One of the biggest challenges in cancer is predicting its initiation and course of progression. In this issue of Cancer Research, Rockne and colleagues use state transition theory to predict how peripheral mononuclear blood cells in mice transition from a healthy state to acute myeloid leukemia. They found that critical transcriptomic perturbations could predict initiation and progression of the disease. This is an important step toward accurately predicting cancer evolution, which may eventually facilitate early diagnosis of cancer and disease recurrence, and which could potentially inform on timing of therapeutic interventions.

See related article by Rockne et al., 3157

Cancer is a highly complex and heterogeneous disease, with many different processes contributing to its development and progression, including mutations, copy number alterations, epigenetic states, and (post-) transcriptional regulatory mechanisms. These processes do not act in isolation, but rather in a complex network of interacting molecules and interactions between tumor, stroma, and immune cells. Many of these processes are also dynamic, and this can result in rewiring of this network of interplaying cells and molecules upon environmental changes or other triggers, such as cancer therapies. To correctly model cancer development, evolution, and its complex phenotypes, we therefore need to transition from analyzing single molecules and time points to using an integrative approach that takes into account the complexity and the dynamics of the tumor ecosystem (1).

Biological networks and signaling pathways are highly structured, allowing for a limited number of stable “states” of a system (2). Cancer can be thought of as one of such states—it is a robust, often irreversible state that can transition from the healthy state when the biological network that comprises the cell and its environment is rewired. Understanding the dynamics of the transitions that are critical for cancer development will help improve our understanding of the disease and its evolution. These dynamics can be modeled with state transition theory, which explains how, given certain conditions or triggers, one state of a system transforms into another (1). In biology, state transition theory has been traditionally applied to understand cellular differentiation (3). In cancer biology, it has been previously used to predict phenotype changes in epithelial-to-mesenchymal transition (4). State transitions have also been argued to play a role in therapeutic response to drugs, for example, immunotherapy (5). In this issue, Rockne and colleagues (6) apply state transition theory to dynamic data obtained from a leukemia mouse model to understand and predict cancer development and evolution.

Rockne and colleagues hypothesized that, in healthy cells, a large energy barrier exists that minimizes the chance to transition to a state of cancer. However, perturbations can lower this energy barrier, facilitating a transition to the cancer state. They postulated that, as a system will tend to its lowest energy state, a cancer cell can get locked in the cancer state if it remains untreated. This theory draws parallels to the well-known epigenetic landscape of Waddington (7), which has been widely used to illustrate the progressive restriction of cell differentiation. Here, cells are described as marbles that roll through a hilly landscape. Their trajectories end in local minima—differentiated states of the cell. The state transition theory used by Rockne and colleagues includes a similar landscape—a two-dimensional state-space with two minima, the states of health and cancer. As the authors show in their study, this energy landscape changes upon perturbed hematopoiesis, allowing for an easier transition of blood cells toward the leukemic state.

To construct the above-mentioned energy landscape, Rockne and colleagues used RNA-seq time-series data obtained from peripheral blood mononuclear cells. They derived these cells from a mouse model that represented a specific subclass of acute myeloid leukemia (AML) driven by fusion gene CBFB-MYH11 (CM). Expression of this fusion gene perturbs hematopoiesis and can result in the development of leukemia. The authors used a conditional knock-in mouse model so that they could control the starting point of this perturbation in each mouse. They sampled peripheral blood in a cohort of CM mice and littermate controls before inducing expression of the fusion gene, as well as after induction at monthly intervals. This yielded a rich dataset of in vivo transcriptomic dynamics. Rockne and colleagues then used a dimension reduction technique on these data to identify the component of the data that was associated with leukemia progression. Using this information, they could estimate the shape of the energy landscape for the induced mice and controls. These landscapes included three critical points—a reference stable state, an unstable transition state during which the cancer became overt, and a state of overt leukemia.

By analyzing the genes and pathways associated with these different states, the authors found that the early state was enriched for processes involved in cytokine signaling. The unstable transition state was enriched for DNA damage and repair response genes, while the leukemia state was enriched for mitosis-related pathways. Pathways that were persistently affected at all critical points were involved in cell proliferation and survival and included PI3K signaling. These processes are all known hallmarks of cancer (8), and the sequence in which these pathways are disrupted makes sense in the context of CM-induced leukemia. These results can potentially inform us on the timing of specific treatment approaches in leukemia. In future studies, it would be interesting to use a similar approach on data obtained from other cancer types or cancer subtypes to determine whether they are characterized by a specific sequence of events that disrupt the different hallmarks of cancer.

A major finding of the study by Rockne and colleagues was that, often several months before the detection of circulating leukemic blasts, the transcriptome had already moved toward the leukemic state. This highlights one of the strengths of using state transition theory to predict cancer development, as it means that it can potentially be used for early diagnosis of cancer. One of the most striking advantages of the approach taken by the authors is that it allowed for alignment of transcriptomic signals that are not synchronized over time. In fact, genes that had the strongest contributions to the leukemic trajectory had remarkably consistent dynamics in the state-space, even though there was high variability in expression levels between the different mice when comparing individual time points. This is a very important result. While some of the mice developed overt leukemia already at 3 months after induction, for other mice it took several months more. It is difficult to take these differences into account when analyzing cancer data. Often, instead of attempting to correct for these differences, groups of patients are compared on the basis of fixed (event-free) survival thresholds, or samples are grouped based on specific phenotypic variables. In humans, there is even more variability in disease onset after an initial triggering event than in mouse models, due to variability in genetics, environment, and other factors. Being able to correct for this variability by directly modeling disease trajectories would be tremendously helpful in moving toward personalized predictions of cancer evolution.

Predicting the timing of cancer onset is challenging. Rockne and colleagues used differential equations—a mathematical approach that is particularly useful in modeling the behavior of complex systems—to estimate the motion of the transcriptome across the above-mentioned energy landscape. Importantly, instead of studying the trajectories of single molecules across this landscape, Rockne and colleagues focused on the transcriptome as a whole, taking a systems-view on cancer development. This allowed for predicting the timing of the unstable transition state, which was found to be associated with an accelerated increase in leukemic blasts. Their predictions could also discriminate between early and delayed onset of leukemia. While this a step toward our understanding of leukemia development with a potential impact on early diagnosis of cancer, it will likely remain a challenge to predict the exact onset of disease once a cancer is in, or close to, the unstable state, as there will likely be high variability in the genotypes and environment of individual patients with cancer. Including more information on this variability will likely improve these models. A potential future extension would be to include other 'omics data types, as well as information on single-nucleotide polymorphisms, known risk factors, and phenotypic variables of each individual. Also, as it became recently possible to profile transcriptomics and other types of 'omic data at the single-cell resolution, one may envision expanding this approach to single-cell measures in the future (9).

How do we place these results in the context of the increasing number of 'omics studies in cancer? Recently, hospitals and cancer centers have started to sequence targeted mutation panels, and sometimes whole-exome sequencing, for new incoming cancer samples (10). The study by Rockne and colleagues provides evidence for whole transcriptomic perturbations to play a role in shifting a healthy cell toward a cancer state. This means that collecting transcriptomic profiles for new incoming cancer samples will likely help fine-tune estimates of how far the cancer has progressed and help predict when the cancer is likely to transition to a more aggressive state or when it will metastasize. In addition, once we have larger datasets and clear knowledge on what treatment strategies are effective in each of the different states, it could help suggest treatment strategies that are adapted to the individual patient with cancer at a particular point in time during the tumor's evolution. Finally, profiling transcriptomic data through screening of risk groups might help to detect cancer at an early stage. For this to materialize, large numbers of tumors would probably need to be profiled, ideally at several time points before and during the patient's disease course. We envision this will be the standard in the future, with an integration of genomic data and the use of computational tools in the clinic.

No potential conflicts of interest were disclosed.

This work was supported by grants from the Norwegian Research Council, Helse Sør-Øst, and University of Oslo through the Centre for Molecular Medicine Norway (NCMM). The author thanks Alessandro Marin for helpful comments.

1.
Clarke
MA
,
Fisher
J
. 
Executable cancer models: successes and challenges
.
Nat Rev Cancer
2020
;
20
:
343
54
.
2.
Kolch
W
,
Halasz
M
,
Granovskaya
M
,
Kholodenko
BN
. 
The dynamic control of signal transduction networks in cancer cells
.
Nat Rev Cancer
2015
;
15
:
515
27
.
3.
Moris
N
,
Pina
C
,
Arias
AM
. 
Transition states and cell fate decisions in epigenetic landscapes
.
Nat Rev Genet
2016
;
17
:
693
703
.
4.
Pastushenko
I
,
Brisebarre
A
,
Sifrim
A
,
Fioramonti
M
,
Revenco
T
,
Boumahdi
S
, et al
Identification of the tumour transition states occurring during EMT
.
Nature
2018
;
556
:
463
8
.
5.
Lesterhuis
WJ
,
Bosco
A
,
Millward
MJ
,
Small
M
,
Nowak
AK
,
Lake
RA
. 
Dynamic versus static biomarkers in cancer immune checkpoint blockade: unravelling complexity
.
Nat Rev Drug Discov
2017
;
16
:
264
72
.
6.
Rockne
R
,
Branciamore
S
,
Qi
J
,
Frankhouser
DE
,
O'Meally
D
,
Hua
W-K
, et al
State-transition analysis of time-sequential gene expression identifies critical points that predict development of acute myeloid leukemia
.
Cancer Res
2020
;
80
:
3157
69
.
7.
Waddington
CH
.
The strategy of the genes; a discussion of some aspects of theoretical biology
.
London, United Kingdom
:
Allen & Unwin
; 
1957
.
8.
Hanahan
D
,
Weinberg
RA
. 
Hallmarks of cancer: the next generation
.
Cell
2011
;
144
:
646
74
.
9.
Rozenblatt-Rosen
O
,
Regev
A
,
Oberdoerffer
P
,
Nawy
T
,
Hupalowska
A
,
Rood
JE
, et al
The Human Tumor Atlas Network: charting tumor transitions across space and time at single-cell resolution
.
Cell
2020
;
181
:
236
49
.
10.
AACR Project GENIE Consortium
. 
AACR Project GENIE: powering precision medicine through an international consortium
.
Cancer Discov
2017
;
7
:
818
31
.