3.3 Single Cell Metabolomics
Single cell metabolomics (SCM) is designed to evaluate the metabolic
profiles of cells with single cell resolution. Like many other single
cell omics approaches, SCM provides a means of assessing heterogeneity
between single cells.58 Majority of SCM pipelines rely
on mass spectrometry-based methods to quantify the metabolomic profiles.
As with all single cell methodologies, the first step is to isolate
single cells. This can be accomplished while keeping the morphology
intact using techniques such as FACS or microfluidic arrays, or by
atomic force microscopy (AFM) which completely isolates the single
cell’s metabolites in a probe. The next logical step is to quench all
metabolic activity within the single cells. This is accomplished with
the use of organic acids or solvents to denature enzymes and impede
further conversion of metabolites, or by a technique known as snap
freezing using liquid nitrogen to derail metabolic activity and promote
membrane lysis. Snap freezing limits the use of reagents that could be
considered contaminants in the downstream MS analysis but requires
further workup to obtain pure metabolic profiles. The use of organic
solvents in the quenching phase is beneficial because it also aids in
metabolite extraction from the lysate. Depending on the type of MS used,
different combinations of organic solvents are utilized. Guo et al.
provide an excellent review on the appropriate use of solvent mixtures
for each type of MS. After quenching metabolic activity and extracting
the metabolites, samples are ionized. Ionization can be broken down into
two distinct mechanisms, vacuum based or ambient methods. The review by
Liu et al. provides an excellent summary of these sub variations
including applications for each technique and data
preprocessing/analysis strategies for the obtained MS
spectra.59
M1 macrophages undergo metabolic reprogramming from oxidative
phosphorylation to glycolysis upon phenotypical
differentiation.60 Sustained M1 macrophage activation
leads to sustained inflammation and can cause tissue damage if not
appropriately regulated. The M2 phenotype is acquired after polarization
with IL-4 and macrophage colony stimulating factor (M-CSF). M2
macrophages mediate the inflammatory response by secreting
anti-inflammatory cytokines such as IL-10 and TGFβ. This gives them an
important role in tissue repair after the host inflammatory response.
Unlike the M1 phenotype, M2 macrophages rely upon the citric acid cycle
(TCA) to support their production of ATP through oxidative
phosphorylation. Phenotypical assays to assess macrophage population
heterogeneity rely upon detection of cytokines or membrane bound surface
markers. However, cytokines and surface markers are expressed in both
phenotypes, thus presenting the need to develop a more specific
phenotypical classification assay. To reliably quantify phenotypical
differences, metabolic profiling with an LC-MS approach can be used,
specifically, time of flight SIMS MS (TOF-SIMS) coupled with an Orbitrap
analyzer (3D OrbiSIMS) which in contradiction to ordinary SIMS or
TOF-SIMS, allows for MS/MS of the metabolite. TOF-SIMS provides highly
localized spatial resolution of cell surfaces and does not require
extensive sample preparation compared to most other LC-MS techniques.
Historically, TOF-SIMS approach has not been reliably used to document
endogenous metabolic profiles due to the poor mass resolving power.
However, when coupled with the high mass resolving power and mass
accuracy of an Orbitrap, characterizing metabolic profiles at the single
cell level is possible. Using a targeted approach to analyze the lipid
palette (matching lipid ion peaks to LIPID MAPS database), it was found
that M1 macrophages had the highest lipid counts and different lipid
composition compared to M2 or M0.38 The amino acid
composition and other metabolites showed notable differences as well.
Overall, the study represents a novel approach for in-situcharacterization of metabolic profiles to assess phenotypical
differences between closely related cell types with single cell
resolution.
HIGH THROUGHPUT IMAGING:
High throughput imaging (HTI) represents a robust set of methodologies
which provide information on cellular morphology and includes large
scale automated sample preparation and image
analysis.61 All HTI workflows are based upon a
targeted approach, where a dye, fluorescent reagent, fluorophore
conjugated antibodies/oligonucleotides, or genetic construct which
expresses fluorescent protein are used to label a particular component
of the cell. This can include proteins, nucleic acids, or specific
organelles within the cell. The workflow begins with perturbing cells
from steady state, by the addition of ligands to activate cellular
signaling pathways and thus inducing differential gene expression or by
utilizing short hairpin RNA (shRNA) and clustered regularly interspaced
short palindromic repeats (CRISPR/Cas9) which can respectively knockdown
and knockout genes in a targeted manner, giving scientists the ability
to diagnose the role of the labeled target in response to cellular
perturbation. The ligand used to disrupt cellular steady state dictates
the cellular pathways involved in regulating the change in phenotype for
the target of interest.
