Identifying and Measuring Bias
There are simple in vitro techniques that can be used to detect
and quantify bias but the most reliable is to simply express two
signaling responses for an agonist for the same receptor as functions of
each other; this is a ‘bias’ plot and clearly indicates bias without
models or equations. For example, Fig 1 shows the obviously different
bias of four Κ-opioid agonists for G-protein and β-arrestin response;
the data indicate that these agonists stabilize two distinctly different
active receptor conformations, one preferring G proteins and one
preferring β-arrestin (White et al, 2014). For exploitation of biased
signaling in discovery programs, models have been derived to put numbers
on the magnitude of bias to allow medicinal chemists to systematically
modify chemical scaffolds (Onaran et al, 2017; Kenakin, 2019; Kolb et
al, 2022). Basically these procedures identify exemplar agonists as
being special, the assumption being that they stabilize a unique active
receptor state conformation. When these effects are seen, there are
important criteria that must be met before biased signaling is
considered seriously.
- Lack of response does not necessarily signify bias but rather only
that the agonist efficacy is insufficient to demonstrate where along
the concentration axis of a dose-response (DR) curve the response
occurs in a particular assay. Therefore, a measurable CR curve
must be observed to accurately determine bias.
- If no response in a particular pathway is observed for a molecule
(appearance of ‘perfect’ bias) it should be tested as an antagonist of
other molecules that do activate the pathway to assess the interaction
with the receptor and determine affinity.
- Bias has no translatable meaning in itself but only as a ratio to
another agonist, i.e. bias always refers to a comparison to another
agonist, usually the natural agonist.
- Potency ratios must be used as this cancels measurement bias caused by
differences in the sensitivities of different functional assays, i.e.
second messenger assays are more sensitive than BRET assays. In this
regard, the separation between two pathway DR curves does not
designate the relative concentrations producing effect in vivo ;
this occurs over the same concentration range but just with different
magnitudes.
Finally, it should be recognized that ‘bias’ does not designate whether
or not selective agonism will be seen; this is determined by the
intrinsic efficacy of the agonist. However, when the system is sensitive
enough and/or when the agonist has sufficient efficacy, bias determines
the relative strength of signal for a given pathway when agonism is
observed.
It should be recognized that assays identifying biased signaling,
whether it be therapeutically applicable or not, identify an important
agonist species, i.e. an agonist that produces a unique active state
receptor of the receptor. This is the important deliverable of anin vitro bias assay but the selected pathways used to
detect the receptor conformations may not be the relevant pathways for
therapeutic bias. In essence, the in vitro detection bias
measurements can be considered a method of detecting a harbinger of
complex signaling heterogeneity to come that eventually may culminate in
therapeutically beneficial bias. The main utility of bias scores is that
they enable classification of new agonists for further testing in
vivo and once bias is identified at the level of the receptor, it can
be considered a rogue property and a unique uncontrolled signal
initiator that then imparts a unique activation pattern on cells.
In terms of thermodynamics, the complete description of a ligand
activity on a receptor must include both affinity and efficacy and the
operational model index of
Log(τ/KA) fulfils
that criterion (Kenakin et al, 2012). However, while bias can arise
through differences in either affinity or efficacy (or both), the most
robust differences come from differences in efficacy (Rajagopal et al,
2011). This is in keeping with the fact that agonism is much more
resistant to changes in tissue sensitivity when it is based on high
efficacy as opposed to high affinity (Kenakin, 1984).
From it’s inception, bias has been considered a binary readout, i.e. if
cyclic AMP and β-arrestin assays are used to discern signaling,
compounds would fall into one of two groups (or perhaps a third that did
not discern pathways). However, these are crude indicators of a more
sophisticated phenomenon, namely the stabilization of a unique receptor
active state and this active state could go on to selectively interact
with a number of other proteins in the cell to show further
differentiations. Therefore, from the first binary test of agonism in
two pathways, a further series of experiments could be done to
delineate finer texture in the signaling produced by the molecule.
Currently, sophisticated readouts of receptor activation of signaling
components are being used to detect and quantify bias in ways that go
beyond a binary readout. For example, measuring biased signaling through
monitoring G protein subunits allows estimation of imbalances in
neurotransmission (Park et al, 2023). New assays have been developed to
determine micro-interactions of receptors with the complete complement
of G proteins in a cell (Olsen et al, 2020). The application of
pathway-selective bioluminescence resonance energy transfer (BRET)
biosensors that monitor the engagement and activation of signaling
effectors downstream of G proteins (including phospholipase C (PLC),
p63RhoGEF, protein kinase C (PKC), and Rho), allow clustering of
compounds into different subfamilies of biased ligands. These effects,
in turn, show bias among G protein subtypes, in terms of not only
functional selectivity between Gαq and β-arrestin but, interestingly,
bias through the engagement of different G proteins to activate a common
effector. Through a suite of BRET biosensors, a fingerprint of
angiotensin receptor agonism has the ability to provide a view of
signaling at various levels of GPCR activation – see Fig 2 (Namkung et
al, 2018). Nanoluciferase-based complementation assays also have been
used to provide textured readouts of biased signaling (Laschet et al,
2019) and genetically-encoded
fluorescent biosensors have been employed to illuminate spatiotemporal
biased signaling (Kayser et al, 2023).