minutes
at -80◦C. Each sample was centrifuged to isolate the soluble extract, and the insoluble
material was extracted twice more with 1 mL 80:20 methanol:water for 5 minutes at
0◦C and the supernatant again isolated by centrifugation. The supernatants from the
three rounds of extraction were combined, dried under nitrogen gas, and reconstituted
in LC/MS grade water (1 mL of water per 25 mg initial tissue weight).
LC/MS analysis Cell culture and tissue samples were analyzed by three separate
LC/MS systems: (i) Stand-alone orbitrap MS (Exactive; Thermo Scientific) oper-
ating in negative full scan mode at 100,000 resolution coupled to C18 ultra per-
formance reversed-phase ion pair LC [62], (ii) triple quadrupole mass spectrometer
(TSQ Quantum Discovery Max; Thermo Scientific) operating in negative multiple
19
reaction monitoring mode coupled to C18 high-performance reversed-phase ion pair
LC, and (iii) triple quadrupole mass spectrometer (TSQ Quantum Ultra; Thermo
Scientific) operating in positive multiple reaction monitoring mode coupled to Hy-
drophilic interaction liquid chromatography (HILIC) chromatography [3]. Metabo-
lites were identified by accurate mass (<5 ppm deviation, Exactive) or characteristic
fragmentation product (triple quads), in combination with retention time match to
validated standards, using in-house software [70]. Linearity of response was verified
by running 2-fold dilutions of most samples and observing 2-fold decreases in peak
intensities. For a subset of tissue samples, amino acid concentrations were determined
using 13C-labeled amino acid standards. For quantification of amino acid pool sizes
in cultured cells, metabolite intensities were normalized by PCV. Cell culture leucine
measurements were obtained using a modified HILIC method [80].
Ex vivo Macropinocytosis assay Fresh PDAC tumor tissue obtained from surgi-
cal resections was cut into slices with an approximate 3-mm cuboidal shape. Tissue
was immersed into serum-free DMEM containing 1 to 2 mg/mL of TMR-dextran and
incubated at 37◦C for 20 to 30 minutes. Tissue was rinsed twice in PBS and imme-
diately frozen in optimal cutting temperature (OCT) compound. Tissue processing
and image analysis was performed as previously described [17].
Data normalization and processing of tissue metabolomic data Ion counts
were normalized to correct for differences in total metabolite abundances across sam-
ples, and for any sample-to-sample drift in the overall instrument response factor. A
normalization factor (γi) was calculated for each LC/MS run i. To calculate γi, every
known metabolite peak Pmi in LC/MS run i was quantified, and compared with the
median value of peak m across all samples, µm. The scaling factor was calculated
according to Eq. 2.1:
γi = median(Pmiµm
)(2.1)
20
and ion counts were corrected accordingly: P ∗mi = Pmi
γi.
The normalized ion count matrix was log2-transformed and averaged over repli-
cates. When a metabolite was measured on multiple instruments, the results were
averaged.
Significance testing of tumor/benign adjacent tissue metabolite differences
To determine whether a subset of metabolites are systematically higher or lower in
cancerous than in benign adjacent pancreatic tissue, P values were computed using a
paired t test with the null distribution generated by bootstrapping [25]. This method
was chosen as a more conservative alternative to determining the test significance
against a t-distribution, because the parametric t-distribution approach makes the
assumption that the log-metabolite abundances are each normally distributed. This
assumption is not valid for many metabolites.
For a given metabolite m, measured in n patients, tumor metabolite abundance
(Cmi) was compared within the same patient to the benign tissue metabolite abun-
dance (Bmi). These abundances, [Bmi, Cmi], were jointly standardized so that they
collectively have a mean of 0 and a standard deviation of 1. The systematic difference
between pairs can be captured by a paired t test statistic (Eq. 2.2).
tm =
∑ni=1
(Bmi−Cmi)n√∑n
i=1(Bmi−Cmi)
2
n−1
n
(2.2)
To generate samples for an empirical null distribution, we need to generate data
where the systematic variation between the benign and cancer samples has been
removed and then use the remaining variation to determine how often we would have
seen such a large systematic difference (tm) by chance. To generate these null data, the
paired difference was removed from the benign abundances (Eq. 2.3a), the modified
abundances were recentered (Eq. 2.3b) and then these residuals were corrected for
21
the 1 degree of freedom eliminated by removing the paired difference (Eq. 2.3c), as
per Efron and Tibshirani [25].
B∗mi = Bmi −
∑ni=1(Bmi − Cmi)
n(2.3a)
Mm = mean[B∗mi, Cmi] (2.3b)
Brmi = (B∗
mi −Mm)
√n
n− 1(2.3c)
Crmi = (C∗
mi −Mm)
√n
n− 1
For each bootstrap sample (500,000 were used), n patients were randomly sampled
(with replacement) and the pairs of null data from these patients (drawn from Brm
and Crm) formed Br
m and Crm. Because there are no systematic differences between
the means of Brm and Cr
m, t statistics of these null pairs (Eq. 2.4) can be used to
approximate the distribution of t statistics expected under the null hypothesis.
tbmr =
∑ni=1
(Bbmi−Cb
mi)
n√∑ni=1
(Bbmi
−Cbmi
)2
n−1
n
(2.4)
A P value for metabolite m can be calculated by determining how often a t
statistic more extreme than the observed statistic (tm) would be expected under the
null hypothesis (Eq. 2.5).
pm = 1−∑R
r |tm| > |tbmr|R
(2.5)
False discovery rate calculation Correction for multiple hypothesis testing fol-
lowed the procedure of Storey and Tibshirani [108]. Briefly, if there are no metabolites
22
that systematically differ between benign tissue and tumors (i.e., all of the metabolites
are true negatives), then the expected distribution of P values across all metabolites
is uniform on [0, 1]. To the extent that some of the metabolites do systematically
differ, the associated P value histogram will be a mixture of true positive P values
(skewed toward zero) and uniform true negatives. This histogram allows us to esti-
mate the fraction of true negatives in the dataset: πo. We can then find a P value
cutoff, q, corresponding to the desired false discovery rate (FDR), by taking the ratio
of the expected number of false positives (πomq) to the number of P values less than
q. Metabolites with P values less than this q-value were treated as discoveries.
2.5 Results - Metabolomic analysis of human
PDAC tumors
Paired PDAC tumor and benign adjacent tissue specimens were acquired by surgical
resection from 49 patients (Fig. 1A). To minimize metabolic changes during the
tissue acquisition process, the preferred technique is freeze-clamping in situ with
liquid nitrogen-cooled Wollenberger tongs [127, 87]. Such in situ freeze-clamping
might compromise clinical outcomes, for example, by precluding proper identification
of tumor margins. Accordingly, we instead relied on excised samples, with the surgical
approach chosen to maintain perfusion until just before excision. Thereafter, tumor
and benign adjacent tissue samples were rapidly quenched in liquid nitrogen and
stored at ≤ −70◦C. Metabolome analysis was conducted at the whole sample level,
without distinguishing between epithelial and stromal tumor components. Paired
samples were extracted in parallel and analyzed by three complementary LC/MS
methods that enabled quantitation of 127 water-soluble metabolites across a majority
of the samples [62].
23
Of the 127 metabolites examined, 57 displayed significantly different levels in
pair-wise analysis of tumor and benign adjacent tissue (Fig. 1B). The most strongly
depleted metabolites in tumors were glutamine, cytidine (whose amino group is de-
rived from glutamine), guanidoacetic acid (a precursor to creatine phosphate, which
was also down), glucose, and several phosphorylated compounds derived from glucose
(glucose-6-phosphate, sedoheptulose-7-phosphate, and glycerol-3-phosphate). The
most strongly increased metabolites in tumors were the DNA base thymine and several
hydrophobic essential amino acids (valine, isoleucine/leucine, and tryptophan). The
tryptophan degradation product kynurenine, which has immunosuppressant bioactiv-
ity [30], was also strongly elevated in the tumors.
Within central carbon metabolism glucose, glucose-6-phosphate, and fructose-6-
phosphate were all decreased in the tumors, as were most TCA cycle compounds.
In contrast, the 3-carbon glycolytic intermediates dihydroxyacetone phosphate and
3-phosphoglycerate were slightly increased, as was lactate. Collectively, these obser-
vations are consistent with increased propensity for aerobic glycolysis but decreased
glucose availability in the tumors.
As a class, the 20 proteogenic amino acids showed particularly strong changes be-
tween the tumor and benign adjacent tissue with some amino acids strongly increased
in the tumors, and others strongly decreased. There was the propensity for “nonessen-
tial” amino acids to be depleted in the tumors, whereas “essential” amino acids ac-
cumulated to higher levels (compare green and red bars in Fig. 1B). This trend,
however, was not absolute. Most importantly, while glutamine was the most depleted
amino acid (with an average depletion of 2.5-fold verified using 13C-labeled internal
standards; Supplementary Table S1), glutamate (which differs from glutamine by a
single amine moiety) was slightly increased. This led us to consider the hypothesis
that the observed patterns of amino acid depletion and accumulation might not re-
flect amino acid essentiality, but rather rates of individual amino acid consumption
24
by anabolic pathways. Specifically, the two most depleted amino acids, glutamine
and serine, play key anabolic roles as amine- and one-carbon donors, respectively. In
addition, serine is a key precursor of lipid head groups, and both CDP-choline and
CDP-ethanolamine levels were increased in tumors. In contrast, levels of amino acids
used primarily for protein synthesis were increased in PDAC.
This pattern of amino acid levels in the tumors is consistent with amino acids
being acquired by protein catabolism to fuel anabolic metabolism (Fig. 1C) [16].
Although uptake or synthesis of monomeric amino acids would be expected to produce
each amino acid in the appropriate amount, protein catabolism instead produces all
amino acids in proportion to their abundance in protein. Those amino acids that
are consumed by multiple anabolic processes (such as glutamine) accordingly become
depleted relative to those used solely or primarily for protein synthesis.
25
Figure 2.1: Cancer Research paper - Figure 1
26
2.6 Results - Macropinocytosis in PDAC tumors
Macropinocytosis mediates the endocytic uptake of extracellular protein in RAS-
driven cancer cells and murine tumors [16]. Therefore, to examine whether this
protein internalization mechanism is active in human PDAC tumor tissue, freshly
acquired human tumor specimens were incubated with high molecular weight
tetramethylrhodamine-conjugated dextran (TMR-dextran), an established marker
of macropinosomes, and intracellular uptake of TMR-dextran was assessed by fluo-
rescent microscopy of tissue sections. TMR-positive macropinosomes were detected
in CK19-positive tumor cells (Fig. 2). Quantitatively lower, but nevertheless sub-
stantial, TMR-dextran staining was also detected in CK19-negative tumor tissue,
which may include both stromal cells and PDAC cells that have undergone epithelial-
mesenchymal transition (thereby losing CK19; Supplementary Fig. S1). In contrast,
few macropinosomes were detected in normal adjacent tissue (Supplementary Fig.
S2). Although some intratumoral variability was observed, stimulated macropinocy-
tosis was evident in each of the five analyzed PDAC tumor samples. These data
indicate that macropinocytosis is an attribute of human pancreatic tumors and that
pancreatic cancer cells have the capacity to take up fluids and their constituents
from the tumor extracellular environment.
27
Figure 2.2: Cancer Research paper - Figure 2
2.7 Results - Support of cultured tumor cell
growth by albumin in the absence of free
amino acids
Prior work has shown that macropinocytosis enables KRAS-activated cells to prolif-
erate in low glutamine media supplemented with albumin [16]. However, the extent
to which protein scavenging can provide other amino acids in quantities sufficient to
support cellular proliferation remains unknown. To address this question, we incu-
bated KRAS-driven pancreatic cells in medium lacking one or more essential amino
acids and supplemented with physiologic levels of albumin. Murine-derived pancreatic
cancer cells with oncogenic KRASG12D and silenced P53 (KRPC cells) [60], growing
in complete medium were switched to leucine-deficient medium either (i) not sup-
plemented with BSA or (ii) supplemented with the typical circulating concentration
28
of BSA (50 g/L or 5%). As expected, KRPC cells did not survive in leucine-free
medium without added BSA. In contrast, when switched to leucine-free medium sup-
plemented with BSA, leucine removal initially resulted in extensive cell death, but
surviving cells proliferated indefinitely (months) with a doubling time of approxi-
mately 24 hours (Fig. 3A and B and Supplementary Fig. S3). After 24 hours in this
medium, the intracellular leucine concentration in these cells was approximately 12.5
pmol/mL cell volume, roughly 100-fold lower than in the same cells grown in com-
plete medium, but twice as high as in cells cultured in the absence of both leucine and
BSA (Fig. 3C). KRPC cells were also able to proliferate indefinitely in the absence
of lysine or phenylalanine (Supplementary Fig. S4).
One trivial explanation for the growth of KRPC cells in medium without leucine
is trace leucine contamination from serum or other additives in quantities sufficient
to support cell growth. To rule out this possibility, we added 12 mmol/L leucine to
leucine-free medium that was not supplemented with BSA. LC/MS measurements
revealed that, while this leucine-spiked medium contained more free leucine than the
BSA-supplemented leucine-free medium (Supplementary Fig. S5), it did not support
cell growth. Therefore, the continuous proliferation of KRPC cells in leucine-free
medium supplemented with albumin is not due to contaminating leucine.
Given that KRPC cells can use intact protein as their sole source of leucine,
lysine, or phenylalanine, we wondered whether these cells could proliferate in BSA-
supplemented medium without any free amino acids. We switched KRPC cells grow-
ing in leucine-free medium to amino acid-free medium either (i) not supplemented
with BSA or (ii) supplemented with 5% BSA. Although the amino acid-free medium
without BSA supplementation did not support cell survival, cells switched to the
BSA-supplemented amino acid-free medium grew to confluence, doubling once over
a period of 5 days (Fig. 3D and E). Thus, KRAS-driven cancer cells in vitro can
29
grow faster than PDAC tumor cells in vivo solely through scavenging and subsequent
catabolism of extracellular protein.
Figure 2.3: Cancer Research paper - Figure 3
2.8 Results - Isotope tracing of serum protein
catabolism
The capability of RAS-mutant cells to proliferate in the absence of essential amino
acids suggests that serum protein catabolism might contribute substantially to their
amino acid pools. To quantitatively measure this contribution, we developed an iso-
30
tope tracer-based method enabling separate quantitation of (i) amino acids initially
taken up as monomers and (ii) amino acids acquired via catabolism of serum protein.
We prepared growth medium in which natural glucose and all amino acids were re-
placed by uniformly 13C- and 15N-labeled glucose and amino acids at standard DMEM
concentrations (13C- and 15N-DMEM). KRPC cells were grown in this medium for five
doublings, such that cellular protein in the resulting population was predominantly
labeled. Then, cells were transferred to 13C- and 15N- medium equivalent to DMEM
except that free amino acids were present at 10% of their DMEM concentrations (13C-
and 15N- DMEM, 10% AA). These amino acid concentrations more closely resemble
physiologic conditions [e.g., average human PDAC glutamine concentration is 700
mmol/L (Supplementary Table S1), and 10% of DMEM glutamine concentration is
400 mmol/L]. Use of this medium is important for detecting unlabeled amino acids
coming from protein catabolism, whose concentrations are otherwise overwhelmed by
the high amino acid concentrations in full DMEM. The cells were cultured for 24 hours
in this medium, and metabolites were extracted and analyzed by mass spectrometry
(Fig. 4A).
In cells grown without added albumin, less than 12% of intracellular and less than
5% of extracellular essential amino acids were unlabeled. These observed unlabeled
amino acids are presumably derived primarily from catabolism of the available serum
protein (the cells were cultured in 5% FBS). The addition of albumin dramatically
increased the observed levels of unlabeled (i.e., serum protein-derived) intracellular
amino acids (Fig. 4B). Moreover, substantial concentrations of unlabeled amino acids
were also observed in the medium, suggesting rapid exchange of intracellular and
extracellular amino acid pools (Fig. 4C). Contribution from albumin to intracellular
amino acid pools was also observed in MIA PaCa-2 and E3 human pancreatic cancer
cells (Supplementary Fig. S6). Thus, we were able to directly track amino acids
derived from protein catabolism in cultured PDAC cells.
31
We next sought to validate the hypothesis that catabolism of serum protein oc-
curs in the lysosome by treating KRPC cells with bafilomycin A1, which impairs
lysosomal function by inhibiting the vacuolar-type H+-ATPase [133]. Treatment of
KRPC cells with bafilomycin A1 resulted in a dose-dependent reduction of amino
acids derived from serum protein catabolism (Fig. 4D and E). Treatment with EIPA,
a canonical inhibitor of macropinocytosis [16], also results in a reduction in serum
proteinderived amino acids in KRPC cells (Supplementary Fig. S7). Taken together,
these data demonstrate that, even in the absence of amino acid deprivation, lyso-
somal degradation of extracellular protein contributes substantially to PDAC amino
acid pools.
Figure 2.4: Cancer Research paper - Figure 4
32
2.9 Results - Amino acid patterns in cells fed by
serum protein macropinocytosis
We next used our isotope tracing strategy to confirm that cells grown in amino acid-
free media supplemented with 5% BSA acquire most of their amino acids from the
unlabeled extracellular protein. With the exception of alanine, serine, and glycine,
which were synthesized from glucose (which, in contrast to the pancreatic cancer mi-
croenvironment, was abundant in the culture media), all amino acids were largely
unlabeled, that is, derived from extracellular protein (Fig. 5A). In addition, amino
acid nitrogen in these cells was predominantly extracellular protein derived (Supple-
mentary Fig. S8).
We next asked how amino acid concentrations differed between cells grown in
amino acid-free medium supplemented with physiologic albumin and cells grown in
standard DMEM. Growth in the albumin-supplemented, amino acid-free medium
resulted in strong depletion of intracellular glutamine after 24 hours in culture (Fig.
5B). However, paradoxically we observed accumulation of essential amino acids to
levels comparable with, and in some cases greater than, those seen in cells grown in
rich medium (Fig. 5C). Moreover, significant concentrations of free amino acids were
excreted into the medium (Fig. 5D and E). As the tumor metabolomic analysis does
not allow for discrimination between cancer cell intracellular and extracellular amino
acid pools, we are unable to determine whether amino acid excretion occurs in vivo. In
addition, perhaps due either to differences in nutrient availability between the actual
tumor and this cell culture model (e.g., of glucose, free amino acids, albumin, and
other proteins) or to the substantial fraction of stroma in the tumor, there was not a
direct correspondence between amino acid concentration changes in the PDAC tumors
and cells grown in albumin-supplemented amino acid-free medium. Nevertheless,
33
these data confirm that growth fed by scavenging of extracellular protein can lead to
glutamine depletion and essential amino acid accumulation.
Figure 2.5: Cancer Research paper - Figure 5
2.10 Discussion
The definition of cancer is the uncontrolled division and growth of cells. To facilitate
this, cancer cells need nutrients that can be used to generate the necessary energy
and cellular building blocks. It is commonly assumed that cancers primarily rely
on glucose and glutamine as their nutrient sources. The dependence on glucose is
certainly true for many cancers and is perhaps best illustrated by the clinical value of
FDG-PET, which uses a glucose analogue to image and stage tumors [116]. However,
there are exceptions, with a notable example being PDAC: less than 25% of tumors are
markedly FDG-PET positive, and up to 35% do not take up FDG above background
34
[40]. This may be caused by restricted blood perfusion due to the high interstitial
pressure and desmoplasia that are characteristic for PDAC [15]. Despite a limited
availability of free glucose and glutamine, PDAC is notoriously aggressive, suggesting
that other nutrients may play an important role in fueling PDAC growth.
What could these alternative nutrients be? We recently reported the ability of
KRAS-transformed cells to uptake extracellular protein through macropinocytosis
[16], and to use serum lipids to support growth [47]. PDAC cells reuse the amino
acids and fatty acids from these extracellular proteins and lipids to support growth,
much as they also rely on recycling of intracellular materials via autophagy to sur-
vive metabolic stress [131]. Leaky tumor vasculature and lymphatics can result in
accumulation of albumin and other serum proteins in the tumor interstitium. Scav-
enging of proteins and lipids diminishes demand for de novo biosynthesis and thus
the need for glucose and free glutamine-derived carbon, reducing equivalents (NADH
and NADPH), and ATP.
As some key metabolic pathways are oxygen-dependent (TCA cycle and fatty acid
desaturation), bypassing them facilitates growth in hypoxia. Oncogenic RAS, even
in nutrient and oxygen replete conditions, reduces oxygen consumption and increases
macropinocytosis and lipid scavenging [26, 27, 47]. It thus appears that RAS prepares
cells to survive and grow in metabolically harsh conditions, including hypoxia.
Here, we provide an in-depth analysis of the metabolic state of primary human
PDAC tumors. PDAC tumor tissues are heterogeneous, characterized not only by the
presence of tumor cells, but also cancer-associated fibroblasts, immune cells, and ex-
tracellular matrix. Our analyses do not allow us to differentiate between the metabolic
contributions from each of these compartments. Rather, they represent the metabolic
state of the tumor tissue as a whole. Using this approach, we found that relative to the
benign adjacent tissues, PDAC tumors were consistently low in both glucose and glu-
tamine. We further found a pattern of amino acid levels consonant with what would
35
happen if the tumors were largely reliant on protein scavenging (Fig. 1C): a differen-
tial utilization of amino acids for purposes other than protein synthesis (nucleotide
and lipid synthesis etc.) leaves amino acids primarily used for protein synthesis to
build up, whereas amino acids used for other purposes as well (glutamine, serine, and
alanine) deplete.
By culturing cells in the absence of one or all free essential amino acids, we were
able to demonstrate the capacity of extracellular protein scavenging to provide amino
acids to support growth. Intriguingly, cells were able to grow rapidly and indefinitely
by this mode of consumption in the absence of free leucine, lysine, or phenylalanine
(which are all relatively abundant in albumin; Supplementary Fig. S9), and tran-
siently in the absence of all free amino acids. Moreover, in the absence of all amino
acids, BSA scavenging was sufficient to produce elevated intracellular levels of selected
essential amino acids.
An increased scavenging ability allows cells to access vast resources of cellular
building blocks and energy: assuming a total plasma protein concentration of 75 g/L,
the amino acid content of plasma proteins exceeds free amino acids by approximately
200-fold. Thus, while poor perfusion may limit flow of all nutrients through the
tumor, in such flow restriction, plasma protein may increase in relative importance as
an amino acid source. In addition, plasma proteins are also a major potential energy
source, exceeding energy available in glucose by about 75-fold. Similarly, the ability
to scavenge fatty acids from various serum lipids, rather than free fatty acids alone,
increases available fatty acids by at least 4-fold [47].
Although blood flow is often low in PDACs due to the high interstitial pressure,
tumor blood vessels are leaky. In combination with the fact that tumors are lym-
phatic deficient, this may result in plasma protein accumulation [107]. In light of
this, the recent clinical success of a protein-drug conjugate albumin-paclitaxel (nab-
36
paclitaxel, Abraxane) in PDAC is particularly tantalizing [121], and warrants further
investigation into metabolic scavenging and how it can be exploited therapeutically.
37
Chapter 3
mTOR Inhibition Restores Amino
Acid Balance In Cells Dependent
on Catabolism of Extracellular
Protein
This chapter is a paper published in Molecular Cell in 2017 [73]. Figure numbering
from the original paper is preserved. Supplementary methods and figures are available
on-line at the Molecular Cell website.
3.1 Authors
Michel Nofal, Kevin Zhang, Seunghun Han, and Joshua D. Rabinowitz
38
3.2 Abstract
Scavenging of extracellular protein via macropinocytosis is an alternative to
monomeric amino acid uptake. In pancreatic cancer, macropinocytosis is driven
by oncogenic Ras signaling and contributes substantially to amino acid supply.
While Ras signaling promotes scavenging, mTOR signaling suppresses it. Here, we
present an integrated experimental-computational method that enables quantitative
comparison of protein scavenging rates across cell lines and conditions. Using it, we
find that, independently of mTORC1, amino acid scarcity induces protein scavenging
and that under such conditions the impact of mTOR signaling on protein scavenging
rate is minimal. Nevertheless, mTOR inhibition promotes growth of cells reliant
on eating extracellular protein. This growth enhancement depends on mTORC1’s
canonical function in controlling translation rate: mTOR inhibition slows transla-
tion, thereby matching protein synthesis to the limited amino acid supply. Thus,
paradoxically, in amino acid-poor conditions the pro-anabolic effects of mTORC1 are
functionally opposed to growth.
39
Figure 3.1: Molecular Cell paper - Graphical Abstract
40
3.3 Introduction
Amino acids are required substrates for protein synthesis and thus cell growth. While
some organisms can synthesize all proteinogenic amino acids from primitive carbon
and nitrogen sources, mammals cannot. For this reason, mammalian cells have been
thought to strictly depend on the availability of amino acid monomers in their extra-
cellular environment to support growth. Recently, it was shown that Ras signaling
stimulates an alternative route of amino acid acquisition whereby cells take up ex-
tracellular protein via macropinocytosis and catabolize it in lysosomes to yield free
amino acids. This process enables K-Ras-mutant pancreatic ductal adenocarcinoma
(PDAC) cells to survive and proliferate despite amino acid scarcity [16, 19].
The mechanistic target of rapamycin complex 1 (mTORC1) is a master growth
regulator that promotes anabolism [58]. In the presence of amino acids, mTORC1 is
recruited to the cytoplasmic surface of lysosomes, where it can be activated by growth
factor signaling [95]. Upon activation, it phosphorylates multiple targets, which col-
lectively stimulate amino acid uptake and protein synthesis [64] while suppressing
autophagy [46]. Amino acid depletion renders mTORC1 inactive, and protein syn-
thesis rates decline as a result. Meanwhile, cells engage in autophagy – i.e. they
catabolize pre-existing intracellular protein, yielding amino acids necessary to pre-
vent starvation. These amino acids reactivate mTORC1, attenuating autophagy and
restoring protein synthesis [135]. The implications of mTORC1 reactivation in the
context of prolonged starvation remain poorly understood.
Recently, Palm et al. showed that mTORC1 activity inhibits the growth of cancer
cells fed the major serum protein, albumin, in place of free essential amino acids.
Torin1, an ATP-competitive mTOR inhibitor, promoted growth in such conditions.
The authors reasoned that in addition to blocking autophagy, mTORC1 suppresses
the catabolism of extracellular protein [78]. This claim was supported by an assay for
extracellular protein degradation which uses a fluorescently labeled form of bovine
41
serum albumin (BSA) known as DQ-BSA, whose fluorogenic component is initially
hidden – many self-quenching BODIPY molecules are conjugated to each albumin
molecule – and only de-quenches once this albumin has been degraded [88]. Indeed,
Torin1 increases protein scavenging as measured with DQ-BSA. While an elegant tool
for visualizing serum protein catabolism, DQ-BSA fluorescence does not provide an
absolute measure of protein catabolic rate.
To this end, we previously reported a method that, using stable isotope tracers,
distinguishes amino acids derived from the catabolism of extracellular protein from
amino acids imported as monomers. Cells are pre-incubated for multiple generations
in medium containing uniformly 13C-labeled amino acids (13C-AA medium), such
that intracellular amino acids and cellular protein become almost completely labeled.
Cells are then switched to 13C-AA medium supplemented with physiologic levels of
unlabeled BSA. At this point, when cells scavenge and degrade the unlabeled albumin,
they release unlabeled amino acids into otherwise labeled amino acid pools. High
amounts of unlabeled amino acids produced by these cells reflect fast serum protein
uptake and catabolism [48].
In murine pancreatic cancer cells grown in physiological nutrient conditions, we
found that almost half of both intracellular and extracellular amino acid pools were
derived from BSA (i.e. unlabeled), demonstrating that protein catabolism can be
a major contributor to amino acid pools in pancreatic cancer [48]. The measured
unlabeled fractions, however, depend not only on the rate of serum protein catabolism,
but also on the number of cells present in the experiment and on their rate of protein
synthesis. Thus, differences in amino acid labeling patterns between cell lines or
growth conditions do not reliably reflect differences in protein scavenging rates.
Here, we present an integrated experimental-computational method that enables
reliable and quantitative comparison of protein scavenging rates across cell lines and
conditions. We then apply this method to investigate the mechanism by which mTOR
42
inhibition enhances the growth of cells fed by protein scavenging. We find that,
independently of mTORC1, amino acid scarcity strongly turns on protein scavenging,
and that under such conditions the impact of Torin1 on protein scavenging rate is
small. Inhibition of mTOR by Torin1 promotes growth of protein-scavenging cells
instead by decreasing their translation rate and thereby matching protein synthesis
to the limited amino acid supply.
3.4 Results - Isotope-Tracer Method Measures
Amino Acid Release Due to Extracellular
Protein Catabolism
Our general strategy for measuring extracellular protein catabolic rate involves pre-
labeling cells with 13C-AA medium and then feeding them a mixture of 13C-AA
medium and unlabeled albumin. Protein scavenging is then the only source of un-
labeled amino acids, and the rate of appearance of such amino acids can be used to
calculate the protein scavenging rate. The challenge is making such measurements in
a manner that accurately reflects per cell protein scavenging rates.
To this end, as cells grew in 13C-AA medium and unlabeled albumin, we took se-
rial time point measurements of intracellular amino acid labeling, extracellular amino
acid labeling, and total cell volume (Figure 1A). We further developed a simple model
of amino acid metabolism, which includes the following reactions for production and
consumption of intracellular amino acid monomers: (i) import and export from the
cell via amino acid transporters, (ii) incorporation into protein, (iii) catabolism of
extracellular protein, and (iv) catabolism of intracellular protein (i.e. via autophagy)
(Figure 1B) [138, 102]. This model applies exclusively to essential amino acids, which
are not synthesized, and it assumes that catabolism of essential amino acid monomers
43
is negligible. Using this model, the dynamic cell growth data and the extracellular
amino acid labeling data (both unlabeled fractions and absolute amounts) are suffi-
cient to uniquely determine the per cell rate of protein scavenging. Because intracel-
lular amino acid pools mix rapidly with extracellular pools (Supplementary Figure
1), data from intracellular amino acids is not required, making this method relatively
facile (see Methods).
When implementing this method, the rate of production of each essential amino
acid from albumin can be measured independently. Thus, as a first test of the method,
we assessed whether the measurements for different amino acids were in agreement,
focusing on five amino acids that we can easily and accurately measure (Figure 1C,
D). We anticipated that the release rates of different amino acids would reflect their
relative abundances in BSA. Indeed, this was the case: there are 59 lysines and only
17 histidines in BSA, and the measured rate for lysine exceeded that for histidine
by approximately 59:17 fold, with the other amino acids intermediate between these
two (Figure 1D). Dividing the release rate of each amino acid by the number of times
that amino acid appears in BSA yields the protein scavenging rate in units of moles
albumin per cell per unit time (Figure 1E).
As validation of this method, we sought to confirm the effect of constitutive
Ras activation on extracellular protein catabolism. To do so, we compared the
protein scavenging rate of immortalized baby mouse kidney cells (iBMK) with or
without constitutively active Ras or Akt alleles. While Akt activation did not in-
duce any change, constitutive Ras signaling roughly doubled the rate of extracellular
protein catabolism, consistent with the long-standing observation that Ras induces
macropinocytosis (Figure 1F) [4].
44
Figure 3.2: Molecular Cell paper - Figure 1
As further validation, we examined the protein scavenging rate of cells before and
after extended growth in conditions that select for accelerated protein scavenging. For
these experiments, we used KRPC cells, which were originally isolated from sponta-
neously arising, K-Ras-driven murine pancreatic tumors that resemble human PDAC
45
[60]. These cells were grown in leucine-free medium supplemented with 5% BSA [48].
Initial growth was slow, but after prolonged culture (100 generations), the resulting
adapted population (KRPCA cells) doubled approximately every 24 hours despite
the absence of free leucine (Figure 2A). Using the isotope-tracer method, we found
that KRPCA cells have roughly 5-fold higher rates of extracellular protein catabolism
(Figure 2B). Thus, we provide a quantitative method for assessment of the rate of
albumin catabolism by protein-eating cells.
Figure 3.3: Molecular Cell paper - Figure 2
46
3.5 Results - Impact of Intracellular Protein
Catabolism on Scavenging Measurements
An important element of the isotope-tracer method is the extended pre-labeling in
13C-AA medium. Extended pre-labeling ensures that autophagy and other modes of
intracellular protein degradation do not generate unlabeled amino acids and thereby
do not confound measurements of extracellular protein catabolism. To examine the
relative magnitudes of intracellular protein degradation and extracellular protein scav-
enging, we conducted analogous experiments with only 1 h pre-labeling, which is insuf-
ficient to substantially label intracellular protein. These experiments were conducted
in murine embryonic fibroblasts harboring an oncogenic K-RasG12D allele (K-RasG12D
MEFs) and KRPCA cells. In K-RasG12D MEFs, the pre-labeling duration did not
significantly impact the production rates of unlabeled amino acids, suggesting that
extracellular protein scavenging predominates over intracellular protein degradation.
In contrast, in KRPCA cells, we observed a two-fold increase in unlabeled amino acid
production with the brief pre-labeling, indicating similar magnitudes of extracellu-
lar and intracellular protein catabolism (Supplementary Figure 2). To confirm that
the measurements of extracellular protein scavenging do not reflect autophagy in the
murine pancreatic cancer cells, we used a well-established KPC cells line harboring
inducible shRNA against the essential autophagy gene Atg5. With the extended pre-
labeling that results in selective measurement of extracellular protein degradation,
knockdown of Atg5 did not significantly impact the measured scavenging rate (Sup-
plementary Figure 3), validating the selectivity of this isotope-tracer approach for
extracellular protein scavenging.
47
3.6 Results - Excessive mTOR Inhibition Slows
Growth on Extracellular Protein
Recent evidence suggests that Ras-activated cells, even without adaptation, can grow
robustly on extracellular protein if mTORC1 activity is suppressed [78]. We wondered
if KRPCA cells achieve high levels of protein scavenging in amino acid-rich medium
by suppressing mTORC1 signaling. We observed a modest decrease in mTORC1
signaling in adapted KRPC cells, as measured by the phosphorylation of S6 kinase 1,
ribosomal protein S6, and 4E-BP1 (Figure 2C).
Given that mTOR inhibition has been shown to promote protein scavenging and
that mTORC1 remains at least partially active in the adapted KRPC cells, which
have high basal scavenging rates, we tested the impact of the ATP-competitive mTOR
inhibitor Torin1 on KRPC cell growth. These experiments were conducted for a range
of Torin1 doses (100-2000 nM) in both parental KRPC and KRPCA cells, in amino
acid-replete, leucine-free, arginine-free, and glutamine-free medium, all supplemented
with 5% BSA (Figure 3). Among these amino acid-deficient conditions, we anticipated
that leucine deprivation would be most easily overcome by albumin scavenging, as
leucine is the most abundant amino acid in albumin. In contrast, we anticipated that
glutamine deprivation would be hardest to overcome, as glutamine is not particularly
abundant in albumin but required by cells in unusually large amounts. We expected
arginine deprivation to be intermediate. Deprivation of other amino acids was not
examined.
48
Figure 3.4: Molecular Cell paper - Figure 3
As expected, in amino acid-replete conditions, mTOR inhibition slowed the growth
of both parental and adapted KRPC cells. Importantly, however, cells doubled in 24
hours despite high doses of Torin1, indicating that these cells are capable of consid-
erable growth even when mTOR signaling is pharmacologically inhibited. In leucine-
deficient conditions, parental cells grew faster in the presence of Torin1, but optimal
growth was achieved in the middle of the Torin1 dose range, indicating that some
mTOR signaling is beneficial. Strikingly, in KRPCA cells, only the lowest dose of
Torin1 promoted growth; higher doses slowed it. While parental cells struggled to
grow without arginine or glutamine, KRPCA cells were able to grow without these
amino acids, with optimal growth occurring in the middle of the Torin1 dose range.
Collectively, these findings show that for cells fed by protein scavenging, mTOR
inhibition has both growth-promoting and growth-suppressing effects. The relative
strengths of these effects seem to depend on the protein scavenging rate of the treated
cells and the inherent difficulty of overcoming the amino acid deprivation. In more
deprived states (e.g. parental cells, glutamine-free medium), mTOR signaling in-
hibits growth. Conversely, in more favorable states (e.g. adapted cells, leucine-free
medium), mTOR signaling promotes growth.
49
3.7 Results - Amino Acid-Deficiency Induces Pro-
tein Scavenging Flux Independently of mTOR
To further investigate the effects of mTORC1 inhibition, we measured, using the
above isotope-tracer approach, the effect of Torin1 on protein scavenging flux. In both
KRPCA cells and K-RasG12D MEFs cultured in amino acid-replete medium, Torin1
increased protein scavenging in dose-dependent fashion (Figure 4A). The largest in-
crease we observed was less than 2-fold, however, whereas Palm et al. reported that
in K-RasG12D MEFs, Torin1 induced a 10-fold increase in DQ-BSA fluorescence and
a 5-fold increase in growth in leucine-free medium.
We next asked if the effect of Torin1 on protein scavenging rates depends on
amino acid availability. We measured the rates of extracellular protein catabolism
in the same three amino acid drop-out media as above in the presence or absence of
high-dose Torin1 (1000 nM). Amino acid deprivation increased protein catabolism at
least as much as high-dose Torin1 (Figures 4B). Interestingly, the degree to which
scavenging was induced aligned with the severity of amino acid starvation. For exam-
ple, in K-RasG12D MEFs, leucine deprivation, the least severe, increased scavenging
by 60%; glutamine deprivation, the most severe, by 220%. One might expect that
a reduction in mTORC1 activity upon amino acid deprivation accounts for these in-
creases. However, mTORC1 activity persists in these cells (Figure 4C). Thus, amino
acid deprivation turns on scavenging independently of mTOR.
50
Figure 3.5: Molecular Cell paper - Figure 4
We were initially puzzled that mTORC1 was active in amino acid-deficient con-
ditions. Others have demonstrated, however, that in cells deprived of amino acids
for long periods of time, mTORC1 signaling is re-activated once protein catabolic
programs begin to take effect [78, 135]. Indeed, we observed that when K-RasG12D
51
MEFs were switched to media lacking all amino acids, phosphorylation of S6 Kinase 1
immediately declined, but eventually returned, although phosphorylation of another
key substrate of mTORC1, 4E-BP1, was maintained throughout the time course.
Notably, removal of leucine alone resulted in no initial decline in the phosphoryla-
tion of either mTORC1 substrate (Supplementary Figure 4). Thus, at least over 24
h, leucine-, arginine-, and glutamine-deprived cells maintain mTORC1 activity. In
fact, at 24 h, glutamine-deprived cells displayed increased mTORC1 signaling, po-
tentially because glutamine deprivation resulted in accumulation of essential amino
acids within the cell (Supplementary Figure 5).
Given this persistent mTORC1 activity, we studied the impact of mTOR inhi-
bition on protein scavenging in amino acid-deprived cells. In leucine-free medium,
Torin1 increased extracellular protein catabolism by only 14% in KRPCA cells and
by 7% in K-RasG12D MEFs. While these enhancements in scavenging flux may con-
tribute to the pro-growth effects of mTOR inhibition in scavenging cells, they are
not quantitatively commensurate with the substantial enhancements in cell growth
observed upon mTOR inhibition. Thus, the growth-promoting effects of Torin1 in
cells reliant on protein scavenging extend beyond enhancing protein catabolism.
3.8 Results - mTOR Inhibition Induces Punctate
DQ-BSA Fluorescence
We were struck by the difference in magnitude of the effect of Torin1 on protein
scavenging flux measured here via isotope tracing (less than 2x) versus previously
via DQ-BSA fluorescence (roughly 10x). To address this discrepancy, we repeated
the DQ-BSA fluorescence experiment which produced this result, using identical cells
and conditions to Palm et al. Specifically, DQ-BSA was added concurrently with 250
nM Torin1 to K-RasG12D MEFs grown in serum-free DMEM, and cells were imaged
52
after 6 h [78]. As expected, Torin1 induced a visible increase in DQ-BSA fluorescence
(Figure 5A) and a corresponding rightward shift in the histogram of pixel fluorescence
intensity (Figure 5B). Thus, we confirmed that mTOR inhibition induces an increase
in DQ-BSA fluorescence.
We next sought to quantify this induction. When we included all measurable
fluorescence in our calculation, we found that Torin1 increases DQ-BSA fluorescence
per cell by less than two-fold, in line with our isotope tracing results (Figure 5C).
Some fluorescence, however, is inevitably noise. To minimize noise, standard methods
for quantification of fluorescence ignore lower intensity signals, only using pixels that
exceed an arbitrarily chosen fluorescence intensity threshold. We found that the choice
of fluorescence intensity threshold greatly affected the apparent magnitude of the
Torin1 effect: low thresholds produced effects less than 2-fold, while high thresholds
produced effects greater than 5-fold (Figure 5D). To explore this phenomenon further,
we divided the distribution of pixel intensities into five ranges and calculated the sum
of the intensities within each range. This revealed that mTOR inhibition dramatically
increases only very high-intensity fluorescence, which accounts for a minor portion
of the total fluorescent signal while having a modest effect on overall fluorescence
(Figure 5E). The relative contributions of each range of pixel intensities are apparent
in discretized images, which enable simultaneous visualization of all ranges of green
fluorescence (Figure 5F and Supplementary Figure 6).
Comparing our results to those of Palm et al. [78], we note no major differences
in the raw data: analysis of our data using a high fluorescence intensity threshold
yields images and quantitative results comparable to Palm et al. We believe, however,
that lower thresholds are more accurate, as they encompass a substantially greater
amount of total fluorescent signal and give quantitative results in line with the our
isotope-tracer data. In essence, the isotope-tracer data, which was not available to
53
Palm et al., inform the proper choice of threshold for quantitation of the fluorescence
data.
Focusing specifically on the high-intensity fluorescence which was strongly in-
duced by Torin1, we observed this fluorescence in discrete punctae that overlap with
lysosomal staining (Figure 5A). One possible explanation for this strong increase in
lysosomal DQ-BSA signal is that mTORC1 may simultaneously inhibit protein scav-
enging and promote egress of scavenged material from the lysosome. In summary, the
combined isotope tracing and fluorescence results show that mTOR, while profoundly
impacting lysosomal DQ-BSA fluorescence accumulation, has a modest overall impact
on protein scavenging rate.
54
Figure 3.6: Molecular Cell paper - Figure 5
55
3.9 Results - mTOR Inhibition Restores Amino
Acid Balance and Prevents Cell Death in
Amino Acid-Deprived Cells
How does mTOR inhibition promote growth of amino acid-deprived cells on extracel-
lular protein if not by directly increasing their rate of extracellular protein catabolism?
mTORC1 is well-known for its role in regulating protein synthesis, phosphorylating
multiple proteins which collectively activate 5’ cap-dependent translation [64]. We
reasoned that reduction of protein synthesis rates upon Torin1 treatment might pre-
vent cells deprived of free extracellular amino acids from promoting translation to the
point of cellular stress.
As others have shown and we have confirmed, mTORC1 activity persists in amino
acid-deprived cells fed extracellular protein, perhaps because scavenged protein feeds
directly into the lysosomal amino acid pool that is sensed by mTORC1 [78]. Corre-
spondingly, protein synthesis rates are not limited by low mTORC1 activity in these
cells. However, protein scavenging cannot support the high protein synthesis rates
of cells growing in copious free monomeric amino acids. GCN2, which senses amino
acid depletion by binding uncharged tRNAs [7, 124, 22], attenuates global translation
independently of mTORC1, while inducing specific translation of genes involved in
maintaining amino acid homeostasis, including the transcription factor ATF4 [35].
ATF4 induces expression of proteins that collectively promote cell survival during
amino acid deprivation by up-regulating amino acid uptake and enhancing protein
folding capacity. Nevertheless, this cellular response to amino acid starvation (known
as the Integrated Stress Response) fails to prevent apoptosis in cells which are chron-
ically unable to translate mRNAs into properly folded proteins. Moreover, some
proteins induced by the Integrated Stress Response, such as CHOP, promote cell
death if amino acid starvation remains unresolved [34, 139].
56
To test our hypothesis that mTOR inhibition promotes growth on extracellu-
lar protein by reducing translation rates and thereby preventing severe amino acid
starvation, we measured the expression of Integrated Stress Response proteins. In K-
RasG12D MEFs deprived of leucine, we observed a strong induction of both ATF4 and
CHOP, regardless of whether cells were supplemented with 5% BSA. This induction
was suppressed in Torin1-treated cells, suggesting that mTOR inhibition in amino
acid-deprived cells restores amino acid homeostasis (Figure 6A). We next sought to
measure cell death directly, to confirm that mTORC1 activity results in apoptosis.
After 48 hours in leucine-free media, close to 50% of cells grown in leucine-free medium
were either apoptotic or dead. mTOR inhibition prevented cell death. This preven-
tion did not depend on the presence of added BSA. Thus, the effects of mTORC1
on the survival of amino acid-deprived cells do not directly depend on the uptake or
catabolism of extracellular protein (Figure 6B).
To verify that mTOR inhibition prevents cell death by suppressing protein syn-
thesis, we tested the effect of direct inhibition of translation on the viability of
leucine-deprived cells. Low doses of harringtonin, which inhibits translation initi-
ation, prevented apoptosis and cell death to a similar extent as mTOR inhibition
(Figure 6C). We were not, however, able to stimulate cell growth in leucine-free,
BSA-supplemented medium with harringtonin (i.e. to replicate the pro-growth ef-
fects of Torin1), suggesting that the growth-promoting effects of mTOR inhibition
go beyond non-specific translation inhibition. This is consistent with the idea that
mTOR inhibition coordinately suppresses translation of a specific subset of genes and
increases protein catabolism. Collectively, these results show that, in cells reliant on
protein scavenging, excessive translation can result in lethal amino acid depletion.
57
Figure 3.7: Molecular Cell paper - Figure 6
While K-RasG12D MEFs require exogenous translation inhibition to maintain
amino acid balance when using extracellular protein in place of free leucine, KRPCA
cells have adapted to such conditions and can grow robustly without pharmacological
mTOR inhibition. We hypothesized that these cells rely on other translational
regulators to properly tune protein synthesis rates to limited amino acid availabil-
ity. We reasoned that GCN2, which slows translation upon amino acid depletion,
might play such a role. Using CRISPR-Cas9 technology, we generated KRPCA
cell lines deficient in GCN2 activity, as measured by lack of ATF4 induction in
leucine-free medium (Supplementary Figure 7A). We tested the ability of GCN2-
deficient KRPCA cells to grow in leucine-free medium supplemented with 5% BSA.
Remarkably, these GCN2-deficient cells almost completely lost the ability to grow
using extracellular protein in place of free leucine (Supplementary Figure 7B-C). If
this defect is due to excessive translation, it should be rescued by mTOR inhibition.
Indeed, GCN2-deficient KRPCA cells, much like parental KRPC cells, benefited
from high-dose Torin1 treatment (Supplementary Figure 7D). Thus, the ability to
58
turn down translation rates to match limited amino acid availability is essential for
growth via protein scavenging.
3.10 Discussion
All cells require amino acids for cell growth. Classically, mammals maintain a steady
concentration of circulating amino acids, which individual cells import as needed. If
perfusion is impaired, however, cells may struggle to support growth requirements us-
ing only amino acids from the environment. This appears to be the case in pancreatic
tumors, which are marked by dense fibrosis and poor perfusion [66, 85]. Pancreatic
tumor cells, driven by K-Ras mutations, mitigate the shortage of monomeric amino
acids in their immediate environment by taking up and catabolizing extracellular pro-
tein. This process enables these cells to survive and proliferate despite amino acid
deprivation [16, 19, 48].
Recently, mTOR inhibition was shown to promote growth of amino acid-deprived
cells on extracellular protein. To explain this, Palm et al. proposed that mTORC1 ac-
tivity represses extracellular protein catabolism and that mTOR inhibition alleviates
this repression [78]. In the present study, we demonstrate that mTORC1 signal-
ing does not prevent extracellular protein catabolism in amino acid-deprived cells.
Rather, these cells simultaneously maintain mTOR activity and increase protein eat-
ing. Nevertheless, cells relying on extracellular protein for amino acids cannot support
the high rates of translation that are possible in amino acid-replete conditions. Be-
cause mTORC1 remains active, these cells are prone to death by over-translation.
Thus, mTOR inhibition enhances growth on extracellular protein in part by restrict-
ing translation and restoring amino acid balance (Figure 7).
59
Figure 3.8: Molecular Cell paper - Figure 7
We also show that cells deprived of free leucine or glutamine increase extracel-
lular protein catabolism while maintaining mTORC1 signaling. This implicates an
mTOR-independent signaling pathway as an activator of this process during amino
acid deprivation. The other ubiquitous amino acid sensing pathway involves GCN2,
which, upon binding to uncharged tRNA, phosphorylates and inhibits translation
initiation factor eIF2α [7, 124, 22]. While inhibiting translation of most mRNAs,
phosphorylation of eIF2α promotes translation of ATF4 and other transcription fac-
tors which induce genes involved in adaptation to amino acid starvation, including
amino acyl-tRNA synthetases, amino acid transporters, and protein folding chap-
erones [34, 35]. Other proteins expressed upon eIF2α phosphorylation are involved
in diverse cellular processes such as expansion of the endoplasmic reticulum, which
houses a substantial fraction of nascent peptides [34]. It is tempting to speculate that
these proteins might also include unknown activators of protein scavenging.
This study highlights the inability of cancer cells fed extracellular protein to op-
timally adjust levels of mTORC1 signaling to match amino acid availability. These
cells maintain mTORC1 activity even when free leucine or glutamine is absent from
60
the extracellular environment. mTORC1 signaling is insensitive to such amino acid
scarcities in part because multiple amino acids are activators of mTORC1, but even
when no free amino acids are present, mTORC1 signaling, initially suppressed, can
be re-activated by protein catabolism via either autophagy [135] or catabolism of ex-
tracellular protein [78]. Thus, while mTORC1 can sense acute amino acid starvation,
it is insufficient to balance biosynthesis and catabolism in response to chronic amino
acid deprivation in cells with constitutive growth factor signaling. In accordance
with this, proline starvation was recently shown to result in mTORC1 hyperactiva-
tion, unresolved ER stress, and decreased tumorigenesis of multiple cancer cell lines
[94]. In a different context, dysregulated mTORC1 renders cells dependent on an
exogenous supply of unsaturated fatty acids (whose production requires oxygen) in
hypoxia [134]. Thus, excessive mTORC1 signaling can push cells into fatal stress
when biosynthetic substrates are limiting.
These findings have implications for mTOR inhibition in cancer therapy. While
mTOR inhibitors have shown anti-tumor activity in certain cancers, they have unex-
pectedly had limited efficacy in most cases. In assessing the therapeutic potential of
these agents, the deleterious activity of mTORC1 in cells deprived of amino acids may
have been overlooked. We find that moderate mTOR inhibition protects these cells
from cell death by restricting translation. Moreover, if these cells catabolize extra-
cellular protein, mTOR inhibition facilitates robust growth. The pro-survival effects
of mTOR inhibition on amino acid-deprived cells may explain the minimal clinical
activity of mTOR inhibitors on pancreatic tumors [45, 128], which are glutamine-poor
[48]. In accordance with this idea, Palm et al. showed that inhibition of mTORC1
enhances the growth of pancreatic tumors in a murine PDAC model. Specifically,
rapamycin decreased the fraction of proliferating cells in outer, vascularized regions
of these tumors, but increased the proliferation of cells in interior, hypovascularized
regions [78]. The present work suggests that mTOR inhibition promotes the growth
61
of these cells not only by promoting protein scavenging, but also by reducing biosyn-
thetic demands. As a result, cells enduring prolonged nutrient shortages can stably
assimilate biosynthetic substrates for anabolism (e.g. by degradation of extracellu-
lar protein) while simultaneously avoiding the lethal cellular stresses associated with
starvation. More generally, many chemotherapeutics target upregulated biosynthesis
in cancer cells. Our results emphasize the importance of finding new ways to amplify
cellular stresses associated with excessive biosynthesis, rather than focusing solely on
slowing these biosynthetic processes down. Indeed, in tumors poorly supplied with
nutrients, slowing anabolism can paradoxically promote growth.
3.11 Materials and Methods - Cell lines
Cell lines and culture All cell lines used in this study are listed in the Key Re-
sources Table. All cells were propagated in DMEM with 25 mM glucose and 4 mM
glutamine and without pyruvate (Mediatech). DMEM was supplemented with 10%
FBS (HyClone) and 25 IU/mL penicillin and 25 mg/mL streptomycin (MP Biomed-
icals), unless specified otherwise.
Knockout cell lines Oligonucleotides targeting murine Gcn2 (also known as
Eif2ak4) were cloned in lentiCRISPR v2 (Addgene #52961) [96]. Virus was pro-
duced in HEK293FT cells, and KRPCA cells were infected. Infected cells were
selected in puromycin, and clonal knockout cell lines were produced by isolation of
single cells from this infected population. Oligonucleotide sequences are listed in the
Key Resources Table.
62
3.12 Materials and Methods - Measuring catabolism
of extracellular protein
Custom Media Preparation Custom media were prepared using DMEM powder
containing all DMEM salts and vitamins, low glucose, and no amino acids or pyruvate
(US Biologicals). Glucose was added to a final concentration of 25 mM glucose, and
sodium bicarbonate to a final concentration of 3.7 g/L. Pyruvate was not added to
any media. To facilitate custom media preparation, concentrated (20-100X) amino
acid stock solutions were prepared and stored at 4◦C. Such solutions were used to
add all amino acids except glutamine (unstable) and tyrosine (insoluble), which were
added directly in powder form. 13C-AA medium, with or without supplemented
BSA, contained uniformly 13C-labeled histidine, lysine, phenylalanine, threonine, and
valine; all other amino acids were unlabeled.
In 13C-AA medium not supplemented with BSA and in all amino acid-deficient
media, amino acid concentrations were identical to standard DMEM (glutamine: 4
mM; isoleucine, leucine, lysine, threonine, and valine: 0.8 mM; arginine, glycine,
serine, phenylalanine: 0.4 mM; cystine, histidine, methionine, and tyrosine: 0.2 mM;
and tryptophan: 0.078 mM). For 13C-AA medium supplemented with BSA, 13C-
labeled amino acids were added at reduced concentrations to facilitate amino acid
uptake measurements (lysine, threonine, and valine: 0.32 mM; phenylalanine: 0.16
mM; and histidine: 0.08 mM). All BSA-supplemented media contained 5% w/v BSA.
All custom media was adjusted to pH 7.2 immediately before sterile filtration and
was additionally supplemented with 5% dialyzed FBS.
Stable isotope-labeled amino acids (including U-13C6 L-Lysine:2HCl, U-13C9 L-
Phenylalanine, U-13C4 L-Threonine, U-13C5 L-Valine, and U-13C6 L-Histidine) were
from Cambridge Isotope Laboratories. All other components were standard tissue
63
culture-grade reagents (Sigma). Tissue culture-grade BSA, which was not delipidated,
was from Sigma.
Isotope-Tracer Experiments Cells were grown for five doublings in 13C-AA
medium, as described above. After five doublings, cells were seeded at low cell
density in 60 mm tissue culture dishes and switched to 2 mL of 13C-AA medium
supplemented with 5% BSA. After 16 hr and 24 hr (and, where indicated, additional
time points), medium amino acids and total cellular volume were measured as below.
Absolute concentrations were determined by comparison of peak intensities in sam-
ples of interest and samples from fresh medium, in which amino acid concentrations
are known.
Metabolite Extraction and LC-MS Analysis For analysis of intracellular amino
acids, medium was aspirated and plates were rinsed three times with room temper-
ature PBS. Metabolism was quenched and amino acids extracted in ice-cold 80:20
methanol:water extraction solution. Plates were scraped and cell extracts were trans-
ferred to eppendorf tubes, which were vortexed and centrifuged at 16,100 g for 5
min. The resulting supernatant was dried under nitrogen flow and resuspended in
HPLC-grade water. 40 μL of the resulting solution was added to 160 μL HPLC-grade
methanol in a new tube. 10 μL triethylamine and 2 μL benzyl chloroformate were
added sequentially, and the resulting mixture was vortexed and incubated at room
temperature for 30 min to derivatize and thereby enhance measurement sensitivity of
amino acids.
For analysis of amino acids in culture medium, 50 μL of medium was directly
added to 200 microliters of HPLC-grade methanol. This mixture was vortexed then
centrifuged at 16,100 g for 5 min. 200 μL supernatant was transferred to a new tube.
10 μL triethylamine and 2 μL benzyl chloroformate were added sequentially, and the
resulting mixture was vortexed and incubated at room temperature for 30 min.
64
After derivatization, samples were diluted such that amino acids fell within the
linear range of a triple quadrupole mass spectrometer (TSQ Quantum Discovery Max;
Thermo Scientific), operating in negative multiple reaction monitoring mode, coupled
to C18 high-performance reversed-phase ion pair liquid chromatography [63, 61]. Data
were analyzed using open-source software [70].
Extracellular Protein Catabolism Rate Computation To derive an expression
for the rate of amino acid release due to serum protein catabolism, we start with the
following basic relationship: any cellular reaction rate (in units of moles per unit time
per unit cell volume) is equal to the total amount of product being produced by this
reaction in all cells (in units of moles per unit time) divided by the total volume of all
cells. In this case, for a given amino acid, the rate of amino acid release due to serum
protein catabolism is equal to the amount of amino acid being released by all cells
divided by total cell volume. Recalling that amino acids generated by extracellular
protein scavenging are unlabeled:
VAArelease =dAA0/dt
V ol(t)(3.1)
After integrating this equation with respect to time, the rate of amino acid release
is equal to the total amount of amino acid released by all cells over the course of the
experiment divided by the time-integral of total cellular volume:
VAArelease =AA0(T )∫ T0V ol(t)dt
(3.2)
Unlabeled amino acids released by extracellular protein catabolism can meet one
of three fates: they can end up as (i) intracellular amino acid monomers, (ii) amino
acid monomers in the medium, or (iii) amino acids which have been incorporated
into cellular protein. (We assume catabolism of essential amino acid monomers is
negligible.) Thus:
65
VAArelease =AA0
intra(T ) + AA0extra(T ) + AA0
prot(T )∫ T0V ol(t)dt
(3.3)
Because the aggregate volume of cells is very small relative to the volume of the
medium in each dish, the first term in the numerator of Equation (3.3) is negligible.
The second term in the numerator, which represents the molar amount of unlabeled
amino acids in the medium at the end of the experiment, is directly measurable.
The amount of unlabeled amino acids in protein at the end of the experiment was
determined indirectly, assuming recycling of cellular protein is negligible (Figure S2):
AA0prot(T ) = Vsynth
∫ T
0
AA0cyto
AAtotalcyto
(t)× V ol(t)dt (3.4)
We assume metabolic steady state to derive a simple expression for Vsynth:
Vsynth = Vin − Vout + VAArelease (3.5)
Substituting Equations (3.4) and (3.5) into Equation (3.3) gives us an expression
for the rate of amino acid release:
VAArelease =AA0
extra(T ) + (Vin − Vout + VAArelease)∫ T0
AA0cyto
AAtotalcyto
(t)× V ol(t)dt∫ T0V ol(t)dt
(3.6)
Finally, we solve for the rate of amino acid release due to extracellular protein
catabolism:
VAArelease =AA0
extra(T ) + (Vin − Vout)∫ T0
AA0cyto
AAtotalcyto
(t)× V ol(t)dt∫ T0V ol(t)dt−
∫ T0
AA0cyto
AAtotalcyto
(t)× V ol(t)dt(3.7)
Finally, dividing each amino acid release rate by the number of times that amino
acid appears in BSA yields protein scavenging flux estimates:
66
Vserumproteincatabolism =VAAreleaseαAA
(3.8)
To demonstrate how to compute the rate of amino acid release from extracellular
protein catabolism and corresponding protein scavenging flux, we provide an example
in which we compute the rate of lysine release in K-RasG12D MEFs growing in amino
acid-replete medium, using data shown in Figure 1C-E. K-RasG12D MEFs pre-grown
for five doublings in 13C-AA medium were switched to 2 mL of 13C-AA medium
supplemented with 5% BSA. The first term in the numerator of Eq. (3.7) is directly
measurable: after 24 hr, we measured 22,700 pmol unlabeled lysine in the medium:
AA0extra(T = 24h) = 22, 700 pmol (3.9)
The second term in the numerator is equal to the net uptake rate multiplied by
the time-integral of the product of the instantaneous unlabeled amino acid fraction
and the instantaneous total cell volume. Net uptake rate can be measured by tracking
amino acid abundance in the medium over time and normalizing to total cell volume.
In this example, we found that net lysine uptake was 2,900 pmol / μL cell / hr:
Vin − Vout = 2, 900 pmol / μL cell / hr (3.10)
To calculate the product of the instantaneous total cell volume and the instan-
taneous cytosolic unlabeled amino acid fraction integrated with respect to time, we
first fit the dynamic amino acid labeling data and the dynamic total cell volume data,
separately, to exponential functions of the following form:
f(t) = Aekt (3.11)
67
After fitting, the equations describing the unlabeled fraction of lysine in the
medium over time (in hours) and the total cell volume (in μL) over time (in hours)
are the following:
AA0cyto
AAtotalcyto
(t) = 0.014e(0.0756)t (3.12)
V ol(t) = 2.718e(0.0278)t (3.13)
Multiplication of Equation (3.12) by Equation (3.13) gives a single exponential
function under the integral in Eq. (7). Integration with respect to time (from 0 h to
24 hr) yields 4.19 μL cell × hr.
∫ T
0
AA0cyto
AAtotalcyto
(t)× V ol(t)dt = 0.040e(0.0756)t+(0.0278)tdt = 4.19 μL cell× hr (3.14)
The first term in the denominator is equal to the time-integral of total cellular
volume. For this, we can use the fitted exponential function describing cellular growth
from the previous step. We found that this integral was equal to 92.3 μL cell × hr.
∫ T
0
V ol(t)dt = e1.0+(0.028)tdt = 92.3 μL cell× hr (3.15)
The second term in the denominator, which also appears in the numerator and
was calculated above, is equal to 4.19 μL cell × hr. Plugging in Equations (3.9)-(3.15)
into Eq. (3.7):
VAArelease =22, 700 + (2, 930× 4.19 = 12, 300)
92.3− 4.19= 397 pmol / μL cell / hr (3.16)
68
Thus, the release rate of lysine from extracellular protein catabolism is 397 pmol
/ L cell / hr. To compute the corresponding protein scavenging flux, we divide this
number by the number of lysines per BSA molecule (59), as per Eq. (8), to yield the
following protein scavenging flux: 6.73 pmol / uL cell / hr:
Vserumproteincatabolism =VAAreleaseαAA
=397
59= 6.73 pmol / μL cell / hr (3.17)
As a final note, the equation for the rate of amino acid release due to extracellu-
lar protein catabolism includes a term containing the unlabeled amino acid fraction
in the cytoplasm over time. For this, we can either use the intracellular unlabeled
fraction, which contains cellular compartments other than the cytoplasm, or the ex-
tracellular unlabeled fraction. We observed that intracellular amino acid pools rapidly
exchange with amino acids in the medium: when we switched cells growing in stan-
dard unlabeled medium to 13C-AA medium, intracellular amino acid pools became
predominantly labeled (>90%) in roughly 10 min (Figure S1). Given this rapid ex-
change, we use extracellular amino acid labeling to represent cytosolic labeling in our
calculations. This has the benefit of requiring only extracellular, not intracellular,
amino acid measurements.
3.13 Materials and Methods - Other experimental
methods
Proliferation Assays For absolute measurements of proliferation (i.e. using cell
volume, cell number), 500K (parental KRPC) or 200K (adapted KRPC) cells were
seeded in standard 60 mm tissue culture dishes in DMEM supplemented with 5%
FBS. After 24 hr, cells were washed once with PBS and switched to amino acid-
69
deficient medium supplemented with 5% (w/v) BSA. Cell number was measured
using a Countess Automated Cell Counter (Invitrogen), and total cell volume was
measured using Packed Cell Volume tubes (Techno Plastic Products).
For relative measurements of proliferation (i.e. using absorbance of resorufin),
60K (parental KRPC) or 20K (adapted KRPC) cells were seeded in standard 24-well
tissue culture plates in DMEM supplemented with 5% FBS. After 24 hr, cells were
washed once with PBS and switched to amino acid-deficient medium supplemented
with 5% (w/v) BSA. After the indicated time in culture, cells were washed twice with
PBS, and standard DMEM supplemented with 10% FBS and 0.1 mg/mL resazurin,
but without additional BSA, was added. After 2 hr, absorbance was measured.
Western Blotting Cells were washed 3x with PBS, then lysed with ice-cold RIPA
buffer (Cell Signaling) with cOmplete protease inhibitor and PhosSTOP phosphatase
inhibitor cocktails (Roche). Soluble lysate fractions were isolated by centrifugation
at 16,100 g for 10 min. Relative protein content was estimated using total cellular
volume as a surrogate, and equal amounts of protein per sample were analyzed by
SDS-PAGE and Western Blotting.
Fluorescence Microscopy 5,000 cells were seeded in DMEM supplemented with
10% FBS in each well of a fibronectin-coated 8-well Chamber Slide (Nunc Inc). After
48 hr, cells were washed once with serum-free DMEM and switched to medium con-
taining 1 mg/mL DQ Green BSA, 50 nM LysoTracker Red, and 0.5 ug/mL Hoechst.
After 6 hr, cells were washed three times with PBS and fixed in 4% paraformalde-
hyde for 15 minutes. After three more washes to remove fixative, the polystyrene
chamber was removed, mounting medium was applied, and a coverslip was mounted.
The mounting medium was allowed to set overnight, and samples were imaged on a
Nikon A1 Confocal Microscope, with imaging parameters set such that no pixels were
saturated. Images were analyzed in Matlab.
70
Cell Viability Measurements Cell viability was assayed by flow cytometric de-
tection of caspase activity. After 48 hr in the specified condition, medium from each
sample was collected. Cells were washed once with room temperature PBS, which was
added to the collected medium. Cells were then detached with trypsin and added to
the collected medium and saline. The resulting cell suspension was centrifuged for 5
min at 2,000 g, and the pellet was resuspended in DMEM supplemented with 2% FBS,
containing CellEvent Caspase-3/7 Green Detection Reagent (Invitrogen). After 2 hr
at 37◦, SYTOX AADvanced Dead Cell Stain was added, and samples were analyzed
by fluorescence-activated cell sorting using a BD LSRII Multi-Laser Analyzer.
Statistical Analysis For proliferation, fluorescence microscopy, and apoptosis ex-
periments, p-values were calculated using a two-tailed unpaired t-test; for relative
protein scavenging rates, a two-tailed paired t-test. 95% confidence intervals were
calculated as the standard error of the mean multiplied by 1.96.
71
Chapter 4
A Genome-Wide Screen Identifies
The Proteins Behind Protein
Eating: GCN2 and cathepsin L
This chapter includes the beginnings of a manuscript. The hypothesis that has
emerged from the data presented – GCN2 upregulates the synthesis of cathepsin
L in amino acid-deprived cells – has not been confirmed.
4.1 Proposed Manuscript Title
GCN2 Upregulates the Translation of Cathepsin L-Encoding mRNAs, Increasing the
Degradative Capacity of Amino Acid-Deprived Cells
4.2 Abstract
Mammalian cells require amino acids to support growth. In nutrient-poor pancreatic
tumors, cancer cells deprived of free amino acids grow by taking up extracellular
protein and catabolizing it in lysosomes. To decipher the molecular mechanisms
72
underlying this growth, we conducted a genome-wide screen for genes essential for
growth using serum protein as the sole source of leucine. The most essential genes
were GCN2, which canonically represses translation initiation in amino acid-deprived
cells, and its putative binding partner GCN1. Using isotope tracers, we found that
GCN2 supports an increase in protein catabolism in amino acid-deficient conditions.
Proteomics in GCN2 wild-type and knockout cells revealed that GCN2 is required
to maintain levels cathepsin L, the most important lysosomal protease for protein
eating according to the screen. Cathepsin L depletion impairs catabolism. Thus,
GCN2 supports catabolism through regulation of translation, enhancing survival and
growth of cells in amino acid-poor conditions.
4.3 Introduction
All cells require amino acids for growth. Classically, mammalian cells take up
monomeric amino acids through transporters in the plasma membrane. Recently,
however, extracellular protein has emerged as an alternative source of amino acids.
Extracellular protein can be taken up via macropinocytosis, a process by which cells
engulf large amounts of extracellular fluid in bulk. The internalized protein can
then be degraded in lysosomes. This activity protein eating promotes the survival
and growth of cells in amino acid-poor environments [16, 48]. In the context of
whole-organism physiology, the role of protein eating remains poorly understood.
To our knowledge, protein eating has only been examined in one setting: pancreatic
ductal adenocarcinoma (PDAC).
PDAC is a devastating disease, with a median overall survival of less than a year
[103]. Pancreatic tumors are nutrient-poor due to substantial fibrosis that limits
the flow of nutrients into the tumor [74]; glutamine is particularly depleted [48].
By contrast, extracellular protein – for example, fibrotic protein secreted by cancer-
73
associated fibroblasts or serum protein leaked by blood vessels – is present in the
tumor in inexhaustible quantities [5]. Increasing evidence suggests that pancreatic
tumor cells rely on this protein as a source of amino acids, including glutamine.
These cells, nearly universally, express a constitutively active form of K-Ras, and K-
Ras signaling stimulates macropinocytosis. This enables cultured pancreatic cancer
cells to use extracellular protein to grow in culture medium lacking essential amino
acids [48]. In slices of excised tissue from human patients, pancreatic tumor cells
actively engaged in macropinocytosis, and in a murine model of PDAC, isotope tracer
studies have provided direct evidence of extracellular protein catabolism in vivo [19].
Pharmacological inhibition of protein eating reduced tumor growth and decreased
amino acid levels in murine tumors [16, 19]. These experiments relied on intratumoral
injection of the macropinocytosis inhibitor EIPA, which is too toxic to administer
systemically [16]. Indeed, while protein eating appears to be an attractive therapeutic
target, no clinically viable inhibitors of this process exist to date.
Little is known about the biochemical mechanisms underlying protein eating. Be-
sides Ras signaling, several other signaling pathways have been implicated in the
regulation of macropinocytosis: phospho-inositol signaling, Rac signaling, and phos-
pholipase C signaling [77]. Sub-membranous pH has also been implicated in regu-
lation of macropinocytosis. (EIPA, the toxic macropinocytosis inhibitor, blocks a
Na+-H+ plasma membrane antiporter, causing proton accumulation that results in
dissociation of Rac1, an inducer of membrane ruffling, from the plasma membrane
[51].)
While K-Ras signaling is indeed sufficient to stimulate macropinocytosis and im-
prove survival and growth on extracellular protein, K-Ras-mutant pancreatic tumor
cells do not initially grow robustly when cultured in leucine-free medium supple-
mented with physiological concentrations of albumin. Most cells switched into this
medium die after several days, but some cells do survive. As we reported previ-
74
ously, starting with KRPC cells, which are derived from an autochthonous murine
model of PDAC, we cultured these survivors continuously in leucine-free, albumin-
supplemented medium, and after several months, the resulting population grows ro-
bustly in this medium, with low levels of cell death [48]. We called these cells adapted
KRPC cells, or KRPCA cells.
Recently, Palm et al. showed that pharmacological inhibition of mTOR complex
1 (mTORC1) kinase activity also enables robust growth on extracellular protein,
proving that long-term adaptation is not required for efficient growth by protein eat-
ing [78]. mTORC1 is a master growth regulator, stimulating protein synthesis and
suppressing protein catabolism upon stimulation by growth factor signaling. Amino
acid levels also modulate mTORC1 activity; in the absence of amino acids, mTORC1
signaling is suppressed. Eventually, however, catabolism of intracellular or extracel-
lular protein in lysosomes reactivates mTORC1 [78, 135]. Thus, mTORC1 signaling
persists in cells cultured in amino acid-deficient medium supplemented with a physio-
logical amount of albumin. For cells starved of amino acids, this persistent mTORC1
signaling is deleterious, as Torin1, a potent inhibitor of mTORC1, promotes growth
in this condition [78].
Palm et al. also proposed a mechanism by which mTOR inhibition enhances
growth fueled by protein eating. They found that Torin1 treatment did not impact
macropinocytosis. Rather, the authors showed that Torin1 induced the degradation of
extracellular protein. Degradation was measured using a fluorescent tool called DQ-
BSA – bovine serum albumin (BSA) conjugated to the fluorophore BODIPY. DQ-
BSA is formulated such that each BSA molecule is associated with several BODIPY
molecules, which quench each other when in close proximity. Only upon degradation
of the BSA do the molecules de-quench (DQ) and become fluorogenic. Thus, the
authors propose that mTORC1 suppresses the utilization of extracellular protein as
nutrients [78].
75
Using a quantitative isotope tracer-based method, we confirmed that Torin1 treat-
ment increases protein eating, but we also observed that amino acid deprivation in-
duces an increase in protein eating, even as mTORC1 activity persists [73]. This
suggests that (i) cells have another way of sensing amino acid shortage and (ii) this
second amino acid sensor increases protein eating independently of mTORC1. The
only two ubiquitously expressed kinases known to be sensitive to amino acid levels are
mTORC1 and GCN2 [24]. We suspect that GCN2 induces the mTORC1-independent
increase in protein observed in amino acid-deprived cells. GCN2 kinase activity is
regulated by uncharged tRNAs, which accumulate in amino acid-deficient conditions.
Upon binding to uncharged tRNA, GCN2 phosphorylates translation initiation factor
eIF2α, thereby inhibiting translation initiation [124, 22, 7].
Some mRNAs are resistant to this GCN2-mediated inhibition of translation initi-
ation. The Atf4 mRNA is one example of an mRNA whose expression is upregulated
by eIF2α phosphorylation. Through a complex mechanism involving short upstream
open reading frames (ORFs) that, in the absence of GCN2 activity, divert ribosomes
from the protein-coding ORF of Atf4, GCN2 activity increases the synthesis of ATF4
[35]. ATF4, a transcription factor, induces expression of amino acid biosynthesis
genes [34]. Other GCN2-resistant mRNAs have been described, but in general the
fact that different mRNAs may have varying sensitivities to eIF2α phosphorylation
has not been explored.
Here, we exploited the capacity of KRPCA cells to grow in leucine-free medium
to conduct a genome-wide screen. Genome-wide CRISPR-based screens have been
employed effectively to systematically identify the genes required for growth in var-
ious cell lines and conditions [123, 122]. We used this technology to identify genes
essential for growth dependent on protein eating. Our screen data revealed that there
are three major molecular processes required for growth on catabolized protein: (i)
uptake of extracellular protein from the environment via macropinocytosis, (ii) degra-
76
dation of endocytosed extracellular protein, and (iii) regulation of protein synthesis.
The two most important genes were Gcn2 and Gcn1. We validated that Gcn2 is
required for growth fueled by protein eating, and we further showed that, through
regulation of translation, GCN2 signaling also promotes growth by increasing the
catabolic capacity of amino acid-deprived cells. This is achieved through regulation
of translation; specifically, GCN2 blocks translation initiation on most mRNAs, di-
recting limited amino acids to other favored mRNAs. Among the GCN2-resistant
mRNAs was cathepsin L, the most important lysosomal hydrolase identified in the
screen. Thus, through phosphorylation of eIF2α, GCN2 increases catabolic capacity,
promoting the survival and growth of amino acid-deprived cells.
4.4 Results - Genome-wide screen systematically
identifies genes required for growth fueled by
catabolized extracellular protein
We have shown previously that the uptake and catabolism of extracellular protein can
fuel the growth of amino acid-starved cancer cells in pancreatic tumors. This process
enables cultured tumor-derived cells to grow in the absence of essential amino acids
if the culture medium is supplemented with a physiological amount of serum protein
(50 g/L BSA) [48]. Cells cultured in this condition are growth-limited mainly by their
ability to eat extracellular protein and use the resulting amino acids efficiently. Thus,
we expected that these cells would be sensitive to loss of any cellular components
essential for growth fueled by protein eating.
To systematically identify these components, we conducted a genome-wide screen
for genes selectively essential for growth in leucine-free medium supplemented with
serum protein – that is, essential for growth in this medium, not essential for growth
77
in amino acid-replete medium. KRPCA cells, which have been adapted for robust
growth fueled by protein eating, were ideal for this screen. We infected these cells
with an expression cassette with the Cas9 endonuclease and a library of single guide
RNAs (sgRNAs). The sgRNA library targets 18855 genes with 184371 unique guide
RNAs. The guide RNA sequences were stably integrated into host cell genomes, so
the relative frequency of each guide could be measured by high throughput sequencing
of these sequences amplified from genomic DNA.
After infection, we split our cells into three populations to be cultured separately.
One population was grown in leucine-free medium supplemented with serum protein
– protein eating is required for growth in this medium. Another population was
grown in amino acid-replete medium supplemented with serum protein, and the third
population was grown in amino acid-replete medium without serum protein supple-
mentation. We cultured each population for 12 doublings, then extracted genomic
DNA. Guide sequences were amplified and sequenced, and the relative frequencies of
all individual guide sequences were compared across populations.
To determine which genes are essential for growth fueled by protein eating but
not for growth in nutrient-rich conditions, we compared the sgRNA frequencies of the
population grown in leucine-free medium and the populations grown in amino acid-
rich media. For each sgRNA, the log-ratio of sgRNA frequencies between populations
was calculated. For each gene, the “selective essentiality” score is the median of the
log-ratios of all sgRNAs targeting that gene (Figure 4.1). (A high selective essentiality
score indicates that a gene was essential in leucine-free medium but not in amino acid-
replete media.) This screen was performed twice, and the selective essentiality scores
shown are the average of the scores from the two independent screen replicates (Figure
4.1). Hereafter, I refer to the genes with the 100 highest selective essentiality scores
as “selectively essential genes.”
78
Figure 4.1: (A) Pooled genome-wide screen design. (B) Selective essentiality forall genes (top) and the top 100 most selectively essential genes (bottom). These top100 genes included several genes involved in three processes: uptake of extracellularprotein, degradation of extracellular protein, and regulation of translation. Selectiveessentiality was calculated for each gene based on the equation in (A). (C) Depictionsof selectively essential genes involved in each process, and scatter plots showing theselectively essentiality scores for these genes. Not all genes plotted are depicted. Theselective essentiality rank of the gene (or the highest ranking gene of the complex) islisted in parenthesis.
79
Figure 4.2: Screen results for Gcn2 and Gcn1. Log2-ratios of sgRNA frequenciesin final to initial populations are shown. Data from each screen replicate is shownseparately. Several sgRNAs targeting both Gcn2 and Gcn1 were strongly depleted inthe final populations grown in leucine-free medium, but not in the final populationsgrown in amino acid-replete medium.
The two genes with the highest selective essentiality scores were Gcn2 (also known
as Eif2ak4) and its putative binding partner Gcn1 (also known as Gcn1l1). GCN2,
as discussed previously, is a kinase that is activated by binding to uncharged tRNA
molecules in amino acid-poor conditions and slows translation initiation [124]. GCN1
is required for GCN2 activation in yeast [67]; the role of GCN1 in mammalian cells
is unknown. In the populations grown in amino acid-rich conditions, the frequencies
of the majority of sgRNAs targeting Gcn2 or Gcn1 did not change. In the popu-
lations grown in leucine-free conditions, however, the frequencies of most sgRNAs
targeting these two genes decreased dramatically (Figure 4.2). This indicates that
cells expressing sgRNAs targeting Gcn2 or Gcn1 were unable to grow in leucine-free
medium.
In general, the selectively essential genes were highly expressed (Figure 4.3). This
suggests that the false positive rate was low – the top 100 most selectively essential
80
Figure 4.3: Selectively essential genes were highly expressed. (A) Using RNA se-quencing data from KRPCA cells (data not shown), log10 reads per kilobase permillion (RPKM) was plotted for all genes. Selectively essential genes were all highlyexpressed. (B) Non-expressed genes (genes with no reads measured in the RNA se-quencing experiment) were not selectively essential.
genes encode proteins that likely play some role in promoting growth fueled by protein
eating. In support of this notion, no non-expressed genes were in these top 100 genes
(Figure 4.3). Moreover, the members of multi-protein complexes generally scored
similarly. For example, the genes encoding all six members of the HOPS complex
(Vps11, Vps16, Vps18, Vps33, Vps39, and Vps41) scored in the top 100, and the
gene encoding the Rab GTPase that regulates HOPS complex activity (Rab7) was
also selectively essential [105]. The fact that all members of this complex scored
suggests that the false negative rate was low – if a protein is required for growth
by protein eating but is not required for growth in nutrient-rich conditions, the gene
encoding that protein was most likely identified as selectively essential.
Not all proteins with essential roles in protein eating-dependent growth were se-
lectively essential, however. Some genes encoding such proteins are essential in amino
acid-replete medium too. For example, K-Ras was essential in all growth conditions;
81
Figure 4.4: K-Ras and the V-ATPase are essential in all growth conditions. For K-Ras, log2-ratios of sgRNA frequencies in final to initial populations are shown. For theV-ATPase, medians of the sgRNA frequency log2-ratios in final to initial populationsfor all subunits are shown. Data from each screen replicate is shown separately.
several vacuolar ATPase components were also essential in all conditions (Figure 4.4).
Other proteins with essential roles in protein eating were not essential in any con-
dition because they have homologs with redundant functions. The subunits of actin
filament capping protein serve as a good example of this. Capping protein is a het-
erodimer composed of CAPZB and either CAPZA1, CAPZA2, or CAPZA3. CAPZB,
which does not have a homolog with redundant function, is selectively essential. Any
CAPZA protein can bind to CAPZB to form a functional complex; because both
CAPZA1 and CAPZA2 are expressed in KRPC cells (data not shown), these pro-
teins are less selectively essential than CAPZB (Figure 4.5). There are undoubtedly
many homologous proteins that perform functions essential to protein eating but are
not selectively essential when knocked out individually and thus escape our attention.
While some complexes include exchangeable proteins like the alpha subunits of
capping protein, other complexes are composed of proteins without co-expressed ho-
mologs. The genes of such complexes had remarkably similar selective essentiality
82
Figure 4.5: Screen results for actin capping protein isoforms. Capzb is selectivelyessential, while the Capza proteins, which have redundant function, are not.
83
scores. For example, the two genes in the TSC1/TSC2 complex ranked 16th and
19th in selectively essentiality and had remarkably similar essentiality profiles. The
same was true for the genes in the GATOR1 complex, the HOPS complex, and the
BORC complex (Figure 4.6). This enables comparison of essentiality between com-
plexes. For example, the HOPS complex appears to be more essential the BORC
complex. In other cases, genes in the same complex, like the AP1 adaptor complex,
had very different Selective Essentiality scores without apparent reason. Ap1m1 was
by far the most selectively essential subunit, despite expression of Ap1m2. Mean-
while, Ap1b1 was not selectively essential, despite being the only known AP1 beta
subunit in the genome.
In general, complexes as a whole seem to be at least as essential as the most
essential gene in the complex; knockout of the most important gene in the complex
more closely resembles the complete absence of the complex than knockout of the
least important gene. (The essentialities of the least important genes in complexes
provide little information about the essentiality of the complex as a whole, so it does
not make sense to average the selective essentiality scores of all genes in a complex.)
Deriving biochemical mechanism from selective essentiality scores is not straight-
forward for many reasons. The degradation of extracellular protein requires not only
the delivery of this protein to lysosomes but also the delivery of lysosomal components
to lysosomes. Thus, a selectively essential gene with an annotated role in vesicle traf-
ficking may be involved in the macropinosome-to-lysosome route or the endoplasmic
reticulum (ER)-to-lysosome route.
84
Figure 4.6: Screen results for various protein complexes. The genes encoding sub-units of a common complex were similarly selectively essential. This enables compar-ison of essentiality between complexes. For example, the HOPS complex appears tobe more essential the BORC complex.
85
4.5 Results - The Three Major Categories of Selec-
tively Essential Genes: Uptake, Degradation,
and Regulation of Translation
Many of the selectively essential genes can be grouped into three major categories:
uptake of extracellular protein via macropinocytosis, trafficking and degradation
of extracellular protein, and negative regulation of protein synthesis (Figure 4.1).
Macropinocytosis is driven by mobilization of the actin cytoskeleton, and accord-
ingly, the selectively essential genes encoding proteins involved in macropinocytosis
included several actin cytoskeleton genes. β-actin was one of the three most selectively
essential genes. There are two ubiquitously expressed actins; the other, γ1-actin, was
also selectively essential [115]. We were somewhat surprised that actin was not es-
sential in cells grown in amino acid-rich conditions. We reasoned that perhaps just
one of these two actin monomer proteins is required in amino acid-rich conditions, in
which macropinocytosis is not critical for growth.
Several actin cytoskeletal proteins beyond the actin subunits themselves were also
identified as selectively essential. These included several of the subunits of the actin-
related protein 2/3 (ARP2/3) complex, which catalyzes the nucleation of new fila-
ments and filament branches [65], and Vasp, which encodes a protein that promotes
actin filament elongation, in part by blocking filament capping [93, 6, 56] (Figure
4.7). (Other ARP2/3 subunits were essential even in amino acid-rich conditions, sup-
porting the notion that a partially functional actin cytoskeleton is required for cell
viability.) As mentioned previously, F-actin capping protein, which caps the barbed
(growing) end of actin filaments [68, 69], blocking them from further growth and
stabilizing them, was selectively essential. How is it possible that both Vasp and
capping protein, which have mutually antagonistic functions (anti-capping and cap-
ping), are both essential? The two activities may be required at different points in
86
Figure 4.7: Selective essentiality of actin-related proteins. Log2-ratios of sgRNAfrequencies in final to initial populations are shown. Data from each screen replicateis shown separately.
the multi-step process of taking up extracellular protein. The precise biochemical
mechanisms underlying this process, however, are complicated and remain poorly un-
derstood. The initial step of actin filament polymerization, which creates membrane
protrusions and eventually membrane ruffles, requires actin filament branching and
elongation [111]. The basis for the requirement of actin filament capping is less clear.
It is possible that capping is required to prevent membrane ruffles from turning into
bigger, more permanent structures like filopodia and invadopodia. It is also possi-
ble that without capping protein, the actin cytoskeletal network of a cell collapses,
making the formation of membrane ruffles impossible.
The idea that the actin cytoskeleton plays a role in macropinocytosis is not new –
cytochalasin D, which inhibits actin polymerization, blocks macropinocytosis [9, 12]
– but the following question remains: How is the actin cytoskeleton mobilized and
controlled to specifically produce membrane ruffles and macropinocytosis? Many
signaling pathways have been implicated: PI3K signaling, Rho GTPase signaling,
87
phospholipase C signaling, and tyrosine kinase signaling, among others [77, 49]. In-
deed, several selectively essential genes are involved in these pathways. Myo9b has
Rho GTPase activity and has been shown to promote invasiveness in lung cancer cells
[52]. Iqgap1, a target of Rho GTPases Cdc42 and Rac1 [36, 57], was also selectively
essential. Additional evidence comes from genes without which cells in leucine-free
medium grew faster. These genes encode proteins deleterious for growth fueled by
extracellular protein. (To identify such genes, we focused on the ratio of sgRNA fre-
quencies between the final leucine-free population and the initial population, rather
than the final amino acid-rich populations. This analysis avoids genes essential in
amino acid-rich conditions but not leucine-free conditions, because these genes do
not inhibit growth in leucine-free medium.) Cells grew fastest without the PTEN
tumor suppressor, a phosphatase that antagonizes PI3K signaling [119]. Cells also
grew faster in leucine-free medium without Ptpn12, a tyrosine phosphatase that likely
dephosphorylates Vasp [110]. The fact that no single signaling protein was much more
essential for protein eating than the rest reflects the high degree of redundancy among
signaling pathways that activate macropinocytosis.
The second major group of selectively essential genes is involved in the degra-
dation of extracellular protein. Among the genes in this group, only one encodes
a protein directly involved in the breakdown of extracellular protein: cathepsin L
(Ctsl). Cathepsin L is a cysteine protease that degrades protein in lysosomes [50].
Many of the other genes are involved in protein trafficking along two major routes.
The first is the route from macropinosomes, where internalized protein first appears
in the cell, to lysosomes, where this internalized protein is degraded. The second is
the route from the endoplasmic reticulum, where lysosomal hydrolases like cathepsin
L are synthesized, to the lysosome, where they degrade extracellular protein.
Much research investigating the delivery of internalized macromolecules to the
lysosome has focused on low-density lipoprotein (LDL). Upon binding to LDL re-
88
ceptor, LDL is taken up into small vesicles called early endosomes. These early
endosomes fuse with one another, growing in size, until they eventually mature into
late endosomes. Late endosomes eventually fuse with lysosomes, at which point their
cargo is degraded [90]. (This classical endocytic pathway seems to be essential for
cell growth fueled by protein eating. LDL receptor is selectively essential, as is Npc1,
which mediates cholesterol efflux from lysosomes. Indeed, cholesterol is required for
macropinocytosis, most likely due its contribution to membrane fluidity [31].)
Compared with LDL uptake, macropinocytosis is poorly understood. The prod-
ucts of macropinocytosis are known as macropinosomes, which are much bigger than
early endosomes. Rather than fusing with one another and increasing in size, these
vesicles appear to decrease in size as they move away from the plasma membrane, to-
ward the center of the cell [112]. The subsequent steps required to deliver the contents
of these macropinosomes to lysosomes are poorly understood. Some of the contents
of macropinosomes are likely sorted into recycling endosomes, which fuse with the
plasma membrane; recycling endosomes dump their cargo back into the extracellular
space and return cell surface receptors back to the plasma membrane. The rest of the
material internalized into macropinosomes, if not recycled, must be degraded. Where
do macropinosomes fit in the early endosome-late endosome-lysosome pathway? The
results of the screen provide some clues.
Membrane trafficking is governed by Rab GTPases. Rab proteins, which are
geranylgeranylated, localize to distinct membrane domains and orchestrate vesicle
docking, membrane fusion, transport along microtubules, and other related activ-
ities [82, 81]. The Rab proteins most essential for growth fueled by protein eat-
ing were Rab35, Rab10, and Rab7. While no single Rab5 homolog was very selec-
tively essential, Rab5c and Rab5a were in the top 10 most selectively essential Rab
genes. (There were 64 total Rab genes that passed through all computational filters.)
Rab5 has been shown to be required for macropinocytosis. Expression of a consti-
89
Figure 4.8: Selective essentiality of Rab proteins and Rabankyrin-5. The average se-lective essentialities, across both screen replicates, of the 10 most selectively essentialRab proteins are shown. For Rabankyrin-5, log2-ratios of sgRNA frequencies in finalto initial populations are shown. Data from each screen replicate is shown separately.
tutively active Rab5 mutant increased pinocytosis of horseradish peroxidase (HRP),
whereas expression of dominant negative Rab5 mutant blocked HRP uptake [59].
Rab5 effector Rabankyrin-5 (also known as Ankfy1) was selectively essential (Figure
4.8). Like Rab5, Rabankyrin-5 overexpression caused an increase macropinocytosis,
and Rabankyrin-5 knockdown caused a reduction [98]. Thus, Rab5 and its effector
Rabankyrin-5 seem to play important roles in uptake of extracellular protein. (One
could make the case that Rab5 belongs in the uptake category.)
The role of Rab5, however, is not restricted to regulation of macropinocytosis.
Simultaneous knockdown of all three Rab5 proteins caused a collapse in the en-
dolysosomal system – a reduction in early endosomes, late endosomes, and lyso-
somes [137]. Indeed, Rab5 localizes to endosomes, not the plasma membrane [90].
Thus, it seems likely that Rab5 both induces macropinocytosis and governs the ac-
tivity of macropinosomes once they are formed, and because Rab5 canonically gov-
90
erns membrane domains of early endosomes, it seems reasonable to imagine every
macropinosome as a special kind of early endosome.
Early endosomes – and assumedly macropinosomes – undergo a process of matu-
ration into late endosomes that involves loss of Rab5 and gain of Rab7 [90]. Based
on the screen data, the most important Rab7 effector is likely the HOPS complex.
The HOPS complex is a tethering complex that orchestrates both late endosome-
late endosome fusion and late endosome-lysosome fusion [101, 84]. Fusion depends
on SNARE proteins [44], although no individual SNARE protein was selectively es-
sential. Thus, through the sequential action of Rab5 and Rab7 and their effector
proteins, extracellular protein is internalized and delivered to lysosomes.
Besides Rab5 and Rab7, two other Rab proteins were selectively essential: Rab35
and Rab10 (Figure 4.8). Rab10 is responsible for recycling of material from early
endosomes back to the plasma membrane [14, 97]. Recycling is required to maintain
homeostasis; without recycling, levels of both plasma membrane lipids and membrane
proteins would decline until macropinocytosis can no longer occur, with the plasma
membrane taut around the cell. The role of Rab35 is less clear. This protein has been
implicated in cytokinesis, recycling from early endosomes, and exosome secretion
[55, 2, 43]. Intriguingly, activating Rab35 mutations have been found in human
tumors. Expression of one of these mutant Rab35 alleles bestowed fibroblasts with
oncogenic potential. Furthermore, in HEK293 cells expressing mutant Rab35, platelet
derived growth factor (PDGF) receptor was trafficked to lysosomes, suggesting that
Rab35 may play an important role in cells fueled by protein eating [125].
Delivery of extracellular protein to the lysosome cannot yield amino acids if lyso-
somal hydrolases like cathepsin L have not also been delivered there. Lysosomal
hydrolases are synthesized into the endoplasmic reticulum (ER), the production site
for all proteins in the endolysosomal and secretory systems [76]. All proteins syn-
thesized into the ER are modified by the addition of a pre-formed oligosaccharide,
91
composed of several N-acetylglucosamine, mannose, and glucose molecules, to an as-
paragine residue of the nascent peptide chain – this is called asparagine (N)-linked
glycosylation. Fully synthesized proteins are next transported from the ER to the
Golgi apparatus. There, lysosomal hydrolases are recognized and modified by the
Golgi resident enzyme GlcNAc-1-phosphotransferase, which adds phosphate groups
to mannose residues of the N-linked oligosaccharide, producing mannose-6-phosphate
(M6P) [18]. (Mechanistically, this phosphorylation is the product of the sequen-
tial addition of GlcNAc-1-phosphate and removal of GlcNAc to a mannose residue.)
GlcNAc-1-phosphotransferase has three subunits: both the α and β subunits are en-
coded by the gene Gnptab, and the γ subunit is encoded by Gnptg. Gnptab, but not
Gnptg, was identified as selectively essential. In humans, mutations in Gnptab cause
mucolipidosis II (or I-cell disease), a fatal lysosome storage disorder [37]. Fibroblasts
from patients with this disease display hypersecretion of lysosomal hydrolases into
the culture medium and accumulation of lysosomes within the cells [75].
Properly M6P-tagged lysosomal hydrolases are recognized by M6P receptors,
which mediate their packaging into clathrin-coated vesicles. The contents of these
vesicles are delivered to early endosomes, the entry point into the endolysosomal sys-
tem. Several genes involved in this process were identified as selectively essential for
growth fueled by protein eating. These included the cation-independent M6P recep-
tor (Igf2r), but not the cation-dependent M6P receptor (M6pr). IGF2R is a receptor
for both insulin-like growth factor and M6P-tagged lysosome hydrolases [71]. Inter-
estingly, Igf1r was also identified as selectively essential. As mentioned previously, the
AP1 adaptor complex, which is required for M6P receptor packaging into clathrin-
coated vesicles destined for early endosomes, is also selectively essential, as is Arf1,
which recruits and activates the AP1 clathrin adaptor complex [89].
Once clathrin-coated vesicles containing lysosomal hydrolases have departed from
the Golgi, they must travel to and fuse with endosomes. This process is not well
92
understood. I propose that Vps45, a poorly characterized SM protein, functions
at this step. SM proteins are required for SNARE-mediated vesicle fusion [109].
(Vps33a, a HOPS complex member, is an SM protein.) In yeast, Vps45 mediates
fusion of Golgi-derived vesicles with the pre-vacuolar compartment [10]. Vps45 is
selectively essential. Since no other vesicle fusion steps are required along the two
major vesicle trafficking routes described, it seems likely that Vps45 is required for
delivery of lysosomal hydrolases to the endolysosomal network.
The BORC complex is a multi-protein complex involved in the peripheral position-
ing of lysosomes. Knockout of any of the subunits of this complex caused lysosomes of
cultured cells to collapse toward the center of the cell. This complex is also required
for cell motility, as determined by scratch assay [86]. The BORC complex has not
been studied in the context of catabolism, but it seems that the peripheral positioning
of lysosomes might be particularly important for efficient catabolism of extracellu-
lar protein. One would think that lysosomes close to the plasma membrane reduce
the amount of trafficking required to deliver extracellular protein to the degradative
compartment.
Uptake and degradation of extracellular protein are sub-processes in the larger
process of supplying cells with amino acids. The third major category of selectively
essential genes (regulation of translation) contains genes encoding proteins that regu-
late amino acid consumption, not amino acid supply. While pancreatic tumor cells in
vivo must regulate the consumption of all scarce amino acids, cultured cells grown in
the absence of leucine must only carefully regulate leucine consumption. As leucine
is a strictly proteinogenic amino acid, regulation of leucine consumption means reg-
ulation of protein synthesis.
Besides the olfactory receptors, which are not expressed widely, there are two
amino acid sensing pathways in mammals: the mTORC1 pathway and the GCN2
pathway [24]. Both mTORC1 and GCN2 are kinases that regulate protein synthesis.
93
mTORC1 is a positive regulator: when activated by insulin signaling, mTORC1
stimulates protein synthesis [58]. Specifically, mTORC1 induces initiation of cap-
dependent translation, which affects only a subset of transcripts [113]. In the absence
of amino acids, mTORC1 signaling is suppressed. However, mTORC1 can be re-
activated by protein catabolism, which yields lysosomal amino acids. Indeed, after
a few hours, leucine-starved cells cultured in a physiological concentration of serum
protein re-activate mTORC1 [135]. Thus, the mTORC1 pathway fails to respond to
prolonged leucine scarcity in cells fed extracellular protein.
GCN2 is a negative regulator of protein synthesis. When activated by uncharged
tRNA binding, GCN2 phosphorylates eukaryotic translation initiation factor 2
(eIF2α). eIF2α is one of three subunits comprising eIF2, which forms a ternary com-
plex with GTP and the initiator codon iMet-tRNA. This ternary complex initiates
translation. eIF2 GTP-binding is promoted by the guanine exchange factor (GEF)
eIF2B; phospho-eIF2α inhibits the activity of eIF2B, thereby inhibiting formation of
the ternary complex [38, 42]. In this manner, GCN2 inhibits translation initiation
upon amino acid depletion. We postulated that Gcn2 is required to protect cells
from the deleterious effects of amino acid starvation, including ribosome stalling and
protein misfolding. This idea has not been confirmed experimentally, however.
Gcn1 was also selectively essential. While we are unaware of any studies inves-
tigating the function of GCN1 in mammals, the yeast homolog of GCN1 has been
shown to be essential for GCN2 signaling. Yeast lacking Gcn1 exhibit a Gcn2-null
phenotype: no eIF2α phosphorylation upon histidine removal. This phenotype can
be rescued by over-expression of histidine tRNA, much of which presumably remains
uncharged when expressed at high levels. Yeast Gcn1 co-immunoprecipitates with
Gcn2, suggesting that Gcn1 facilitates the binding of uncharged tRNA molecules to
Gcn2. Gcn1 also co-sediments with ribosomes. In fact, Gcn1 is homologous to yeast
translation elongation factor 3 (EF3), suggesting that Gcn1 interacts with the A site
94
Figure 4.9: Comparison of the selective essentialities of translation regulators. Theaverage selective essentialities, across both screen replicates, are shown.
of the ribosome where amino acylated (or uncharged) tRNAs enter [67, 118, 28]. (EF3
is not conserved in mammals.) Taken together, this evidence led Alan Hinnebusch,
who discovered Gcn2, to propose that Gcn1 may catalyze the transfer of uncharged
tRNA from the A-site of ribosomes to the uncharged tRNA-binding site of Gcn2,
thereby activating Gcn2 kinase activity [42].
While Gcn2 and Gcn1 were the most essential genes for growth fueled by protein
eating, negative regulators of mTORC1 were less selectively essential (4.9). The idea
that these negative regulators of mTORC1 (TSC and GATOR1) are less important
to amino acid-starved cells than GCN2 and GCN1 is consistent with the idea that the
negative regulators of mTORC1 fail to suppress mTORC1 activity to the extent that
optimizes growth on extracellular protein [73]. The fact that pharmacological inhibi-
tion of mTORC1 promotes robust growth of albumin-fed cells in amino acid-deficient
conditions supports this notion [78]. Thus, in cells dependent on catabolism of extra-
cellular protein, genetic loss of any negative regulator of mTORC1 does not cause,
but rather exacerbates mTORC1 hyperactivity in amino acid-deficient conditions.
95
Beyond GCN2 and GCN1 and the negative regulators of mTORC1, there was
one particularly interesting gene involved in translation initiation that was found to
be selectively essential: Dhx29. DHX29 is an RNA helicase which is required for
in vitro translation of mRNAs with structured 5’ UTRs untranslated regions [83].
The precise role of DHX29 in cells, however, remains poorly understood. Parsyan et
al. showed that Dhx29 is broadly required for translation initiation in HeLa cells,
and Dhx29 knockdown impedes HeLa cell growth in culture and in xenografts. The
authors conclude that DHX29 is a “bona fide translation initiation factor,” similar
to eIF4E, and suggest that DHX29 affects “the initiation of translation of mRNAs
with moderately to extensively structured 5’ UTRs,” including “those involved in
controlling cell proliferation and apoptosis” [79]. To our knowledge, the possibility
that DHX29 is selectively essential in specific conditions has never been proposed.
4.6 Results - Screen Validation and Proteomics
To validate the basic findings of the screen, we generated knockout cell lines lacking
selectively essential genes in each of the three categories described above: Gcn2 (regu-
lation of translation), Vasp (uptake of extracellular protein), and Vps39 (degradation
of extracellular protein). Our knockout cell lines each originated from a single cell;
single cells (and the clonal populations that emerge from them) can vary in phenotype
in ways that do not depend on the gene that has been knocked out. To control for
these variations, we expressed either EGFP or the human homolog of the knocked
out gene. We confirmed knockout and re-expression of Gcn2 and Vasp by Western
Blot (Figure 4.10A). (We could not find a reliable Vps39 antibody, so we verified
knockouts by Sanger sequencing (data not shown).)
We measured the growth of these cells in amino acid-replete medium or leucine-
free medium, both supplemented with 50 g/L albumin. In amino acid-replete medium
96
Figure 4.10: Basic validation of selectively essential genes. (A) Western blots con-firming knockout and re-expression of Gcn2 and Vasp. We could not find a reliableVps39 antibody, so we verified knockouts by Sanger sequencing (data not shown). (B)Growth of knockout and re-expression cell lines in amino acid-rich and leucine-freemedia. (C) Images of knockout and re-expression cell lines cultured in leucine-freemedium for 48 hours.
97
all three knockout cell lines grew as fast as the re-expression control cell lines, but in
leucine-free medium the knockout cell lines all grew significantly more slowly, with
noticeable cell death (Figure 4.10B-C).
As Gcn2 was the most selectively essential gene, we next asked whether GCN2
impacted the rate of catabolism of extracellular protein. While GCN2-wild-type cells
increased their catabolic rates upon glutamine removal, the catabolic rates of GCN2
knockout cells remained the same (4.11A). The isotope tracer data also revealed
that GCN2 activity does not meaningfully alter the rate of amino acid incorporation
into protein (data not shown). These results suggest a role for GCN2 in upregu-
lating catabolism of extracellular protein in amino acid-deficient conditions. How
does GCN2, with only one known phospho-target (eIF2α), induce increased catabolic
rates? Phosphorylation of eIF2α slows translation initiation – could this somehow
result in increased catabolism? If not, perhaps GCN2 phosphorylates a yet-unknown
lysosome-related protein?
For the following reasons, I hypothesized that GCN2 induces increased catabolic
rates through phosphorylation of eIF2α, not through phosphorylation of any other
protein. (GCN2 is not known to phosphorylate any other protein.) First, the increase
in catabolic rates is gradual, not abrupt. If GCN2 were causing this increase by
phosphorylating a vesicle trafficking component that induced trafficking of endosomal
cargo to the lysosome, for example, the increase in catabolism ought to be relatively
immediate – on the timescale of minutes or hours. Instead, the GCN2-dependent
increase in catabolism occurs gradually over several days (4.11B). Second, inhibition
of mTORC1, like GCN2 activation, induces a similar gradual increase in catabolism.
Might these increases in catabolism – one induced by mTOR inhibition, the other by
GCN2 activation – result from a common mechanism? Both mTOR inhibition and
GCN2 activation slow translation initiation. Perhaps mTORC1 and GCN2 regulate
translation initiation to varying extents depending on the mRNA. If so, mTORC1
98
Figure 4.11: GCN2 supports protein catabolism in amino acid-deprived cells. (A)Scavenging flux measurements of GCN2 wild-type and GCN2 knockout cells in aminoacid-rich and glutamine-free media. These measurements were based on 24 hours ofculture in these conditions. (B) Scavenging flux measurements in GCN2 wild-typecells over 72 hours. Scavenging fluxes measured as described previously [73].
inhibition and GCN2 activation might induce catabolism by effectively upregulating
the translation of proteins involved in protein eating.
GCN2 is activated in cells that are starved for amino acids. From an evolutionary
perspective, many of the known downstream consequences of GCN2 activation make
sense as responses to amino acid starvation. GCN2 slows translation initiation; this
prevents ribosome stalling and cell death. GCN2 induces translation of ATF4; ATF4
is a transcription factor that activates expression of amino acid stress response genes
[34]. ATF4 is produced as a result of eIF2α phosphorylation: ATF4 synthesis is
achieved only when the ribosome reads through three decoy start sites in the upstream
untranslated region, then initiates translation at the correct start site [35]. Based on
existing datasets in the literature, it seems unlikely that there are other genes like
ATF4 whose translation is so tightly regulated (completely suppressed unless eIF2α
99
is phosphorylated), but it does seem possible that other genes which are already
translated continue to be translated at relatively high levels. These GCN2-resistant
mRNAs might encode proteins somehow involved in protein catabolism.
The fact that DHX29 is selectively essential supports the notion that translation
of a certain subset of mRNAs may be critical. DHX29 is required for translation
initiation on mRNAs with structured 5’ UTRs [83]. Not all mRNAs have structured
5’ UTRs. Thus, DHX29 facilitates the synthesis of some (unknown) proteins. If
DHX29 is selectively essential because it is required for synthesis of these proteins,
one would expect that some of these proteins are required to support the growth of
cells fueled by protein eating.
To comprehensively assess the effect of GCN2 on cellular protein levels, we mea-
sured the transcriptomes and proteomes of GCN2 wild-type and knockout cells cul-
tured (i) in amino acid-rich medium for 24 hours, (ii)-(iii) in leucine-free medium for
24 hours and 48 hours, and (iv)-(v) in glutamine-free medium for 24 hours and 48
hours. All media were supplemented with 5% dialyzed fetal bovine serum and 5%
bovine serum albumin. Here, I focus on the proteomics data. These data were gen-
erated using isobaric (TMT) tags that enable simultaneous measurement of peptide
abundances across samples. The method (TMT-MS3) quantified the relative levels
of 5738 proteins in Gcn2 WT or KO cells cultured in the five conditions listed above.
Unsupervised clustering of the normalized data revealed that glutamine deprivation
was the factor that altered global protein levels the most, relative to cells grown
in amino acid-rich medium (Figure 4.12A). Interestingly, protein levels in leucine-
deprived GCN2 WT cells were more similar to cells grown in amino acid-rich medium
than leucine-deprived GCN2 KO cells. This suggests that GCN2 is required to main-
tain proteome homeostasis, not to affect proteome remodeling. At present, I do
not understand the details of this process – how exactly GCN2 “maintains proteome
homeostasis.” One general possibility is that GCN2 upregulates the synthesis of short-
100
Figure 4.12: The effect of GCN2 and amino acid depletion on protein levels. UsingTMT-MS3-based proteomics, we quantified the levels of 5738 proteins in GCN2 wild-type and knockout cells cultured in amino acid-rich medium for 24 hours, in leucine-free medium for 24 hours and 48 hours, and in glutamine-free medium for 24 hoursand 48 hours. A heatmap showing the levels of all proteins is on the left. A heatmapshowing the levels of all selectively essential proteins measured is on the right.
lived proteins at the expense of stable proteins, which need not be synthesized unless
cell growth is possible.
We next searched for specific proteins that were upregulated upon amino acid-
deprivation specifically in GCN2-expressing cells, hoping to identify proteins which
might explain the GCN2-dependent increase in catabolism in amino acid-deprived
cells. We reasoned that these unidentified GCN2-regulated proteins that promote
101
catabolism should be essential for growth fueled by protein eating. (These proteins
could also be essential for growth in amino acid-rich medium – in other words, they
might not be selectively essential for growth on extracellular protein, but rather just
essential. For example, GCN2 could conceivably induce synthesis of the components
of the V-ATPase to increase catabolism. However, we did not observe any substantial
changes in the levels of V-ATPase components (data not shown).) Thus, we restricted
our initial search for important GCN2-regulated proteins to the genes that scored as
selectively essential in the screen.
Clustering the proteomics data for just the selectively essential genes – of the top
100, 78 were quantified in the proteomics experiment – did not reveal any outstand-
ing proteins upregulated in amino acid-deprived cells in a strictly GCN2-dependent
fashion (Figure 4.12B). To systematically assess which of these genes encode proteins
whose levels are upregulated by GCN2, we first compared protein levels in GCN2 WT
and KO cells in amino acid-rich medium (DMEM). (While GCN2 is thought to be in-
active in amino acid-rich conditions, DMEM lacks several non-essential amino acids,
and inevitably, amino acid pools are periodically depleted.) We found that SLC7A5,
a high-affinity leucine transporter and known GCN2-ATF4 target [34], was the most
GCN2-upregulated protein in amino acid-replete medium. Among the other selec-
tively essential proteins most upregulated by GCN2 in amino acid-replete conditions
were the lysosomal hydrolase Cathepsin L and the mannose-6-phosphate receptor
IGF2R, which delivers Cathepsin L to the lysosome (Figure 4.13A). We next exam-
ined the changes in protein levels upon amino acid deprivation. To do this, we first
calculated the difference in protein levels in amino acid-deficient medium – after 24
hours in glutamine-free medium, for example – relative to the levels in amino acid-
rich medium. This difference was calculated independently for Gcn2 WT and Gcn2
KO cells. We then subtracted this difference in Gcn2 KO cells from the difference
in Gcn2 WT cells, thereby isolating the GCN2-dependent induction in protein level
102
changes. We found that GCN2 induced higher Cathepsin L and IGF2R levels upon
either leucine or glutamine deprivation. This was true after both 24 hours and 48
hours of amino acid deprivation (Figure 4.13B-C).
To summarize the effect of GCN2 in amino acid-deprived cells in a single statis-
tic, we compared the average protein levels in all amino acid-deprived conditions –
leucine-deprived for 24 hours and 48 hours, glutamine-deprived for 24 hours and 48
hours – separately in GCN2 WT and GCN2 KO cells. Over all amino acid-deprived
samples, GCN2 upregulated cathepsin L and IGF2R more than any other selectively
essential genes; on average, in amino acid-deficient conditions these proteins were
twice as abundant in cells expressing GCN2 (Figure 4.13D). Other proteins upregu-
lated by GCN2 included the leucine transporter SLC7A5, the lysosomal cholesterol
transporter NPC1, and the actin bundling protein plastin 3. These results suggest a
simple explanation for why GCN2 expressing cells exhibit higher catabolic rates in
amino acid-deficient conditions: GCN2 promotes the synthesis of cathepsin L, which
catalyzes a rate-limiting degradative step in the catabolism of extracellular protein,
and cathepsin L receptor IGF2R, which delivers cathepsin L to the lysosome.
Cathepsin L levels did not increase, however, in cells deprived of amino acids.
Rather, in both Gcn2 WT and Gcn2 KO cells, cathepsin L levels decreased dramati-
cally upon glutamine removal. After 24 hours, cathepsin L levels dropped significantly
– 68% in Gcn2 WT, 75% in Gcn2 KO. This suggests that cathepsin L is a short-lived
protein. There are two obvious reasons cathepsin L might be especially short-lived:
cathepsin L might itself be degraded in the lysosome, or it might be slowly lost due
to secretion into the extracellular space. In any case, if cathepsin L is required for
protein eating, the cell must continuously synthesize it. After 48 hours of glutamine
deprivation, cathepsin L levels in Gcn2 WT cells recover to the levels that were mea-
sured in these cells in rich medium. In Gcn2 KO cells, cathepsin L levels rebound
partially but remain low. Upon leucine deprivation, Gcn2 WT cells maintain cathep-
103
Figure 4.13: Cathepsin L levels are increased in amino acid-deprived cells expressingGCN2. (A-D) Relative protein abundances in GCN2 wild-type and GCN2 knockoutcells cultured in amino acid-replete (A) or amino acid-deficient (D) media. The datafrom all amino acid-deficient samples was averaged for each cell line. (B-C) GCN2-dependent induction in protein levels after leucine or glutamine removal after 24 hours(B) or 48 hours (C). (E) Cathepsin L and IGF2R levels in amino acid-deprived cells.
104
sin L levels after one day and increase cathepsin L levels after two days, whereas
in Gcn2 KO cells cathepsin L levels decline (Figure 4.13E). Meanwhile, cathepsin L
mRNA levels did not change significantly (data not shown). Thus, GCN2 seems to
post-transcriptionally regulate cathepsin L levels.
We have repeatedly observed increases in lysosome volume – as measured by Lyso-
Tracker, a pH-sensitive dye – in cells deprived of amino acids, both Gcn2 WT and
Gcn2 KO (data not shown). Amino acid-deprived cells likely achieve this increase in
lysosome volume by sending various vesicles – possibly endosomes – to fuse with lyso-
somes. In doing so, cells target the material within these endosomes for degradation.
These increases in lysosome volume are not paralleled by increases in the concen-
trations of lysosomal proteins; thus, trafficking of endosomal cargo to the lysosome
does not depend on synthesis of new proteins. However, if key degradative enzymes
like cathepsin L are short-lived, they must be replaced. If not, the cell will lose the
capacity to degrade protein and will become irreparably amino acid-starved.
Interestingly, Ctsl-/- MEFs also exhibit increased lysosome volume, even in amino
acid-replete medium [21]. This suggests that increased lysosome volume can be caused
not only by demand for amino acids in starved cells but also by the failure to de-
grade what was trafficked to the degradative compartment. In light of this idea,
the increased lysosome volumes of glutamine-deprived cells may reflect a combina-
tion of increased trafficking to lysosomes and limited degradative capacity, given that
cathepsin L levels drop upon glutamine removal.
To test this theory, I plan to over-express cathepsin L in Gcn2 WT and Gcn2
KO cells. Naively, I predict that cathepsin L overexpression will increase catabolic
rates, regardless of Gcn2 expression, in amino acid-rich conditions. (This may not
necessarily turn out to be true. For example, if cathepsin L overexpression induces the
cathepsin L-mediated degradation of other important lysosomal enzymes, catabolic
rates may decrease.) More importantly, with cathepsin L levels constitutively high,
105
I predict that Gcn2 WT and Gcn2 KO cells will catabolize extracellular protein at
equal rates upon amino acid removal. Cathepsin L over-expression may also impact
lysosome volume.
106
Chapter 5
Ribosomes on the Night Shift: The
universal protein-making machine
becomes a nutrient source between
meals
(This perspective was published in Science in 2018. The authors are Michel Nofal
and Joshua D. Rabinowitz [72].) From an evolutionary perspective, life involves two
simple goals: survival and reproduction. But these goals are fundamentally at odds.
Reproduction depends on growth, but attempts to grow when nutrients are scarce can
jeopardize survival. In cells, growth is accomplished in large part by ribosomes, huge
RNA-protein machines that translate nucleic acid messages into protein, the main
biochemical constituent of cells. In nutrient-rich conditions, cells can be filled with
ribosomes: they comprise over a third of total biomass in rapidly growing Escherichia
coli [100]. But what happens to ribosomes when nutrient levels decline, as occurs
sporadically in microbes and nightly in sleeping humans? Biosynthesis subsides, and
ribosomes now serve as a reservoir of nutrients. Building on recent progress probing
107
the regulation of protein synthesis and degradation from Gu et al. [32] and Abu-
Remaileh et al. [1], on page 751 of this issue, Wyant et al. [129] elucidate a pathway
in which ribosomes are selectively digested, promoting survival in starved cells.
Eukaryotic cells, from yeast to humans, can gobble up parts of their interior
through a process called autophagy, encapsulating them in double membranes and
forming an enclosed compartment known as an autophagosome. Autophagosomes
then deliver their contents to the degradative compartment (lysosomes) where macro-
molecules are recycled into monomeric nutrients [114].
Autophagy was initially thought to be a non-specific process, but it has become
increasingly clear that cells can pick and choose what to digest in this manner. For
example, defective mitochondria are detected and marked for autophagy through a
system involving the proteins PINK1 (PTEN-induced putative kinase protein 1) and
Parkin [29]. In nutrient-poor conditions, however, mitochondria are valuable – they
provide the most efficient way to generate energy from carbon – whereas ribosomes are
no longer needed in large numbers to fuel biosynthesis. These dispensable ribosomes
can be selectively degraded by ribophagy – autophagy of ribosomes.
How do cells balance growth and survival? The mTOR complex 1 (mTORC1)
kinase has emerged as an important regulator of this balance. When conditions are
favorable for growth, mTORC1 stimulates the synthesis of all major biomaterials in
cells, especially ribosomes, while suppressing autophagy. If growth conditions are poor
– for example, during periods of starvation – mTORC1 is inactive, and autophagy
proceeds.
To decide whether or not growth is appropriate, mTORC1 must sense and inte-
grate a diverse set of environmental cues. One of these cues is amino acid availability.
Cells must be well-stocked with amino acids, which are needed to make protein, in
order to grow. Amino acids within the cytosol promote translocation of mTORC1
to the surface of the lysosome, where its activator, a small guanosine triphosphatase
108
(GTPase) called RHEB (Ras homolog enriched in brain), resides [95, 11]. Numer-
ous specific amino acid sensing proteins have been characterized. These include
the Sestrin and CASTOR (cytosolic arginine sensor for mTORC1) families of pro-
teins, which sense leucine and arginine, respectively [126, 13]. Gu et al. added to
this list when they identified SAMTOR (S-adenosylmethionine sensor upstream of
mTORC1), which indirectly senses the essential amino acid methionine by binding
to S-adenosylmethionine.
Why does the lysosome, which degrades macromolecules, play such a central role
in the regulation of mTORC1, which promotes the construction of macromolecules?
Accumulating evidence suggests that mTORC1 preferentially senses nutrients that
are generated in the lysosome. Perhaps by sensing the products of degradation,
mTORC1 can assess whether catabolic processes have generated enough nutrients.
This reasoning assumes that lysosomes differ from the rest of the cell not only in
acidity and protein content, but also in metabolite content. However, most methods
for lysosomal purification involve ultracentrifugation in sucrose gradients for several
hours, during which time metabolites have likely reacted or escaped, and weakly
associated lysosomal proteins have disassociated. Abu-Remaillah et al. reported a
method for rapid isolation of lysosomes called LysoIP. Cells are genetically engineered
to express a protein tag on lysosomal membranes. Magnetic beads linked to antibodies
specific for the tag are added to lysed cells, and lysosomes are purified magnetically.
This method has enabled systematic analysis of the metabolite content of lysosomes
for the first time.
Using LysoIP, the authors compared the cytosolic and lysosomal concentrations of
numerous metabolites. In cells grown in amino acid-rich conditions, metabolite levels
were generally similar in both compartments, but upon impairment of the vacuolar
ATPase, which acidifies the lysosome, the lumenal concentrations of a large number
of metabolites increased. These metabolites included most nonessential amino acids,
109
which are apparently released from lysosomes in a proton-dependent manner. The
lysosomal levels of most essential amino acids, however, did not change, suggesting
their regulation by another factor.
A series of elegant experiments revealed that mTORC1 promotes the efflux from
lysosomes of most essential amino acids, including leucine. Efflux is mediated by
a lysosomal membrane protein called SLC38A9 (sodium-coupled neutral amino acid
transporter 9). In amino acid-poor conditions, leucine transport out of lysosomes is
required to activate mTORC1 [130]. Taken together, these data suggest a paradox:
mTORC1 activity induces SLC38A9-mediated efflux of leucine out of lysosomes, but
mTORC1 remains inactive until leucine leaves the lysosome.
SLC38A9 has another important function, which may explain this paradox: upon
binding to the non-essential amino acid arginine in the lysosomal lumen, SLC38A9
helps to activate mTORC1. Thus, when amino acid levels accumulate in lysosomes,
mTORC1 may initially become partially activated by SLC38A9. In this scenario,
mTORC1 subsequently induces SLC38A9-mediated amino acid efflux from lysosomes.
Finally, effluxed leucine can augment mTORC1 activation. In other words, SLC38A9
may be at the center of a feed-forward loop whereby nutrients derived from lysosomal
catabolism activate mTORC1 only after accumulating above a threshold.
Why might SLC38A9 sense lysosomal arginine levels? Perhaps mTORC1 evolved
to sense the degradation of proteins rich in arginine. One class of proteins stands
out as arginine-rich: ribosomal proteins. These proteins contain high frequencies of
arginine and lysine, the positive charges of which help to bind the negatively charged
phosphate backbone of ribosomal RNA.
How are ribosomes selectively delivered to lysosomes? Wyant et al. applied quan-
titative proteomics to identify proteins that increase their association with lysosomes
in nutrient-starved cells. NUFIP1 (nuclear fragile X mental retardation-interacting
protein 1), which has a previously annotated role in the nucleus, was found at higher
110
concentrations on lysosomes in cells deprived of glucose and amino acids. The authors
then showed that NUFIP1 binds to ribosomes when mTORC1 is inactive and enables
ribophagy by delivering these ribosomes to autophagosomes for degradation.
Ribosomes are arguably the most important biochemical machine. But the im-
portance of translation has overshadowed the role ribosomes can play as a nutrient
source. By elucidating the function of NUFIP1, Wyant et al. provide a genetic handle
to specifically probe the importance of ribosomes as nutrients. Indeed, genetic loss of
NUFIP1 (that is, the inability to use ribosomes as nutrients) impairs survival of cells
starved of glucose and amino acids.
Although it is easy to induce starvation of mammalian cells experimentally in a
culture dish, cells in vivo are never exposed to glucose-free, amino acid-free environ-
ments. Rather, they are bathed in a steady stream of circulating nutrients. What
prevents this stream from running dry? During extended periods between meals,
macromolecules must be degraded. Proteins are depots of amino acids; glycogen is
a depot of sugar. Ribosomes, uniquely, are depots of amino acids, sugar, and nu-
cleobases, and as such, they can support diverse metabolic activity. A recent report
showed that, in mice, liver size and ribosome content oscillate with the diurnal cycle,
increasing while the animals are awake (and eating at will), then gradually falling
during sleep [104]. Thus, after meals, the liver fills with ribosomes. For a time, these
ribosomes perform their canonical role: using ingested amino acids to make protein.
But as nutrient levels drop, these ribosomes, via ribophagy, are recycled into nutri-
ents for the rest of the body. These findings have not been validated in humans,
but they raise intriguing possibilities. While intact ribosomes are essential for diverse
anabolic functions – protein synthesis is required for long-term memory formation
[20] – degraded ribosomes may maintain nutrient levels as we sleep. Perhaps they
quite literally fuel our dreams. Food for thought.
111
Figure 5.1: (See caption above.)
112
Chapter 6
Conclusion and Future Directions
When I began as a graduate student in Princeton, the idea that any mammalian
epithelial cells might use extracellular protein as an amino acid source was foreign to
the cancer metabolism community. Times have changed. It is now clear that pancre-
atic cancer cells, as well as other cells with active Ras signaling, can use extracellular
protein as an amino acid source. Protein eating does not merely supplement amino
acid pools. I proved that this process can fuel the growth of cultured cells deprived
of essential amino acids. In some cell lines, protein eating can support growth in
medium lacking all amino acids.
Protein eating can be measured in several ways. Imaging methods provide qual-
itative information. For example, DQ-BSA imaging was used to confirm that extra-
cellular protein was degraded in lysosomes [16]. In this thesis, I presented a method
that employs isotope tracers and liquid chromatography-mass spectrometry to mea-
sure protein eating in quantitative terms: extracellular protein-derived amino acids
released per µL cell volume per hour [73]. This method has yet to be used by others,
but I believe it is the only method that provides reliable measurements of protein eat-
ing rate. Using this method, I found that the effect of mTORC1 on protein eating in
amino acid-deprived cells is minimal; thus mTOR inhibition enhances cell growth fu-
113
eled by protein eating by another mechanism. I also showed that amino acid-deprived
cells increase their rate of protein eating in mTOR-independent fashion.
Finally, I presented the results of a genome-wide screen that systematically iden-
tified genes selectively required for growth fueled by protein eating. This screen
revealed the importance of GCN2 signaling in amino acid-deprived cells. GCN2 not
only slows translation initiation to prevent cell death but also increases catabolic
rates. GCN2 may support catabolism by promoting the synthesis of the lysosomal
hydrolase cathepsin L. To demonstrate the importance of cathepsin L in maintain-
ing high catabolic rates, I plan to over-express cathepsin L in GCN2 wild-type and
GCN2 knockout cells. I predict that cathepsin L over-expression will rescue – at least
partially – the protein eating defect of GCN2 knockout cells in amino acid-deficient
conditions.
Upregulation of cathepsin L, however, cannot be the only mechanism by which
amino acid-deprived cells upregulate catabolism. Imaging experiments have shown
that GCN2 wild-type and GCN2 knockout cells alike exhibit an increase in lysosome
volume upon glutamine removal. This suggests that, independently of GCN2, cells
respond to amino acid starvation by mediating fusion of endosomes with lysosomes,
thereby committing additional protein for degradation. Meanwhile, cathepsin L levels
decline – a little bit in GCN2 wild-type cells and a lot in GCN2 knockout cells.
Taken together, these data suggest that cells upregulate catabolism by a translation-
independent mechanism that involves modulation of vesicle trafficking. I propose
that these cells cannot maintain high levels of catabolism solely by trafficking protein
to lysosomes because cathepsin L is short-lived. GCN2 upregulates the synthesis of
cathepsin L and, in doing so, supports increased catabolism in these cells. Without
GCN2, cells direct the contents of endosomes to the degradative compartment, but
the degradative compartment lacks a key enzyme.
114
To prove that GCN2 upregulates the synthesis of cathepsin L, I plan to do ribo-
some profiling of GCN2 wild-type and GCN2 knockout cells in amino acid-rich and
various amino-acid deficient conditions. I predict that there will be an increase in
ribosomes bound to cathepsin L in amino acid-deprived GCN2 wild-type cells and not
in amino acid-deprived GCN2 knockout cells. It will also be interesting to investigate
other GCN2-dependent changes in translational efficiency upon amino acid removal.
Many other questions remain unanswered. How is macropinocytosis regulated?
Why is GCN1 required for GCN2 signaling? Do any cells without Ras mutations eat
protein in vivo? Which mRNAs require DHX29 for translation? Is IGF1R another
mannose-6-phosphate receptor? How does GCN2 suppress the synthesis of some
mRNAs more than others? Why do amino acids liberated from intact protein in
lysosomes have a higher probability of being used for protein synthesis than amino
acids that entered the cell as monomers (data not shown)? What does Rab35 do?
Can the isotope tracer methods I developed to quantify catabolic fluxes be repurposed
to measure protein turnover when coupled with proteomics? Can inhibiting GCN2
be an effective therapy in pancreatic cancer?
How do amino acid-deprived cells upregulate trafficking to the lysosome, as dis-
cussed above? This process occurs in GCN2 knockout cells; thus, GCN2 is not re-
sponsible. In addition, mTORC1 signaling remains active in glutamine-deprived cells;
thus, this process is mTORC1-independent. While mTORC1 may not be the answer
to this question, the relationship between mTORC1 and lysosomes is interesting and
incompletely understood. mTORC1 is required for the reformation of small lysosomes
from large ones once the cargo has been degraded [135]. mTORC1 also interacts with
one of the subunits of the BORC1 complex in ways that are poorly understood.
BORCS6 physically interacts with the Ragulator complex, the lysosomal scaffold to
which many mTORC1-related proteins bind, and disrupts the Rag-Ragulator interac-
tion to inhibit mTORC1 [99]. The consequences of these protein-protein interactions
115
are unclear, but it is tempting to speculate that mTORC1 regulates the reformation
and positioning of lysosomes by modulating the activity of the BORC complex.
In general, the movement of vesicles, including lysosomes, is poorly understood.
Most vesicles transport proteins; lysosomes degrade them, as well as other macro-
molecules. Both transport and degradation cannot be easily measured by measuring
quantities (number of vesicles, DQ-BSA fluorescence). These processes are best de-
scribed in terms of flux. As such, perhaps tools developed to study metabolism can
be applied to the field of vesicle trafficking.
Protein eating – as opposed to any other vesicle trafficking-related activity – en-
ables unprecedented exploration of vesicle trafficking, a fundamental element of cell bi-
ology, in three distinct ways. First, because pancreatic cancer cells cultured in amino
acid-deficient medium rely on this process for survival and growth, proliferation-
based genetic screens can be used to systematically identify the key protein catalysts
of this pathway. Second, using isotope tracers, the rate of protein eating can be
measured quantitatively in terms of amino acid release from the lysosome. Com-
parison of protein-eating rates in wild-type and knockout cell lines revealed that no
single gene is absolutely required for degradation of extracellular protein, with the
possible exception of essential genes such as V-ATPase components. Rather, loss of
selectively essential genes results only in a partial decrease in pathway flux (roughly
50%). Third, because extracellular protein is taken up from the extracellular space
in a receptor-independent manner, endocytic vesicles can easily be labeled with flu-
orogenic substrates. No other vesicle trafficking route can be linked to a growth
phenotype, facilitating genetic screens; quantified in terms of overall pathway flux;
and tracked using fluorescent tracers added to the culture medium.
Protein eating can be considered a metabolic pathway. The intermediates of this
pathway are not metabolites with unique chemical structures but vesicles with unique
sub-cellular localization, membrane composition (both lipids and protein), and cargo.
116
Unlike classical metabolic pathways like glycolysis, the steps of this pathway are not
catalyzed by individual enzymes but by groups of proteins with diverse functions.
Many of these proteins – for example, those involved in transport along microtubules
or vesicle fusion – have been characterized in detail in vitro, but this may not be
enough; decades of in vitro experiments were insufficient to determine the control
points of glycolysis in cells. Rather, isotope tracer studies probing the totality of the
glycolytic pathway in the context of the cellular environment revealed these control
points. Analogous systems-level analyses will likely be required to fully understand
the cellular functions of vesicular trafficking proteins.
The protein-eating pathway is amenable to such systems-level analyses for the
reasons listed above. We can now quantify the overall rate of protein eating, but
little is known about regulates this rate. What are the most rate-limiting steps? To
what extent are the steps reversible? And what are the roles of individual proteins in
the context of the whole pathway? The goal is to develop fluorescence microscopy-
based methodology to enable the quantification of vesicle trafficking fluxes. Such
methodology would enable us to move from a world of crude understanding of isolated
proteins to a systems-level understanding of cellular processes.
117
Bibliography
[1] Monther Abu-Remaileh, Gregory A Wyant, Choah Kim, Nouf N Laqtom, MariaAbbasi, Sze Ham Chan, Elizaveta Freinkman, and David M Sabatini. Lysosomalmetabolomics reveals v-atpase-and mtor-dependent regulation of amino acidefflux from lysosomes. Science, 358(6364):807–813, 2017.
[2] Patrick D Allaire, Andrea L Marat, Claudia Dall’Armi, Gilbert Di Paolo, Pe-ter S McPherson, and Brigitte Ritter. The connecdenn denn domain: a geffor rab35 mediating cargo-specific exit from early endosomes. Molecular cell,37(3):370–382, 2010.
[3] Sunil U Bajad, Wenyun Lu, Elizabeth H Kimball, Jie Yuan, Celeste Peter-son, and Joshua D Rabinowitz. Separation and quantitation of water solublecellular metabolites by hydrophilic interaction chromatography-tandem massspectrometry. Journal of chromatography A, 1125(1):76–88, 2006.
[4] Dafna Bar-Sagi and James R Feramisco. Induction of membrane rufflingand fluid-phase pinocytosis in quiescent fibroblasts by ras proteins. Science,233(4768):1061–1068, 1986.
[5] Franz K Bauer, Manuel Tubis, and H Bruce Thomas. Accumulation of homolo-gous radioiodinated albumin in experimental tumors. Proceedings of the Societyfor Experimental Biology and Medicine, 90(1):140–142, 1955.
[6] James E Bear, Tatyana M Svitkina, Matthias Krause, Dorothy A Schafer,Joseph J Loureiro, Geraldine A Strasser, Ivan V Maly, Oleg Y Chaga, John ACooper, Gary G Borisy, et al. Antagonism between ena/vasp proteins and actinfilament capping regulates fibroblast motility. Cell, 109(4):509–521, 2002.
[7] Juan J Berlanga, Javier Santoyo, and Cesar de Haro. Characterization of amammalian homolog of the gcn2 eukaryotic initiation factor 2α kinase. Euro-pean Journal of Biochemistry, 265(2):754–762, 1999.
[8] Andrew V Biankin, Nicola Waddell, Karin S Kassahn, Marie-Claude Gingras,Lakshmi B Muthuswamy, Amber L Johns, David K Miller, Peter J Wilson, Ann-Marie Patch, Jianmin Wu, et al. Pancreatic cancer genomes reveal aberrationsin axon guidance pathway genes. Nature, 491(7424):399, 2012.
118
[9] SL Brenner and ED Korn. Substoichiometric concentrations of cytochalasin dinhibit actin polymerization. additional evidence for an f-actin treadmill. Jour-nal of Biological Chemistry, 254(20):9982–9985, 1979.
[10] Nia J Bryant, Robert C Piper, Sonja R Gerrard, and Tom H Stevens. Trafficinto the prevacuolar/endosomal compartment of saccharomyces cerevisiae: avps45-dependent intracellular route and a vps45-independent, endocytic route.European journal of cell biology, 76(1):43–52, 1998.
[11] Claudia Buerger, Ben DeVries, and Vuk Stambolic. Localization of rheb to theendomembrane is critical for its signaling function. Biochemical and biophysicalresearch communications, 344(3):869–880, 2006.
[12] James F Casella, Michael D Flanagan, and Shin Lin. Cytochalasin d inhibitsactin polymerization and induces depolymerization of actin filaments formedduring platelet shape change. Nature, 293(5830):302, 1981.
[13] Lynne Chantranupong, Sonia M Scaria, Robert A Saxton, Melanie P Gygi,Kuang Shen, Gregory A Wyant, Tim Wang, J Wade Harper, Steven P Gygi,and David M Sabatini. The castor proteins are arginine sensors for the mtorc1pathway. Cell, 165(1):153–164, 2016.
[14] Carlos Chih-Hsiung Chen, Peter J Schweinsberg, Shilpa Vashist, Darren PMareiniss, Eric J Lambie, and Barth D Grant. Rab-10 is required for endo-cytic recycling in the caenorhabditis elegans intestine. Molecular biology of thecell, 17(3):1286–1297, 2006.
[15] Gerald C Chu, Alec C Kimmelman, Aram F Hezel, and Ronald A De-Pinho. Stromal biology of pancreatic cancer. Journal of cellular biochemistry,101(4):887–907, 2007.
[16] Cosimo Commisso, Shawn M Davidson, Rengin G Soydaner-Azeloglu, Seth JParker, Jurre J Kamphorst, Sean Hackett, Elda Grabocka, Michel Nofal, Jef-frey A Drebin, Craig B Thompson, et al. Macropinocytosis of protein is anamino acid supply route in ras-transformed cells. Nature, 497(7451):633, 2013.
[17] Cosimo Commisso, Rory J Flinn, and Dafna Bar-Sagi. Determining themacropinocytic index of cells through a quantitative image-based assay. Natureprotocols, 9(1):182, 2014.
[18] Maria Francisca Coutinho. N-acetylglucosamine-1-phosphate transferase, al-pha/beta and gamma subunits (gnptab, gnptg). Handbook of Glycosyltrans-ferases and Related Genes, pages 1335–1347, 2014.
[19] Shawn M Davidson, Oliver Jonas, Mark A Keibler, Han Wei Hou, Alba Luengo,Jared R Mayers, Jeffrey Wyckoff, Amanda M Del Rosario, Matthew Whitman,Christopher R Chin, et al. Direct evidence for cancer-cell-autonomous extra-cellular protein catabolism in pancreatic tumors. Nature medicine, 23(2):235,2017.
119
[20] Hasker P Davis and Larry R Squire. Protein synthesis and memory: a review.Psychological bulletin, 96(3):518, 1984.
[21] Julia Dennemarker, Tobias Lohmuller, Sebastian Muller, Stephanie VargasAguilar, Desmond J Tobin, Christoph Peters, and Thomas Reinheckel. Impairedturnover of autophagolysosomes in cathepsin l deficiency. Biological chemistry,391(8):913–922, 2010.
[22] Jinsheng Dong, Hongfang Qiu, Minerva Garcia-Barrio, James Anderson, andAlan G Hinnebusch. Uncharged trna activates gcn2 by displacing the proteinkinase moiety from a bipartite trna-binding domain. Molecular cell, 6(2):269–279, 2000.
[23] John P Duffy, Guido Eibl, Howard A Reber, and Oscar J Hines. Influence ofhypoxia and neoangiogenesis on the growth of pancreatic cancer. Molecularcancer, 2(1):12, 2003.
[24] Alejo Efeyan, William C Comb, and David M Sabatini. Nutrient-sensing mech-anisms and pathways. Nature, 517(7534):302, 2015.
[25] Bradley Efron and Robert Tibshirani. Bootstrap methods for standard errors,confidence intervals, and other measures of statistical accuracy. Statistical sci-ence, pages 54–75, 1986.
[26] Jing Fan, Jurre J Kamphorst, Robin Mathew, Michelle K Chung, Eileen White,Tomer Shlomi, and Joshua D Rabinowitz. Glutamine-driven oxidative phospho-rylation is a major atp source in transformed mammalian cells in both normoxiaand hypoxia. Molecular systems biology, 9(1):712, 2013.
[27] Daniela Gaglio, Christian M Metallo, Paulo A Gameiro, Karsten Hiller,Lara Sala Danna, Chiara Balestrieri, Lilia Alberghina, Gregory Stephanopoulos,and Ferdinando Chiaradonna. Oncogenic k-ras decouples glucose and glutaminemetabolism to support cancer cell growth. Molecular systems biology, 7(1):523,2011.
[28] Minerva Garcia-Barrio, Jinsheng Dong, Sandra Ufano, and Alan G Hinnebusch.Association of gcn1–gcn20 regulatory complex with the n-terminus of eif2αkinase gcn2 is required for gcn2 activation. The EMBO journal, 19(8):1887–1899, 2000.
[29] Sven Geisler, Kira M Holmstrom, Diana Skujat, Fabienne C Fiesel, Oliver CRothfuss, Philipp J Kahle, and Wolfdieter Springer. Pink1/parkin-mediated mi-tophagy is dependent on vdac1 and p62/sqstm1. Nature cell biology, 12(2):119,2010.
[30] Alvaro Gonzalez, Nerea Varo, Estibaliz Alegre, Angel Diaz, and Ignacio Melero.Immunosuppression routed via the kynurenine pathway: a biochemical andpathophysiologic approach. Advances in clinical chemistry, 45:156–198, 2008.
120
[31] Stine Grimmer, Bo van Deurs, and Kirsten Sandvig. Membrane ruffling andmacropinocytosis in a431 cells require cholesterol. Journal of cell science,115(14):2953–2962, 2002.
[32] Xin Gu, Jose M Orozco, Robert A Saxton, Kendall J Condon, Grace Y Liu,Patrycja A Krawczyk, Sonia M Scaria, J Wade Harper, Steven P Gygi, andDavid M Sabatini. Samtor is an s-adenosylmethionine sensor for the mtorc1pathway. Science, 358(6364):813–818, 2017.
[33] John E Hall. Guyton and Hall textbook of medical physiology e-Book. ElsevierHealth Sciences, 2015.
[34] Jaeseok Han, Sung Hoon Back, Junguk Hur, Yu-Hsuan Lin, Robert Gilder-sleeve, Jixiu Shan, Celvie L Yuan, Dawid Krokowski, Shiyu Wang, Maria Hat-zoglou, et al. Er-stress-induced transcriptional regulation increases protein syn-thesis leading to cell death. Nature cell biology, 15(5):481, 2013.
[35] Heather P Harding, Isabel Novoa, Yuhong Zhang, Huiqing Zeng, Ron Wek,Matthieu Schapira, and David Ron. Regulated translation initiation controlsstress-induced gene expression in mammalian cells. Molecular cell, 6(5):1099–1108, 2000.
[36] Matthew J Hart, Marinella G Callow, Brian Souza, and Paul Polakis. Iqgap1, acalmodulin-binding protein with a rasgap-related domain, is a potential effectorfor cdc42hs. The EMBO journal, 15(12):2997–3005, 1996.
[37] Andrej Hasilik, Abdul Waheed, and Kurt von Figura. Enzymatic phosphoryla-tion of lysosomal enzymes in the presence of udp-n-acetylglucosamine. absenceof the activity in l-cell fibroblasts. Biochemical and biophysical research com-munications, 98(3):761–767, 1981.
[38] John WB Hershey and William C Merrick. Pathway and mechanism of initiationof protein synthesis. cold spring harbor monograph series, 39:33–88, 2000.
[39] Manuel Hidalgo. Pancreatic cancer. New England Journal of Medicine,362(17):1605–1617, 2010.
[40] Tatsuya Higashi, Tsuneo Saga, Yuji Nakamoto, Takayoshi Ishimori, Koji Fu-jimoto, Ryuichiro Doi, Masayuki Imamura, and Junji Konishi. Diagnosis ofpancreatic cancer using fluorine-18 fluorodeoxyglucose positron emission tomog-raphy (fdg pet)usefulness and limitations in clinical reality. Annals of nuclearmedicine, 17(4):261, 2003.
[41] Sunil R Hingorani, Emanuel F Petricoin III, Anirban Maitra, Vinodh Ra-japakse, Catrina King, Michael A Jacobetz, Sally Ross, Thomas P Conrads,Timothy D Veenstra, Ben A Hitt, et al. Preinvasive and invasive ductal pan-creatic cancer and its early detection in the mouse. Cancer cell, 4(6):437–450,2003.
121
[42] Alan G Hinnebusch. Translational regulation of gcn4 and the general aminoacid control of yeast. Annu. Rev. Microbiol., 59:407–450, 2005.
[43] Chieh Hsu, Yuichi Morohashi, Shin-ichiro Yoshimura, Natalia Manrique-Hoyos,SangYong Jung, Marcel A Lauterbach, Mostafa Bakhti, Mads Grønborg,Wiebke Mobius, JeongSeop Rhee, et al. Regulation of exosome secretion byrab35 and its gtpase-activating proteins tbc1d10a–c. The Journal of cell biol-ogy, 189(2):223–232, 2010.
[44] Reinhard Jahn and Richard H Scheller. Snaresengines for membrane fusion.Nature reviews Molecular cell biology, 7(9):631, 2006.
[45] Milind M Javle, Rachna T Shroff, Henry Xiong, Gauri A Varadhachary, DavidFogelman, Shrikanth A Reddy, Darren Davis, Yujian Zhang, Robert A Wolff,and James L Abbruzzese. Inhibition of the mammalian target of rapamycin(mtor) in advanced pancreatic cancer: results of two phase ii studies. BMCcancer, 10(1):368, 2010.
[46] Yoshiaki Kamada, Ken-ichi Yoshino, Chika Kondo, Tomoko Kawamata, NorikoOshiro, Kazuyoshi Yonezawa, and Yoshinori Ohsumi. Tor directly controls theatg1 kinase complex to regulate autophagy. Molecular and cellular biology,30(4):1049–1058, 2010.
[47] Jurre J Kamphorst, Justin R Cross, Jing Fan, Elisa de Stanchina, RobinMathew, Eileen P White, Craig B Thompson, and Joshua D Rabinowitz. Hy-poxic and ras-transformed cells support growth by scavenging unsaturated fattyacids from lysophospholipids. Proceedings of the National Academy of Sciences,110(22):8882–8887, 2013.
[48] Jurre J Kamphorst, Michel Nofal, Cosimo Commisso, Sean R Hackett, WenyunLu, Elda Grabocka, Matthew G Vander Heiden, George Miller, Jeffrey ADrebin, Dafna Bar-Sagi, et al. Human pancreatic cancer tumors are nutrientpoor and tumor cells actively scavenge extracellular protein. Cancer research,75(3):544–553, 2015.
[49] Kousuke Kasahara, Yuji Nakayama, Izumi Sato, Kikuko Ikeda, Masaki Hoshino,Takeshi Endo, and Naoto Yamaguchi. Role of src-family kinases in formationand trafficking of macropinosomes. Journal of cellular physiology, 211(1):220–232, 2007.
[50] Heidrun KIRSCHKE, Jurgen LANGER, Bernd WIEDERANDERS, SiegfriedANSORGE, and Peter BOHLEY. Cathepsin l: A new proteinase from rat-liverlysosomes. European journal of biochemistry, 74(2):293–301, 1977.
[51] Mirkka Koivusalo, Christopher Welch, Hisayoshi Hayashi, Cameron C Scott,Moshe Kim, Todd Alexander, Nicolas Touret, Klaus M Hahn, and Sergio Grin-stein. Amiloride inhibits macropinocytosis by lowering submembranous ph and
122
preventing rac1 and cdc42 signaling. The Journal of cell biology, 188(4):547–563, 2010.
[52] Ruirui Kong, Fengshuang Yi, Pushuai Wen, Jianghong Liu, Xiaoping Chen,Jinqi Ren, Xiaofei Li, Yulong Shang, Yongzhan Nie, Kaichun Wu, et al. Myo9bis a key player in slit/robo-mediated lung tumor suppression. The Journal ofclinical investigation, 125(12):4407–4420, 2015.
[53] Albert C Koong, Vivek K Mehta, Quynh T Le, George A Fisher, David J Terris,J Martin Brown, Augusto J Bastidas, and Mark Vierra. Pancreatic tumors showhigh levels of hypoxia. International Journal of Radiation Oncology* Biology*Physics, 48(4):919–922, 2000.
[54] Roderik M Kortlever, Nicole M Sodir, Catherine H Wilson, Deborah LBurkhart, Luca Pellegrinet, Lamorna Brown Swigart, Trevor D Littlewood,and Gerard I Evan. Myc cooperates with ras by programming inflammationand immune suppression. Cell, 171(6):1301–1315, 2017.
[55] Ilektra Kouranti, Martin Sachse, Nassim Arouche, Bruno Goud, and ArnaudEchard. Rab35 regulates an endocytic recycling pathway essential for the ter-minal steps of cytokinesis. Current biology, 16(17):1719–1725, 2006.
[56] Matthias Krause, Erik W Dent, James E Bear, Joseph J Loureiro, and Frank BGertler. Ena/vasp proteins: regulators of the actin cytoskeleton and cell mi-gration. Annual review of cell and developmental biology, 19(1):541–564, 2003.
[57] Shinya Kuroda, Masaki Fukata, Kenta Kobayashi, Masato Nakafuku, NobuoNomura, Akihiro Iwamatsu, and Kozo Kaibuchi. Identification of iqgap as aputative target for the small gtpases, cdc42 and rac1. Journal of BiologicalChemistry, 271(38):23363–23367, 1996.
[58] Mathieu Laplante and David M Sabatini. mtor signaling in growth control anddisease. Cell, 149(2):274–293, 2012.
[59] Guangpu Li, Crislyn DSouza-Schorey, M Alejandro Barbieri, Jonathan ACooper, and Philip D Stahl. Uncoupling of membrane ruffling and pinocytosisduring ras signal transduction. Journal of Biological Chemistry, 272(16):10337–10340, 1997.
[60] Piro Lito, Anna Saborowski, Jingyin Yue, Martha Solomon, Eric Joseph, Sun-yana Gadal, Michael Saborowski, Edward Kastenhuber, Christof Fellmann,Kazuhiro Ohara, et al. Disruption of craf-mediated mek activation is requiredfor effective mek inhibition in kras mutant tumors. Cancer cell, 25(5):697–710,2014.
[61] Wenyun Lu, Bryson D Bennett, and Joshua D Rabinowitz. Analytical strate-gies for lc–ms-based targeted metabolomics. Journal of Chromatography B,871(2):236–242, 2008.
123
[62] Wenyun Lu, Michelle F Clasquin, Eugene Melamud, Daniel Amador-Noguez,Amy A Caudy, and Joshua D Rabinowitz. Metabolomic analysis via reversed-phase ion-pairing liquid chromatography coupled to a stand alone orbitrap massspectrometer. Analytical chemistry, 82(8):3212–3221, 2010.
[63] Wenyun Lu, Elizabeth Kimball, and Joshua D Rabinowitz. A high-performanceliquid chromatography-tandem mass spectrometry method for quantitation ofnitrogen-containing intracellular metabolites. Journal of the American Societyfor Mass Spectrometry, 17(1):37–50, 2006.
[64] Xiaoju Max Ma and John Blenis. Molecular mechanisms of mtor-mediatedtranslational control. Nature reviews Molecular cell biology, 10(5):307, 2009.
[65] Laura M Machesky, R Dyche Mullins, Henry N Higgs, Donald A Kaiser, LaurentBlanchoin, Robin C May, Margaret E Hall, and Thomas D Pollard. Scar, a wasp-related protein, activates nucleation of actin filaments by the arp2/3 complex.Proceedings of the National Academy of Sciences, 96(7):3739–3744, 1999.
[66] Daruka Mahadevan and Daniel D Von Hoff. Tumor-stroma interactions in pan-creatic ductal adenocarcinoma. Molecular cancer therapeutics, 6(4):1186–1197,2007.
[67] MJ Marton, D Crouch, and AG Hinnebusch. Gcn1, a translational activator ofgcn4 in saccharomyces cerevisiae, is required for phosphorylation of eukaryotictranslation initiation factor 2 by protein kinase gcn2. Molecular and cellularbiology, 13(6):3541–3556, 1993.
[68] K Maruyama. Effect of β-actinin on the particle length of f-actin. Biochimica etBiophysica Acta (BBA)-Biophysics including Photosynthesis, 126(2):389–398,1966.
[69] Koscak Maruyama, Sumiko KIMURA, Tetsuro ISHII, Masaaki KURODA,Kazuyo OHASHI, and Shigeru MURAMATSU. β-actinin, a regulatory pro-tein of muscle. The Journal of Biochemistry, 81(1):215–232, 1977.
[70] Eugene Melamud, Livia Vastag, and Joshua D Rabinowitz. Metabolomic anal-ysis and visualization engine for lc- ms data. Analytical chemistry, 82(23):9818–9826, 2010.
[71] David O Morgan, Jeffrey C Edman, David N Standring, Victor A Fried,Michele C Smith, Richard A Roth, and William J Rutter. Insulin-like growthfactor ii receptor as a multifunctional binding protein. Nature, 329(6137):301,1987.
[72] Michel Nofal and Joshua D Rabinowitz. Ribosomes on the night shift. Science,360(6390):710–711, 2018.
124
[73] Michel Nofal, Kevin Zhang, Seunghun Han, and Joshua D Rabinowitz. mtorinhibition restores amino acid balance in cells dependent on catabolism of ex-tracellular protein. Molecular cell, 67(6):936–946, 2017.
[74] Kenneth P Olive, Michael A Jacobetz, Christian J Davidson, AarthiGopinathan, Dominick McIntyre, Davina Honess, Basetti Madhu, Mae A Gold-graben, Meredith E Caldwell, David Allard, et al. Inhibition of hedgehog signal-ing enhances delivery of chemotherapy in a mouse model of pancreatic cancer.Science, 2009.
[75] Takanobu Otomo, Katsumi Higaki, Eiji Nanba, Keiichi Ozono, and Norio Sakai.Lysosomal storage causes cellular dysfunction in mucolipidosis ii skin fibrob-lasts. Journal of Biological Chemistry, 286(40):35283–35290, 2011.
[76] George Palade. Intracellular aspects of the process of protein synthesis. Science,189(4200):347–358, 1975.
[77] Wilhelm Palm, Jingwen Araki, Bryan King, Raymond G DeMatteo, andCraig B Thompson. Critical role for pi3-kinase in regulating the use of proteinsas an amino acid source. Proceedings of the National Academy of Sciences, page201712726, 2017.
[78] Wilhelm Palm, Youngkyu Park, Kevin Wright, Natalya N Pavlova, David ATuveson, and Craig B Thompson. The utilization of extracellular proteins asnutrients is suppressed by mtorc1. Cell, 162(2):259–270, 2015.
[79] Armen Parsyan, David Shahbazian, Yvan Martineau, Emmanuel Petroulakis,Tommy Alain, Ola Larsson, Geraldine Mathonnet, Gritta Tettweiler, Christo-pher U Hellen, Tatyana V Pestova, et al. The helicase protein dhx29 promotestranslation initiation, cell proliferation, and tumorigenesis. Proceedings Of TheNational Academy Of Sciences, pages pnas–0909773106, 2009.
[80] Joseph J Pesek, Maria T Matyska, Steven M Fischer, and Theodore RSana. Analysis of hydrophilic metabolites by high-performance liquidchromatography–mass spectrometry using a silica hydride-based stationaryphase. Journal of chromatography A, 1204(1):48–55, 2008.
[81] Suzanne Pfeffer and Dikran Aivazian. Targeting rab gtpases to distinct mem-brane compartments. Nature reviews Molecular cell biology, 5(11):886, 2004.
[82] Suzanne R Pfeffer. Rab gtpases: specifying and deciphering organelle identityand function. Trends in cell biology, 11(12):487–491, 2001.
[83] Vera P Pisareva, Andrey V Pisarev, Anton A Komar, Christopher UT Hellen,and Tatyana V Pestova. Translation initiation on mammalian mrnas with struc-tured 5 utrs requires dexh-box protein dhx29. Cell, 135(7):1237–1250, 2008.
125
[84] Maaike S Pols, Corlinda ten Brink, Prajakta Gosavi, Viola Oorschot, and JudithKlumperman. The hops proteins hvps41 and hvps39 are required for homotypicand heterotypic late endosome fusion. Traffic, 14(2):219–232, 2013.
[85] Paolo P Provenzano, Carlos Cuevas, Amy E Chang, Vikas K Goel, Daniel DVon Hoff, and Sunil R Hingorani. Enzymatic targeting of the stroma ablatesphysical barriers to treatment of pancreatic ductal adenocarcinoma. Cancercell, 21(3):418–429, 2012.
[86] Jing Pu, Christina Schindler, Rui Jia, Michal Jarnik, Peter Backlund, andJuan S Bonifacino. Borc, a multisubunit complex that regulates lysosome po-sitioning. Developmental cell, 33(2):176–188, 2015.
[87] Bjørn Quistorff and Egon Pedersen. A new device for the freeze-clamping oftissue samples. Analytical biochemistry, 73(1):236–239, 1976.
[88] Renata CM Reis, Marcos HF Sorgine, and Tatiana Coelho-Sampaio. A novelmethodology for the investigation of intracellular proteolytic processing in intactcells. European journal of cell biology, 75(2):192–197, 1998.
[89] Xuefeng Ren, Ginny G Farıas, Bertram J Canagarajah, Juan S Bonifacino, andJames H Hurley. Structural basis for recruitment and activation of the ap-1clathrin adaptor complex by arf1. Cell, 152(4):755–767, 2013.
[90] Jochen Rink, Eric Ghigo, Yannis Kalaidzidis, and Marino Zerial. Rab conversionas a mechanism of progression from early to late endosomes. Cell, 122(5):735–749, 2005.
[91] Eugene Roberts and Sam Frankel. Free amino acids in normal and neoplastictissues of mice as studied by paper chromatography. Cancer research, 9(11):645–648, 1949.
[92] Eugene Roberts and Tatsuya Tanaka. Free amino acids of the yoshida ascitestumor. Cancer research, 16(3):204–210, 1956.
[93] Klemens Rottner, Barbara Behrendt, J Victor Small, and Jurgen Wehland.Vasp dynamics during lamellipodia protrusion. Nature cell biology, 1(5):321,1999.
[94] Nisebita Sahu, Darlene Dela Cruz, Min Gao, Wendy Sandoval, Peter M Haverty,Jinfeng Liu, Jean-Philippe Stephan, Benjamin Haley, Marie Classon, GeorgiaHatzivassiliou, et al. Proline starvation induces unresolved er stress and hindersmtorc1-dependent tumorigenesis. Cell metabolism, 24(5):753–761, 2016.
[95] Yasemin Sancak, Liron Bar-Peled, Roberto Zoncu, Andrew L Markhard,Shigeyuki Nada, and David M Sabatini. Ragulator-rag complex targets mtorc1to the lysosomal surface and is necessary for its activation by amino acids. Cell,141(2):290–303, 2010.
126
[96] Neville E Sanjana, Ophir Shalem, and Feng Zhang. Improved vectors andgenome-wide libraries for crispr screening. Nature methods, 11(8):783, 2014.
[97] Hiroyuki Sano, Lorena Eguez, Mary N Teruel, Mitsunori Fukuda, Tuan DChuang, Jose A Chavez, Gustav E Lienhard, and Timothy E McGraw. Rab10,a target of the as160 rab gap, is required for insulin-stimulated translocation ofglut4 to the adipocyte plasma membrane. Cell metabolism, 5(4):293–303, 2007.
[98] Carsten Schnatwinkel, Savvas Christoforidis, Margaret R Lindsay, SandrineUttenweiler-Joseph, Matthias Wilm, Robert G Parton, and Marino Zerial. Therab5 effector rabankyrin-5 regulates and coordinates different endocytic mech-anisms. PLoS biology, 2(9):e261, 2004.
[99] Lawrence D Schweitzer, William C Comb, Liron Bar-Peled, and David M Saba-tini. Disruption of the rag-ragulator complex by c17orf59 inhibits mtorc1. Cellreports, 12(9):1445–1455, 2015.
[100] Matthew Scott, Carl W Gunderson, Eduard M Mateescu, Zhongge Zhang, andTerence Hwa. Interdependence of cell growth and gene expression: origins andconsequences. Science, 330(6007):1099–1102, 2010.
[101] Darren F Seals, Gary Eitzen, Nathan Margolis, William T Wickner, and Al-bert Price. A ypt/rab effector complex containing the sec1 homolog vps33p isrequired for homotypic vacuole fusion. Proceedings of the national academy ofsciences, 97(17):9402–9407, 2000.
[102] Tomer Shlomi, Jing Fan, Baiqing Tang, Warren D Kruger, and Joshua D Ra-binowitz. Quantitation of cellular metabolic fluxes of methionine. Analyticalchemistry, 86(3):1583–1591, 2014.
[103] Rebecca L Siegel, Kimberly D Miller, and Ahmedin Jemal. Cancer statistics,2017. CA: a cancer journal for clinicians, 67(1):7–30, 2017.
[104] Flore Sinturel, Alan Gerber, Daniel Mauvoisin, Jingkui Wang, David Gatfield,Jeremy J Stubblefield, Carla B Green, Frederic Gachon, and Ueli Schibler.Diurnal oscillations in liver mass and cell size accompany ribosome assemblycycles. Cell, 169(4):651–663, 2017.
[105] Jachen A Solinger and Anne Spang. Tethering complexes in the endocyticpathway: Corvet and hops. The FEBS journal, 280(12):2743–2757, 2013.
[106] Jaekyoung Son, Costas A Lyssiotis, Haoqiang Ying, Xiaoxu Wang, Sujun Hua,Matteo Ligorio, Rushika M Perera, Cristina R Ferrone, Edouard Mullarky,Ng Shyh-Chang, et al. Glutamine supports pancreatic cancer growth througha kras-regulated metabolic pathway. Nature, 496(7443):101, 2013.
[107] Gerd Stehle, Hannsjorg Sinn, Andreas Wunder, Hans Hermann Schrenk,J Charles M Stewart, Gernot Hartung, Wolfgang Maier-Borst, and Dieter Lud-wig Heene. Plasma protein (albumin) catabolism by the tumor itselfimplications
127
for tumor metabolism and the genesis of cachexia. Critical reviews in oncol-ogy/hematology, 26(2):77–100, 1997.
[108] John D Storey and Robert Tibshirani. Statistical significance for genomewidestudies. Proceedings of the National Academy of Sciences, 100(16):9440–9445,2003.
[109] Thomas C Sudhof and James E Rothman. Membrane fusion: grappling withsnare and sm proteins. Science, 323(5913):474–477, 2009.
[110] Tingting Sun, Nicola Aceto, Kristen L Meerbrey, Jessica D Kessler, ChunshuiZhou, Ilenia Migliaccio, Don X Nguyen, Natalya N Pavlova, Maria Botero, JianHuang, et al. Activation of multiple proto-oncogenic tyrosine kinases in breastcancer via loss of the ptpn12 phosphatase. Cell, 144(5):703–718, 2011.
[111] Joel A Swanson. Shaping cups into phagosomes and macropinosomes. Naturereviews Molecular cell biology, 9(8):639, 2008.
[112] Joel A Swanson and Colin Watts. Macropinocytosis. Trends in cell biology,5(11):424–428, 1995.
[113] Carson C Thoreen, Lynne Chantranupong, Heather R Keys, Tim Wang,Nathanael S Gray, and David M Sabatini. A unifying model for mtorc1-mediated regulation of mrna translation. Nature, 485(7396):109, 2012.
[114] Miki Tsukada and Yoshinori Ohsumi. Isolation and characterization ofautophagy-defective mutants of saccharomyces cerevisiae. FEBS letters, 333(1-2):169–174, 1993.
[115] Joel Vandekerckhove and Klaus Weber. At least six different actins are ex-pressed in a higher mammal: an analysis based on the amino acid sequence ofthe amino-terminal tryptic peptide. Journal of molecular biology, 126(4):783–802, 1978.
[116] Matthew G Vander Heiden, Lewis C Cantley, and Craig B Thompson. Under-standing the warburg effect: the metabolic requirements of cell proliferation.science, 324(5930):1029–1033, 2009.
[117] MG Vander Heiden, SY Lunt, TL Dayton, BP Fiske, WJ Israelsen, KR Mat-taini, NI Vokes, G Stephanopoulos, LC Cantley, CM Metallo, et al. Metabolicpathway alterations that support cell proliferation. In Cold Spring Harbor sym-posia on quantitative biology, volume 76, pages 325–334. Cold Spring HarborLaboratory Press, 2011.
[118] CR Vazquez de Aldana, MJ Marton, and AG Hinnebusch. Gcn20, a novel atpbinding cassette protein, and gcn1 reside in a complex that mediates activationof the eif-2 alpha kinase gcn2 in amino acid-starved cells. The EMBO journal,14(13):3184–3199, 1995.
128
[119] Igor Vivanco and Charles L Sawyers. The phosphatidylinositol 3-kinase–aktpathway in human cancer. Nature Reviews Cancer, 2(7):489, 2002.
[120] Daniel D Von Hoff, Thomas Ervin, Francis P Arena, E Gabriela Chiorean,Jeffrey Infante, Malcolm Moore, Thomas Seay, Sergei A Tjulandin, Wen WeeMa, Mansoor N Saleh, et al. Increased survival in pancreatic cancer with nab-paclitaxel plus gemcitabine. New England Journal of Medicine, 369(18):1691–1703, 2013.
[121] Daniel D Von Hoff, Ramesh K Ramanathan, Mitesh J Borad, Daniel A Laheru,Lon S Smith, Tina E Wood, Ronald L Korn, Neil Desai, Vuong Trieu, Jose LIglesias, et al. Gemcitabine plus nab-paclitaxel is an active regimen in patientswith advanced pancreatic cancer: a phase i/ii trial. Journal of Clinical Oncology,29(34):4548, 2011.
[122] Tim Wang, Kıvanc Birsoy, Nicholas W Hughes, Kevin M Krupczak, Yorick Post,Jenny J Wei, Eric S Lander, and David M Sabatini. Identification and character-ization of essential genes in the human genome. Science, 350(6264):1096–1101,2015.
[123] Tim Wang, Jenny J Wei, David M Sabatini, and Eric S Lander. Genetic screensin human cells using the crispr/cas9 system. Science, page 1246981, 2013.
[124] Sheree A Wek, Shuhao Zhu, and Ronald C Wek. The histidyl-trna synthetase-related sequence in the eif-2 alpha protein kinase gcn2 interacts with trna andis required for activation in response to starvation for different amino acids.Molecular and cellular biology, 15(8):4497–4506, 1995.
[125] Douglas B Wheeler, Roberto Zoncu, David E Root, David M Sabatini, andCharles L Sawyers. Identification of an oncogenic rab protein. Science,350(6257):211–217, 2015.
[126] Rachel L Wolfson, Lynne Chantranupong, Robert A Saxton, Kuang Shen, So-nia M Scaria, Jason R Cantor, and David M Sabatini. Sestrin2 is a leucinesensor for the mtorc1 pathway. Science, page aab2674, 2015.
[127] Albert Wollenberger, Otto Ristau, and Georg Schoffa. Eine einfache technikder extrem schnellen abkuhlung großerer gewebestucke. Pfluger’s Archiv furdie gesamte Physiologie des Menschen und der Tiere, 270(4):399–412, 1960.
[128] Brian M Wolpin, Aram F Hezel, Thomas Abrams, Lawrence S Blaszkowsky,Jeffrey A Meyerhardt, Jennifer A Chan, Peter C Enzinger, Brittany Allen,Jeffrey W Clark, David P Ryan, et al. Oral mtor inhibitor everolimus in patientswith gemcitabine-refractory metastatic pancreatic cancer. Journal of ClinicalOncology, 27(2):193, 2009.
[129] Gregory A Wyant, Monther Abu-Remaileh, Evgeni M Frenkel, Nouf N Laq-tom, Vimisha Dharamdasani, Caroline A Lewis, Sze Ham Chan, Ivonne Heinze,
129
Alessandro Ori, and David M Sabatini. Nufip1 is a ribosome receptor forstarvation-induced ribophagy. Science, page eaar2663, 2018.
[130] Gregory A Wyant, Monther Abu-Remaileh, Rachel L Wolfson, Walter W Chen,Elizaveta Freinkman, Laura V Danai, Matthew G Vander Heiden, and David MSabatini. mtorc1 activator slc38a9 is required to efflux essential amino acidsfrom lysosomes and use protein as a nutrient. Cell, 171(3):642–654, 2017.
[131] Shenghong Yang, Xiaoxu Wang, Gianmarco Contino, Marc Liesa, Ergun Sahin,Haoqiang Ying, Alexandra Bause, Yinghua Li, Jayne M Stommel, GiacomoDell’Antonio, et al. Pancreatic cancers require autophagy for tumor growth.Genes & development, 2011.
[132] Haoqiang Ying, Alec C Kimmelman, Costas A Lyssiotis, Sujun Hua, Ger-ald C Chu, Eliot Fletcher-Sananikone, Jason W Locasale, Jaekyoung Son, HaileiZhang, Jonathan L Coloff, et al. Oncogenic kras maintains pancreatic tumorsthrough regulation of anabolic glucose metabolism. Cell, 149(3):656–670, 2012.
[133] Tamotsu Yoshimori, Akitsugu Yamamoto, Yoshinori Moriyama, Masamitsu Fu-tai, and Yutaka Tashiro. Bafilomycin a1, a specific inhibitor of vacuolar-type h(+)-atpase, inhibits acidification and protein degradation in lysosomes of cul-tured cells. Journal of Biological Chemistry, 266(26):17707–17712, 1991.
[134] Regina M Young, Daniel Ackerman, Zachary L Quinn, Anthony Man-cuso, Michaela Gruber, Liping Liu, Dionysios N Giannoukos, EkaterinaBobrovnikova-Marjon, J Alan Diehl, Brian Keith, et al. Dysregulated mtorc1renders cells critically dependent on desaturated lipids for survival under tumor-like stress. Genes & development, 27(10):1115–1131, 2013.
[135] Li Yu, Christina K McPhee, Lixin Zheng, Gonzalo A Mardones, YueguangRong, Junya Peng, Na Mi, Ying Zhao, Zhihua Liu, Fengyi Wan, et al. Termi-nation of autophagy and reformation of lysosomes regulated by mtor. Nature,465(7300):942, 2010.
[136] Jihye Yun, Carlo Rago, Ian Cheong, Ray Pagliarini, Philipp Angenendt, HarithRajagopalan, Kerstin Schmidt, James KV Willson, Sandy Markowitz, ShibinZhou, et al. Glucose deprivation contributes to the development of kras pathwaymutations in tumor cells. Science, 325(5947):1555–1559, 2009.
[137] Anja Zeigerer, Jerome Gilleron, Roman L Bogorad, Giovanni Marsico, HidenoriNonaka, Sarah Seifert, Hila Epstein-Barash, Satya Kuchimanchi, Chang GengPeng, Vera M Ruda, et al. Rab5 is necessary for the biogenesis of the endolyso-somal system in vivo. Nature, 485(7399):465, 2012.
[138] Jinghui Zhao, Jeffrey J Brault, Andreas Schild, Peirang Cao, Marco Sandri, Ste-fano Schiaffino, Stewart H Lecker, and Alfred L Goldberg. Foxo3 coordinatelyactivates protein degradation by the autophagic/lysosomal and proteasomalpathways in atrophying muscle cells. Cell metabolism, 6(6):472–483, 2007.
130
[139] Helene Zinszner, Masahiko Kuroda, XiaoZhong Wang, Nikoleta Batchvarova,Richard T Lightfoot, Helen Remotti, James L Stevens, and David Ron. Chopis implicated in programmed cell death in response to impaired function of theendoplasmic reticulum. Genes & development, 12(7):982–995, 1998.
131