Traditionally, drug discovery methodologies are target based, they rely
on pre-cognition of a target molecule (interacting protein, or specific
receptor), and assess the structural dynamics of “hit/lead” compounds
that can bind the target and modulate its effector
function.62 This approach typically involves screening
the target against a library of compounds predicted to have high binding
affinity potential based on knowledge of the binding pocket and or
molecular docking simulations and assessing the mechanism of action for
each compound in further downstream analysis
experiments.63 The structure-based drug discovery is
rapidly developing as artificial intelligence and deep learning
algorithms progress.64 It should be noted that there
is inherent bias with target-based screening methodologies, such that
hit compounds designed for an individual target may have off-target
effects (polypharmacology). To resolve this bias, phenotypic based drug
discovery (PDD) methodologies start by looking at a cellular phenotype
and attempt to correlate hit compounds that can modulate the phenotypic
readout. Lin et al. provide a review that encompasses the details of
workflow design and data analysis for each of these screening
methodologies in greater detail.
HTI can also be used to profile compounds or libraries in a
multiparameter fashion based on statistical analysis and clustering of
compounds that elicit a similar phenotypic readout.65For example, a multiplexed image-based assay termed “Cell Painting”
can measure ~1,500 morphological features pertaining to
different combinations of size, shape, texture, staining intensity, and
so on, in response to multiple perturbations.66 These
perturbations can be chemically induced or could be a result of genetic
manipulation (knockdown/knockout). The technique is capable of sensing
subtle phenotypical differences and it groups perturbing agents
(compounds/genes) based on similar pathways that they affect, thus
shedding light on the markers of disease. The basic workflow involves
genetic or chemical perturbation to a target cell line, staining cells
with fluorescent dyes that label 8 different compartments within the
cell (nucleus, F-Actin, plasma membrane, mitochondria, etc.), microscopy
imaging, and image analysis where the morphological features taken from
the image correlate to a profile that reflects the phenotypic state of
the cell. Comparing profiles taken at different time points or with
different perturbing agents can elucidate the mechanism of action and
cellular signaling pathways pertaining to a particular phenotype.
Lastly, HTI methodologies can be classified into a third category, used
for deep imaging which combines the use of fully automated
high-resolution microscopy and sophisticated computational analysis of
many images.67 Due to the massive amount of cellular
input, this technique can assess rare cellular phenotypes that may only
be present at low probabilities.68 This niche
application of HTI is relatively unexplored, perhaps due to the limited
availability of cost prohibitive high throughput imaging systems.
However, even with the limited number of studies utilizing the deep
imaging approach, the potential of HTI to identify rare cellular
phenotypes in response to a small subset of perturbing agents represents
an important area of study in systems biology, that is, phenotypical
profiling of biomarkers of disease pathogenesis for rare and
understudied host-pathogen interactions.
Nuclear factor kappa-light-chain enhancer of activated B cells (NF-κB)
is a family of transcription factors that regulate innate and adaptive
immune responses, cellular differentiation, proliferation, and
apoptosis.69 The mammalian NF-κB family is dimeric in
nature and includes five different protein monomers (p65/RelA, RelB,
cRel, p50/105, and p52/100) that form either homo or heterodimers and
each dimer differentially binds to DNA. All the monomers have a
conserved N-terminal domain known as the Rel homology domain (RHD),
which is essential for DNA binding, dimerization, nuclear localization,
and inhibitor binding.70 NF-κB p105 and p100 proteins
contain a IκB inhibitory domain which contains, multiple copies of the
ankyrin repeat (ANK) at the C-terminus. Both NF-κB subfamily proteins
(p105 and p100) undergo proteasome-dependent partial proteolysis to
their active DNA binding forms (p50 and p52,
respectively).71 NF-κB dimers reside in the cytoplasm
bound to inhibitor proteins of the IκB family, where upon degradation of
the inhibitor (phosphorylation of IκB-by-IκB kinase (IKK) followed by
ubiquitylation and proteasomal degradation), NF-κB translocated into the
nucleus where it binds DNA and stimulates transcription of target genes.
Importantly, one of the target genes is the inhibitor itself which
provides a mode of NF-κB signaling regulation via a negative feedback
loop. With constant stimulation, the degradation of the inhibitor as
well as NF-κB re-synthesis leads to oscillations of NF-κB nuclear
translocation.72 Oscillations in NF-κB translocation
are signal dependent, for example, sustained stimulation of cells with
tumor necrosis factor a (TNFα) leads to oscillations, whereas a short
one-time stimulation with TNFα leads to only one sharp peak of NF-κB
translocation/activation.73 Induction of cells with
LPS leads to an entirely different response, where NF-κB has been found
to translocate in either one cycle of translocation, persistent
translocation, or oscillations in the patterns of
translocation.74 Since the translocation of NF-κB has
been found to be stimulus-dependent, it is important to consider the
kinetic and spatio-temporal landscape of NF-κB when studying pertinent
signaling dynamics. These studies highlight the importance of high
throughput imaging techniques for studying cellular signaling dynamics
with real time quantification and showcase how HTI workflows can be
utilized to elucidate the mechanisms underlying host-pathogen
interactions.
COMPUTATIONAL SIMULATION AND MODELING:
Computational biology involves the assessment of complex biological
systems through the development of computational models and simulations
which can be used to develop predictive models of the factors involved
in disease pathogenesis. The field is rapidly progressing with
developments in computer hardware, software, and experimental methods,
lowering the computational efforts required to produce these models.
