Figure 1: Approaches for Quality
Assurance in Biopharmaceutical Production (A) Quality by testing
approach – after process steps samples are drawn to determine product
quality and quantity in the laboratory. As this is a retrospective
approach, it does not allow any process control and leads to failure
prone production even though it is costly and time consuming. (B)
Quality by Design approach – the process is monitored in real-time and
therefore product quality and quantity is known at each stage of the
process. If the process is performed within the set specifications, the
product will finally have the desired quality. Real-time monitoring of
the process allows process control and therefore enhances product
quality and reduces time and material consumption for offline analysis.
Then the subsequent quality by design initiative’s main aims (Q8-Q11
quality guidelines of the International Council for Harmonization of
Technical Requirements for Pharmaceuticals for Human Use (ICH) (The
International Council for Harmonization of Technical Requirements for
Pharmaceuticals for Human Use (ICH),) should “ensure that all sources
of variability affecting a process are identified, explained, and
managed by appropriate measures. This enables the finished medicine to
consistently meet its predefined characteristics from the start”. This
considers the raw material variations and all variations of the process
parameters. Furthermore, recently a guidance has been released to
recommend continuous integrated biomanufacturing. This guidance also
recommends an in depth process understanding but does not recommend
process control. The guideline of real-time release has been developed
because “under specific circumstances an appropriate combination of
process controls (critical process parameters) together with pre-defined
material attributes may provide greater assurance of product quality
than end product testing and the context as such be an integral part of
the control strategy principle”. Furthermore, “RTRT may be applied
during the stages of manufacture of chemical and biological products
resulting in the elimination of all, or certain, specific tests in the
specifications of the finished active substance or finished medicinal
product”. The ultimate goals must be to monitor and control the process
in real-time so that we are able to efficiently control the variability
of raw materials, variability of the process parameters and in the
inherent variability of the biological system(Wasalathanthri et al.,
2020). If we want to control the process, we need real-time information
on quantity, purity and potency, - the critical quality attributes (CQA)
- because only then can we control the critical process parameters. This
does not differ from current approaches, but everything will be
real-time instead of retrospective. It is obvious that we do not have
dedicated sensors for quantity, purity and potency. We must use indirect
methods, but there is a reluctance in industry and academia to apply
soft sensors. A soft sensor (Luttmann et al., 2012; Mandenius et al.,
2015; Roch et al., 2016), also known as a virtual sensor, is not a real
existing sensor, but a simulation of interdependencies between
representative measured variables and a target variable such as product
concentration, aggregate content, dsDNA content, or a biological
activity of a biopharmaceutical for instance. Then the soft sensors can
be used to control a system(Ender et al., 2003). Especially in
downstream processing when executed in batch we have limited possibility
to really control the process. Therefore, a further incentive for
continuous integrated manufacturing beyond the economic gain and
improvement of the environmental footprint is the ability to fully
control and automate a process and to arrive at system runs
autonomously. Such a system would also contribute to a digital twin of a
manufacturing plant, but such a twin is not required for an automated
process. Process control is already well established in upstream
processing but not in downstream processing. Control of oxygen and pH
has been already done in the very early days of production of
biopharmaceuticals in bioreactors. In an advanced bioreactor control
systems air, O2, CO2, in the off gas and
dissolved oxygen and temperature are measured in-line together with NIR,
Raman, turbidity and capacitance to infer nutrient concentration and
waste products or on-line sampling is conducted measured it directly by
at-line enzymatic assays or HPLC (Zhao et al., 2015).
In advanced downstream processing, chromatographic workstations are
equipped with UV 280 flow cell, pH, conductivity probes (Carta et al.,
2010). Advanced sensors such as NIR, AT -FTIR, Raman multiangle light
scattering, and dynamic light scattering are not routinely used as
sensors for downstream process monitoring, although they would provide
very useful information that goes far beyond the standard information.
The signal from such sensors yields a spectrum that must be further
interpreted using a so-called deconvolution process, for example, to
assign a particular wavelength to a molecular property. These
chemometric approaches were developed simultaneously with the
development of the sensor hardware. Recently, it was found that the peak
profile itself could be used to correlate it with certain properties of
the process fluid or biomolecule, such as the glycosylation pattern of
antibodies, aggregate content, or truncated variants. In such a case,
the entire spectrum is used instead of a specific wavelength, and if the
spectral information is trained with off-line data, it can be considered
a soft sensor(Feidl et al., 2019; Kornecki et al., 2018).
Decades ago, bioprocessing originally started with fixed time/volume
processes and even processes are performed without a single sensor. The
next stage is threshold-based processing based on either a single sensor
or multiple sensors. More sophisticated methods for using sensor data
include chemometric methods, advanced statistical methods, and hybrid
models (Figure 2). Hybrid models and model predictive control are the
culmination of a monitoring and control system that ultimately enables
real-time release.
Fixed Time/Volume