VIEWPOINT
Translational Proteomics
Addressing the Challenges of High-Throughput Cancer
Tissue Proteomics for Clinical Application: ProCan
Brett Tully,* Rosemary L. Balleine, Peter G. Hains, Qing Zhong, Roger R. Reddel,
and Phillip J. Robinson
The cancer tissue proteome has enormous potential as a source of novel
predictive biomarkers in oncology. Progress in the development of mass
spectrometry (MS)-based tissue proteomics now presents an opportunity to
exploit this by applying the strategies of comprehensive molecular profiling
and big-data analytics that are refined in other fields of ‘omics research.
ProCan (ProCan is a registered trademark) is a program aiming to generate
high-quality tissue proteomic data across a broad spectrum of cancer types. It
is based on data-independent acquisition–MS proteomic analysis of
annotated tissue samples sourced through collaboration with expert clinical
and cancer research groups. The practical requirements of a high-throughput
translational research program have shaped the approach that ProCan is
taking to address challenges in study design, sample preparation, raw data
acquisition, and data analysis. The ultimate goal is to establish a large
proteomics knowledge-base that, in combination with other cancer ‘omics
data, will accelerate cancer research.
This era of possibility for matching specific anti-cancer agents to
points of vulnerability in cancer cells has led to an urgent search
for novel predictive biomarkers. The drive comes from experi-
ence with highly successful therapy-biomarker matches includ-
ing detection of direct therapeutic targets in cancer cells, such as
HER2 amplification in breast cancer,[1] and more indirect indica-
tors of benefit such as BRCA1/2 mutation status and response to
PARP1 inhibitors in ovarian cancer.[2] It has also become integral
to the design of early phase clinical trials of molecularly targeted
agents, with biomarkers forming part of inclusion criteria, even
across disparate cancer types.[3]
The evolution of methods for high-throughput genomic and
transcriptomic profiling of cancer has enabled this field of dis-
covery and practice change. A major limitation is that genomics
analysis currently succeeds in identifying “actionable” abnor-
malities in a minority of cancer patients,[4] and robust predictive
Dr. B. Tully, Prof. R. L. Balleine, Dr. P. G. Hains, Dr. Q. Zhong,
Prof. R. R. Reddel, Prof. P. J. Robinson
ProCan
Children’s Medical Research Institute
Faculty of Medicine and Health
The University of Sydney
Westmead, NSW 2145, Australia
E-mail: btully@cmri.org.au
The ORCID identification number(s) for the author(s) of this article
can be found under https://doi.org/10.1002/pmic.201900109
DOI: 10.1002/pmic.201900109
Proteomics 2019, 1900109 1900109 (1 of 5)
www.proteomics-journal.com
biomarkers are lacking for key targeted
therapeutics, including immune check-
point inhibitors and VEGF-targeted
agents.[5,6]
The cancer tissue proteome is an
under-explored domain with huge po-
tential for novel biomarker discovery
because proteins are the chief structural
and signaling molecules in cancer cells.
Moreover, the opportunity to profile the
complex cancer tissue milieu could be
informative for predicting response to
agents that have indirect modes of ac-
tion. To date, the contribution of cancer
tissue proteomics to biomarker discovery
has been hampered by poor reproduci-
bility. However, the development of high-
sensitivity quantitative MS techniques
coupled with improvements in tissue-
processing methods and big data
analytics now advance the prospects for both biomarker discovery
and clinical application.
There are many challenges to overcome before the potential
of proteomics in oncology is realized. Chief among these is to
establish a knowledge-base of sufficient size and scope to address
complex clinical issues. To date, MS-based tissue proteomics has
been a relatively low-throughput technology and most cohorts in
published cancer studies are small. To enable robust discovery, a
large number of cancer samples must be analyzed in a consistent,
or at least a comparable way. The corollary is that a commitment
is needed to develop core processes, methods, and infrastructure
that can operate at scale, and without compromise to the central
tenets of analytical validity and clinical utility that are critical for
clinical application.
1. ProCan
The Australian Cancer Research Foundation International Cen-
tre for the Proteomics of Human Cancer (ProCan) is an effort
to advance the field of cancer diagnostics by contributing pan-
cancer tissue proteomics data. At its core is a custom-built MS
facility currently housing six instruments that function as a sin-
gle operational unit with capacity of around 10 000 sample runs
per year. Over the next 5 years, ProCan aims to profile can-
cers, pediatric and adult, solid and hematologic, and represent-
ing most known cancer types through a network of collaborations
with expert clinicians and other cancer researchers. The overall
goal is to develop clinically applicable biomarkers by addition of