Computational models can be separated into 2 sub groups, quantitative
and logical models. A quantitative model utilizes sets of differential
equations to define the dynamics of the model which are typically
non-linear. It requires pre-defined knowledge of details regarding the
pathway or cellular event under study and is thus limited to modeling
small portions of a well classified pathway. A logical model is based
upon a Boolean system and qualitatively defines the dynamics of the
model. It does not require a pre-defined knowledge of the system to be
analyzed, and thus can be applied to large scale systems.
A sub field of computational biology utilizes both modeling approaches
and resides at the intersection of systems biology and traditional
bioinformatics, known as systems bioinformatics.75 The
field of systems bioinformatics can be defined as the framework for
integrating the multiomics landscape traditionally used in systems
biology approaches, to provide insight into each individual omics layer
and the cumulative interactions between them. In this way, systems
bioinformatics provides methodologies capable of assessing the
biological mechanisms of the entire interwoven system rather than the
summation of each individual component or omics layer. The generation of
this field is based around systems theory which is holistic in nature,
and it’s use in systems bioinformatics is dependent on graph theory,
network science, and other mathematical approaches which facilitate the
analysis of complex networks derived from the system of interest.
Mathematical models are fundamental for analyzing network topology and
kinetics. As multiomics based quantification methods advance in both
high throughput ability and sensitivity, more accurate parameters can be
fed into models and provide more accurately quantifiable simulations of
signaling dynamics. Networks can also be applied to qualitative models
of pathway modeling. Many open access platforms are available for that
purpose. These software work by taking an input list of gene symbols or
protein names and assessing their gene ontology (GO) to map them. For
example, Cytoscape facilitates the visualization of complex biological
networks with annotated gene symbols and expression
data.76 Reactome enhanced pathway visualization is
another peer reviewed alternative.77 There are several
pathway databases which allow for the visualization of signaling
components based upon GO terms. A review comparing some of the most
widely used databases can be found here.78 Another
detailed review on construction and analysis of biological pathways can
be found here.79
Construction of networks to showcase biological pathways can incorporate
both mathematical modeling functions and qualitative visualization to
help researchers fully understand the molecular dynamics involved in
cell signaling. One such software is Simmune, that generates
computational models incorporating the spatially resolved
reaction-diffusion networks. Simmune utilizes rule-based approaches to
lower computational complexity of simulations. This involves
pre-programming of the simulation with fundamental signaling components
(important proteins) and their pair-wise interactions which allows the
computer to assemble the complexes that constitute the signaling
network.80 This rule-based approach was incorporated
as a response to one of the most traditional challenges in pathway
modeling- combinatorial explosion. Combinatorial explosion can occur
when there is excessively high number of alternative interactions
arising from a network consisting of many different signaling
components, or individual components that have multiple binding sites
and thus many possible interactions. Simmune works by generating a local
network in a multi-step fashion. The first step involves the
construction of a non-spatial network that includes every possible
molecular interaction for each fundamental signaling component. This
‘template network’ is then adjusted to reflect the local molecular
environment which lowers computational extensivity of the simulation.
Simmune is also able to account for morphologically dynamic models which
usually requires rebuilding the network every time the cellular
morphology changes, to account for spatial constraints concerning the
receptor ligand interactions during membrane fluctuation.In-silico approaches for assessing pathway dynamics represent an
important stepping stone in systems bioinformatics and all related
disciplines to either validate or predict experimental findings. In a
study put forth by Manes et al., the chemo sensing pathway
Sphingosine-1-phosphate (S1P) was explored with a combinatorial approach
of RNA sequencing, targeted proteomics, and Simmune based modeling for
computational validation.81,82 This highlights the
importance of in silico based computational models in providing
insights into molecular mechanics of complex signaling networks and
validating the experimental findings.
Protein folding simulations such as AlphaFold2 are also an informative
tool for predicting and assessing protein structure and can be used to
diagnose the protein’s effector function.83 AlphaFold2
can construct a 3-dimesnional representation of how a protein will fold,
based upon its primary sequence. The software uses a deep learning-based
algorithm with multi-sequence alignment which incorporates both physical
and biological knowledge regarding protein
structure.84 In addition to the software’s abilities,
the AlphaFold team has now released accurate structure predictions for
human proteome in a freely available database.85 The
availability of a database with highly accurate protein structures that
are continuously updated is a major step forward for the field of
structural biology as it takes away the burden of generating these
structural models from scratch.86 As AlphaFold2
evolves along with the database, we can expect to see more structural
predictions that are publicly available and provide researchers with
tools that can be exploited for drug discovery, investigating
heteromeric protein-protein interactions, and creating simulations of
pathway dynamics where each component of the signaling pathway includes
a highly accurate 3-dimensional model of its native conformational
shape.87 This will greatly benefit the field of
systems biology as better structural predictions of the human proteome
can help researchers assess all the possible functions of a protein and
build more complex models regarding their kinetics.
INTEGRATING OMICS APPROACHES: