As discussed above and in other chapters in this text, high-throughput screening (HTS) is usually performed with only one or two low-resolution techniques of the type previously considered. The selection of the particular technique is typically based on what is known about degradation pathways of the target, convenience, and availability of appropriate instrumentation as well as speed. It has recently become possible to combine the results from multiple techniques with the goal of providing a more comprehensive picture of a protein’s structure and its response to various environmental perturbations. This can be used to select optimal methods for HTS as well as for various forms of comparative analysis. A number of methods are available for this purpose. We will consider here only the one that has been most thoroughly described in the literature [reviewed in Maddux et al. (2011)], but other approaches such as Chernoff faces and star charts (Yau 2011) in which information is encoded in facial features or abstract geometric shapes are under consideration. The former method is known as the EPD. The word empirical is inserted in front of the phrase “phase diagram” to differentiate it from the well-known thermodynamic or equilibrium phase diagram since equilibrium conditions are not implied in the former.

The basic idea behind the EPD is to represent the protein (application has also been made to peptides, nucleic acids, virus-like particles, viruses, and bacteria cells) as a vector in which the components of the vector are experimental values obtained from the various methods employed as a function of solution variables. Preparation of EPDs generally involves buffer subtraction of the data, peak ­selection for data analysis (entire spectra can also be used), averaging of multiple data acquisitions, normalization, input matrix synthesis, singular value decomposition, and finally color mapping of the most significant data using an RGB color scheme. A detailed description of the method including the mathematics involved is ­presented in Maddux et al. (2011). In general, far-UV circular dichroism (CD) is used to monitor secondary structure although FTIR and Raman spectroscopies can also be employed. Tertiary structure is most commonly analyzed with intrinsic fluorescence, near-UV CD, or high-resolution derivative absorption spectroscopy. Dye binding using compounds such as 8-anilino naphthalene sulfonic acid (ANS) is frequently used to probe the exposure of apolar regions in the protein. Dissociation and association (including aggregation) are typically probed with static and/or DLS (although see below). Overall thermal stability is often studied with differential scanning calorimetry. The most common independent variables (i.e., forms of stress that have previously been employed) are temperature and pH although a wide variety of other variables have been used as described below. Perhaps the major limitation of the EPD approach has been the time and instrumentation necessary to prepare an EPD. This has recently changed with the advent of equipment capable of performing multiple different types of measurements simultaneously. Originally an EPD typically required a fluorometer, CD spectropolarimeter, light scattering system, and perhaps a DSC or FTIR spectrometer. Several newly ­developed instruments now at least partially overcome this limitation. For example, recent improvements in CD instruments now permit both near- and far-UV ­spectra, near-UV absorption spectra, fluorescence, and SLS (turbidity or scattering at the fluorescence emission wavelength) to all be acquired simultaneously in a four-position sample chamber under variable temperature conditions (Hu et al. 2011). This permits EPDs to be generated in less than a day. A similar “protein machine” with a six-position sample chamber has also been recently described (Maddux et al. 2012). Perhaps the simplest version of a system with rapid EPD generation capability is a UV absorption spectrometer, typically of the diode array variety, to permit sufficient resolution of derivative peaks (Kueltzo et al. 2003). At high resolution (usually second), the derivative spectrum of a protein will usually manifest distinct peaks for phenylalanine, tyrosine, and tryptophan (if present). Since the residues are usually buried, Tyr is often interfacial, and Trp present in highly variable environments, temperature, and pH-­induced peaks shifts can frequently provide a fairly detailed picture of a protein’s structural response to various perturbations. Such data can easily be represented in the form of an EPD (Kueltzo et al. 2003). Similarly, fluorescence microtiter plate-­based fluorometers have been developed which employ multiple fluorescence and light scattering measurements as a function of temperature, as also described in the fluorescence section. The latter is of especially high throughput, permitting the generation of many EPDs in a single day. EPDs for four model proteins obtained from multiple instruments, two CD-based spectrometers, and a high-throughput fluorometer are shown in Fig. 2.3, where it can be seen that all produce similar EPDs although small differences are apparent due to the various types of measurements used to construct each EPD.

The EPD method provides a comprehensive overview of how a protein responds to environmental alteration in the form of a colored diagram in which regions of different color correspond to different structural states of the target molecules. By reference to the original data, native partially folded and molten globule, ­extensively unfolded, dissociated, oligomerized, and various aggregated states can all be identified. This provides the scientist with clues to trouble spots in a protein and a basis with which to select assays with which to screen for potential stabilizers. For example, if aggregation or a particular structural change occurs under moderate temperature and/or pH conditions, one can select a less stable condition and one or more techniques sensitive to selected degradation events for screening purposes. Typically, a supplemental GRAS (generally regarded as safe) library containing a selection of buffers, sugars, sugar alcohols, amino acids, polymers, detergents, and osmolytes is used. In the initial screen relatively high concentrations of compounds are used with their concentration dependence and use in combination later optimized. It is usually wise to employ at least two methods: one sensitive to aggregation (e.g., light scattering) and one to structural change (e.g., fluorescence, CD) for this purpose. DSC is also commonly employed especially due to the recent availability of highly sensitive high-throughput instruments. It is also possible to prepare EPDs in the presence of selected stabilizers to permit a more detailed analysis/comparison of their effects on a protein. The information thus obtained by a temperature/pH EPD thus provides a basis for buffer and excipient selection at an early stage of pharmaceutical development. Although not yet published, a new version of the EPD has been developed in which the colors have actual physical measuring in contrast to the arbitrary assignment of color in the original EPD.

High-throughput methods and EPDs can also be applied to a wide variety of different situations, some of which will be briefly described here. Two commonly encountered forms of stress in the protein therapeutic area are freeze/thaw and shear. It is often necessary to freeze and then thaw both during development and manufacturing situations. An EPD can be created using the number of freeze/thaw cycles under defined conditions as an independent variable accompanied by temperature and pH stress. All three variables (temperature, pH, freeze/thaw cycles) can be combined into a three-dimensional representation in which the EPD is presented as a colored surface. Shear stress is also often encountered in the development, manufacture (especially filling), and shipping of protein pharmaceuticals. To explore this potential degrading stress, the intensity of the shear can be varied by a mechanical process such as stirring, shaking, or some other forms of agitation, and this is used as a variable in EPD production.

Protein concentration is another important variable that has assumed increasing importance with the use of high-concentration formulations. This variable can be typically evaluated over the range of 0.05–300 g/L depending on the solubility of the protein and the methods employed in the analysis. Proteins usually alter their structure to little or no extent as a function of protein concentration, but aggregation and surface adsorption are both highly concentration dependent. Thus, aggregation-­sensitive techniques such as light scattering are often of special importance in protein concentration-dependent studies. Another common variable of particular importance is ionic strength. In the Debye–Huckel charge shielding regime (0–0.15 ionic strength), a number of intermediate concentrations should be evaluated to probe electrostatic effects. At higher salt concentrations both preferential hydration and binding effects usually dominate with salt concentration into the molar range appropriately examined.

It is also possible to create EPDs based on phenomena such as aggregation. A variety of different types of aggregates have been identified in protein solutions based on their relative size and the nature of a protein’s conformation (altered or native) within the aggregate. A number of methods are available that are sensitive to these features (see Chap. 9), permitting an aggregation-based EPD to be created. Again, using variables like temperature, pH, ionic strength, freeze/thaw, and shear, soluble protein aggregates can be detected by methods such as size-exclusion HPLC, sedimentation velocity analytical ultracentrifugation, FFF, and DLS. Complimentary structural data can be obtained by the methods described above with FTIR and Raman spectroscopy especially useful because of the particulate nature of such samples. Larger (i.e., submicron) aggregates can be characterized by DLS including single-particle microscope-based approaches (nanoparticle tracking analysis) and classic microscopy-based techniques (atomic force microscopy, scanning electron microscopy, and transmission electron microscopy) although they are difficult to quantitate and new methods such as quartz crystal microbalances and nanomechanical resonators are seeing increasing use. Subvisible and visible particles can be analyzed by methods such as coulter counting, light obscuration, micro-­flow digital imaging, and various visual procedures. Using parameters such as size, composition, structure, and particle number as dependent variables, an EPD can be generated that provides a comprehensive picture of the nature of protein aggregates that form under a wide variety of stress conditions.

Although not yet described in the scientific literature, chemical degradation can also be analyzed in various high-throughput modes and be summarized in EPDs. The application of EPDs described above has all been to various physical processes in which covalent bonds are not broken. Of equal importance to protein degradation, however, are chemical changes such as oxidation and deamidation events. Chemical changes are usually quantitatively determined by peptide mapping combined with mass spectrometry (MS) in the form of HPLC-MS experiments which permit both the amount and location of a residue modification within a protein’s amino acid sequence to be determined. To increase throughput, once the nature of any changes has been identified, HPLC-MS analysis may be replaced by methods such as RP-HPLC and capillary electrophoresis or isoelectric focusing. A convenient way to present and analyze such data is in the form of rate constants for the individual, for example, deamidation and oxidation events. This of course requires time-­dependent measurements. The rate constants can be used as the dependent variables in an EPD since they are typically determined as function-independent variables such as pH and temperature. Of special interest is the comparative use of physical and chemical EPDs which permits an exploration of the relationship between structural changes and chemical events (and vice versa).

As a final example, EPDs can be used in a strictly comparative mode. Structural comparisons between both similar and assumed identical proteins are often a very important element of pharmaceutical analysis. For example, when manufacturing changes are made, during the development of biosimilars or when investigating mutant proteins, a detailed comparison of the various species is a critical part of the analysis. A direct comparison of EPDs of the target molecules provides a ­convenient and sensitive way to see if structural identity has been obtained. As an example, second-generation functional mutants of fibroblast growth factor one (FGF-1) have been compared using EPDs and used to select molecules that are not dependent on heparin for their activity (Alsenaidy et al. 2012). Such EPD-dependent comparisons have even been performed with different rotavirus serotypes despite their individual complexity (Esfandiary et al. 2010). Various mathematical (difference) methods exist to facilitate such comparisons.