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.proteomics-journal.com

Figure 1. ProCan flywheel. The principal objective of ProCan is to build a body of insight and knowledge about cancer. This is an iterative process where
each cycle, driven by a single question of unmet clinical need, enables a body of knowledge to be created and the proteomic landscape of cancer to
be expanded. Around the wheel, there are loci of activities, each of which contain many challenges and decision points, especially those related to the
practicalities of performing high-quality science in a biobank-scale context. As the landscape grows, pan-cancer analyses become possible, offering a
new and unique perspective that we expect will enable insights not obtained through other means.

high-quality proteomics data to existing stores of clinical and
molecular information (Figure 1).
The ProCan program is structured as a series of individual
hypothesis-driven research studies focused on individual cancer
types that combine to form a pan-cancer knowledge-base over the
course of the program. Operational decision-making has been
shaped by the practical requirements of clinical research and
high-throughput proteomic studies performed at biobank-scale.
Here, we describe major challenges we have identified and our
approach to addressing some of these.
1.1. Study Design
Individual studies that comprise the majority of ProCan research
are conducted in collaboration with expert clinical or research
groups. Priority research collaborations are based on the avail-
ability of tissue sample collections that are richly annotated with
clinical data, and ideally have other ‘omic data available for inte-
grated analysis with proteomics (Figure 2). A major considera-
tion in study design is to ensure that the hypotheses to be tested
address areas of “unmet clinical need” relevant to specific can-
cers. In this respect, input from domain-expert collaborators at
the stage of project planning and monitoring is critical. In the fu-
ture, biomarker-seeking sub-studies integrated into clinical trials
may provide opportunities to facilitate these interactions.
A challenge in conducting a large number of collaborative
projects across institutions and jurisdictions has been to de-
velop systems to manage regulatory processes, including human
research ethics/institutional review board approvals, materials
transfer, and other inter-institutional agreements. The time and
resource cost of this aspect is substantial. An advantage is that ex-
perience in this area has been advanced by the conduct of large
international collaborative studies over the past two decades, and
our experience indicates a high level of familiarity with relevant
issues as well as dedicated expertise available in many institu-
tions to support researchers dealing with these challenges.
The key to clinical application of tissue-based proteomics is to
adapt to the practical requirements of clinical workflow. In the
discovery phase, a major implication for ProCan is to prioritize
analysis of formalin-fixed paraffin-embedded (FFPE) over fresh-
frozen (FF) tissue samples where possible. Our experience with
FFPE tissue proteomics is consistent with other reports show-
ing that high-quality data can be generated and that the scale and
scope of quantifiable proteins is comparable with FF tissues.[7]
However, there may be differences in the proteomic profiles of
FFPE and FF samples that require definition of distinct classi-
fiers for the two sample types. Furthermore, in order to ensure
throughput, we are focusing on analysis of unmodified peptides.
However, re-analysis of publicly available ProCan data by us or
others will enable questions currently out of scope to be explored
in the future.
1.2. Sample Preparation
Cancer tissue samples are a complex mix of malignant, reac-
tive, and normal elements and diagnostic interpretation relies
on morphological assessment by a specialist pathologist. To in-
terpret data from proteomics, which is tissue disruptive, it is im-
perative that the components of a sample submitted for MS are
known. Isolation of cancer cells from tissue by microdissection
is a strategy to directly assess their proteomic features. However,
this is a very labor-intensive process that is not well suited to
high-throughput, or clinical application. It is also likely that a pro-
teomic profile from the tumor microenvironment will be infor-
mative and is important to capture.
The strategy that ProCan is using to address these issues
is to prepare adjacent sections from tissue blocks for matched
histopathologic review and proteomics analysis. A technical chal-
lenge that this has created in the case of FF samples is to
develop efficient methods for removal of optimal cutting tem-
perature compound prior to MS. In the future, it may be-
come possible to deconvolute a tissue proteomic profile to

Proteomics 2019, 1900109 1900109 (2 of 5) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.proteomics-journal.com

Figure 2. Schematic of ProCan matrix. ProCan aims to survey the proteomic landscape of all types of cancer via a series of individual collaborative projects
(represented here as radial panels). A typical project is designed to address a clinically relevant question by analyzing a cohort of cancer samples that
has been collected by a collaborating research group and annotated with clinical data, including response to treatment and survival. Wherever possible,
a matched section of the tumor—immediately adjacent to tissue processed for mass spectrometry—is analyzed by histopathology review. Cohorts for
which various combinations of other ‘omic data are available (indicated here by colored segments; WGS, whole genome sequencing; WES, whole exome
sequencing) are prioritized, so that correlation between proteomics and other ’omics can be addressed. The use of a single proteomics technology
across all ProCan projects will facilitate a pan-cancer overview of the human cancer tissue proteome.

estimate the contribution of different tissue elements, as has
been achieved to a degree with transcriptomic data.[8] ProCan is
curating a large-scale database of digitized whole slide images
anticipating that it will become possible for machine learning
techniques to enable more direct integration of histopathology
and tissue proteomic profiles.
After samples are collected, tissue lysis and digestion protocols
must be rapid, efficient, reproducible, and broadly applicable to
tissues of different kinds and from different source laboratories.
In addition, the methodology should be adaptable for integration
of robotics to facilitate throughput where possible. ProCan has
instituted the use of pressure-cycling technology (barocyclers) to
achieve consistent lysis and digestion of tissue samples.[9] In the
establishment phase, we have made considerable effort to refine
protocols that achieve shorter processing times while improving
digestion and peptide yield, and maintaining reproducibility and
sample quality. For example, simultaneous Lys-C and tryptic di-
gestion is used in an accelerated manner as an aid to sample
throughput.[10]
1.3. Raw Data Acquisition
LC-MS/MS is a highly sensitive, mature technology capable
of identifying and providing relative quantitation of many
thousands of proteins in a tissue sample in a 1 to 2 h run
time. ProCan research is largely based on a variant of label-free
LC-MS/MS that is particularly well suited to comparative
analysis of a large number of tissue samples; Sequential
Window Acquisition of all THeoretical mass spectra/Data-
Independent Acquisition (SWATH/DIA).[11] DIA/SWATH gives
close to complete detection of peptides from tryptic digestion
of complex samples, and its applicability to comparative can-
cer tissue analysis has been demonstrated.[9] It does rely on
availability of relevant spectral reference libraries, and our
strategy is to use two-dimensions of chromatography from
pooled samples to make libraries representative of all samples
analyzed.
There are many practical considerations in the design and
operation of a biobank-scale LC-MS/MS facility. A key for Pro-
Can is to use standardized approaches to SWATH/DIA data
capture, and to give priority to time- and cost-saving modifi-
cations, including use of readily available reagents and com-
ponents such as commercially available capillary columns. In
addition, the entire system requires constant monitoring and
quality control (QC). ProCan has enabled continuous QC by
selected use of simple (bovine serum albumin) and complex
(HEK293 cells) digests coupled to automated monitoring sys-
tems and protocols. These external controls are in addition to
synthetic peptide standards that have been selected to span the
retention time of the experiment and are included in every
injection.

Proteomics 2019, 1900109 1900109 (3 of 5) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
LC-MS/MS instrumentation requires intensive maintenance
for optimal performance, especially in a multi-instrument envi-
ronment. Decisions such as the amount of sample injected (2 μg)
and the choice of a 90 min LC gradient, along with pre-emptive
maintenance help limit the downtime of the facility. A challenge
in ensuring this over the long-term is dependency on a highly
skilled scientific workforce who can prepare samples as well as
operate and maintain the equipment. A firm commitment to de-
velop this skill-base is critical to ensuring that proteomics can
make a sustainable contribution to clinical practice.
1.4. Single Study–Based Data Analysis
Proteomics faces the general ‘omics challenges of high dimen-
sionality and sparsity of data that comes from generating a large
number of measurements from a relatively small number of
samples. It is also subject to the universal truths that study
limitations can be overcome by the availability of independent
validation datasets of appropriate size, and that there is no
statistical correction that can truly compensate for the absence
of such data. To make progress through the early period of
collecting data resources, data scientists need to be alert to
potential confounders and choose between a wide range of
strategies for data pre-processing, including batch correction,
data normalization, handling missing data values, and selecting
informative features. Each approach has strengths and weak-
nesses, and while some guidance comes from experience with
other cancer ‘omic data, there are issues of particular relevance to
proteomics.
SWATH/DIA proteomics initially employs a peptide-centric
scoring strategy to generate a peptide matrix, followed by infer-
ence of proteins.[11] These two steps each involve complex statisti-
cal tests and stringent multiple-testing correction methods. Pep-
tide “roll-up” to proteins has the effect of substantially reducing
the size of the dataset, both by absorbing a number of peptides
into an individual protein assignation and by rejecting a volume
of peptide data that cannot be confidently assigned to a protein.
This becomes a form of a priori data filtering; however, it is cur-
rently unclear whether useful information is lost or false signal is
introduced during this process. The proteome is extremely com-
plex, with multiple proteoforms corresponding to a single gene
sequence. This complexity may be better reflected in the peptide
rather than a protein matrix, and it is possible that class discovery
analyses based on the more finely granular peptide-level data
will have advantages. To fully explore this, the performance
of both peptide and protein-level classifiers will need to be
tested.
A challenging feature of proteomic data is the high rate of
missing values. It is likely that a proportion of this is non-random
and related to threshold detection of low abundance peptides
or real differences in protein expression between tissue sample
types, reflecting potentially informative biological differences. A
number of strategies to manage missing values have been pro-
posed, including filtering out features with missing data and im-
puting missing data values. The development of classification
models that are robust to missing values is of particular interest,
and there is some evidence that these may have advantages.[12]
Proteomics 2019, 1900109 1900109 (4 of 5)
www.proteomics-journal.com
1.5. Multi-Study Data Analysis
Ultimately, ProCan’s goal is a highly curated collection of sample-
related data, proteomics, and other ‘omic data to support global
analyses. There remain significant open questions regarding the
synthesis of these disparate data sources into a single cohe-
sive analytics platform. We anticipate many challenges, ranging
from the essentially technical issue of merging spectral libraries
across cohorts, through complex normalization and batch correc-
tion, to the integrative analysis of multi-omic data with clinical
outcomes.
ProCan data will provide a valuable resource beyond their
primary use, once they become accessible to collaborators and
the wider scientific community. To facilitate this, the curated
data must sit within a FAIR framework,[13] that is, Findable by
both humans and machines, Accessible using standard proto-
cols, Interoperable with other systems and data resources, and
Reusable and reproducible via richly described metadata. This
multitude of data exists within a complex taxonomy that must
be contextualized with biological knowledge, either garnered
from existing public databases or from software applications.
Traditional relational databases, although widely adopted across
all disciplines, can limit exploration of highly connected data
such as these; however, graph-structured database technology—
industrialized in fields such as fraud detection, national security,
and social networks—has been shown to facilitate this in biolog-
ical research.[14]
An advantage of SWATH/DIA is that raw data files can form a
permanent digital map available for re-analysis through compu-
tational pipelines as analytical strategies improve, without need-
ing to re-run the physical samples. Data commons–like infras-
tructure enables this across tens or even hundreds of thousands
of samples, ensuring that the current “release” of processed data
reflects the state-of-the-art. Within the commons, data must be
classified for access at various levels from fully public to tightly
controlled, depending on the “trust level” of the user and the sen-
sitivity of the data. Importantly, such a system must adhere to
relevant jurisdictional privacy and ethics regulations.
For cancer clinicians and researchers to develop confi-
dence in high-throughput proteomics, individual studies must
demonstrate intrinsic robustness and consistency between
laboratories and over time. The international proteomics com-
munity has acknowledged this by promulgating a commitment
to transparency and quality in proteomics, chiefly through
sharing data, detailed documentation of study metadata, ref-
erence datasets, and reagents.[15] These efforts are supported
by national and international consortia such as the Clinical
Proteomic Tumor Analysis Consortium and the International
Cancer Proteogenomics Consortium (of which ProCan is a
member) that were given impetus in 2016 by the U.S. Cancer
Moonshot initiative’s focus on proteogenomics.[16] Standards
set by journal editors have an important role here, with a
recent guideline for submission of DIA-MS studies a useful
exemplar.[17] The contribution of large-scale projects such as
ProCan to this endeavor will include a commitment to develop
and test a system of standard operating procedures and critical
evaluation of quality, both internally and through peer review
mechanisms.

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
2. Concluding Comments
Sustained effort in cancer tissue–based proteomics research
is expected to provide substantial clinical benefit. Success will
require input from a broad range of stakeholders, including
cancer consumers, cancer biologists, oncologists, patholo-
gists, proteomicists, software engineers, data scientists, health
economists, regulators, and research funding agencies. Our
experience to date with ProCan indicates that a high level
of community support exists, and that clinical application of
high-throughput proteomics will be achievable.
Acknowledgements
ProCan is supported by the Australian Cancer Research Foundation, the
Cancer Institute New South Wales (NSW) (2017/TPG001, REG171150),
the NSW Ministry of Health (CMP-01), the University of Sydney, the
National Breast Cancer Foundation (IIRS-18-164), the Cancer Council
NSW (IG 18-01), Ian Potter Foundation, the Commonwealth of Australia
through the Medical Research Futures Fund (MRFF-PD), and the Na-
tional Health and Medical Research Council of Australia (GNT1047070,
GNT1170739).
Conflict of Interest
The authors declare no conflict of interest.
Keywords
cancer, data analysis, data-independent acquisition, proteomics, sequen-
tial window acquisition of all theoretical mass spectra–mass spectrometry
Received: April 30, 2019
Revised: July 11, 2019
Published online:
[1] S. Loibl, L. Gianni, Lancet 2017, 389, 2415.
[2] S. A. Cook, A. V. Tinker, BioDrugs 2019, 33, 255. PMID: 30895466.
Proteomics 2019, 1900109 1900109 (5 of 5)
www.proteomics-journal.com
[3] P. Janiaud, S. Serghiou, J. P. A. Ioannidis, Cancer Treat. Rev. 2019, 73,
20.
[4] J. Marquart, E. Y. Chen, V. Prasad, JAMA Oncol. 2018, 4, 1093.
[5] G. T. Gibney, L. M. Weiner, M. B. Atkins, Lancet Oncol. 2016, 17,
e542.
[6] C. Bais, B. Mueller, M. F. Brady, R. S. Mannel, R. A. Burger, W. Wei,
K. M. Marien, M. M. Kockx, A. Husain, M. J. Birrer, N. R. G. On-
cology/Gynecologic Oncology Group, J. Natl. Cancer Inst. 2017, 109,
djx066.
[7] O. J. Gustafsson, G. Arentz, P. Hoffmann, Biochim. Biophys. Acta
2015, 1854, 559.
[8] K. Yoshihara, M. Shahmoradgoli, E. Martinez, R. Vegesna, H. Kim,
W. Torres-Garcia, V. Trevino, H. Shen, P. W. Laird, D. A. Levine, S.
L. Carter, G. Getz, K. Stemke-Hale, G. B. Mills, R. G. Verhaak, Nat.
Commun. 2013, 4, 2612.
[9] T. Guo, P. Kouvonen, C. C. Koh, L. C. Gillet, W. E. Wolski, H. L. Rost,
G. Rosenberger, B. C. Collins, L. C. Blum, S. Gillessen, M. Joerger, W.
Jochum, R. Aebersold, Nat. Med. 2015, 21, 407.
[10] N. Lucas, A. B. Robinson, M. Marcker Espersen, S. Mahboob, D.
Xavier, J. Xue, R. L. Balleine, A. deFazio, P. G. Hains, P. J. Robinson,
J. Proteome. Res. 2019, 18, 399.
[11] C. Ludwig, L. Gillet, G. Rosenberger, S. Amon, B. C. Collins, R. Aeber-
sold, Mol. Sys. Biol. 2018, 14, e8126.
[12] M. Ali, S. A. Khan, K. Wennerberg, T. Aittokallio, Bioinformatics 2018,
34, 1353.
[13] M. D. Wilkinson, M. Dumontier, I. J. Aalbersberg, G. Appleton, M.
Axton, A. Baak, N. Blomberg, J. W. Boiten, L. B. da Silva Santos, P. E.
Bourne, J. Bouwman, A. J. Brookes, T. Clark, M. Crosas, I. Dillo, O.
Dumon, S. Edmunds, C. T. Evelo, R. Finkers, A. Gonzalez-Beltran, A.
J. Gray, P. Groth, C. Goble, J. S. Grethe, J. Heringa, P. A. t Hoen, R.
Hooft, T. Kuhn, R. Kok, J. Kok, et al., Sci. Data 2016, 3, 160018.
[14] B. H. Yoon, S. K. Kim, S. Y. Kim, Genom. Informat. 2017, 15, 19.
[15] C. R. Kinsinger, J. Apffel, M. Baker, X. Bian, C. H. Borchers, R. Brad-
shaw, M. Y. Brusniak, D. W. Chan, E. W. Deutsch, B. Domon, J. Gor-
man, R. Grimm, W. Hancock, H. Hermjakob, D. Horn, C. Hunter,
P. Kolar, H. J. Kraus, H. Langen, R. Linding, R. L. Moritz, G. S.
Omenn, R. Orlando, A. Pandey, P. Ping, A. Rahbar, R. Rivers, S. L.
Seymour, R. J. Simpson, D. Slotta, et al., Mol. Cell. Proteom. 2011, 10,
O111.015446.
[16] C. R. Jimenez, H. Zhang, C. R. Kinsinger, E. C. Nice, Clin. Proteom.
2018, 15, 4.
[17] R. J. Chalkley, M. J. MacCoss, J. D. Jaffe, H. L. Rost, Mol. Cell. Proteom.
2019, 18, 1.

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim