HOS 2019: Chaos to Control in Less Than a Decade

Ten years ago, the buzz among the biotherapeutic community was about protein aggregation. However, in 2019, the intensity on the topic has declined, but it would be a huge mistake to cite the decline as evidence the topic has been resolved. Several new technologies have emerged, and interesting workarounds have also helped. However, the most promising advance is the extension of analytical ultracentrifugation (AUC) into the formulation concentration range.

The formulation range

Even the FDA seems to recognize that scientists developing biotherapeutics are using the best available technology (BAT), but BAT is not solving the problem of aggregation in the formulation range. Techniques such as HPLC, size exclusion chromatography (SEC), field flow fractionation (FFF), light scattering, MS, etc., are suitable for characterizing biotherapeutics in dilute solution. But, the effective dose of biotherapeutic is usually in the range of 50–500 mg. Most are administered by injection. Typical injection volumes are about 1 mL. This means that the active pharmaceutical ingredient (API) in the injection liquid is 50 mg/mL or higher, often 10 times higher. Strange things seem to happen in very concentrated solutions of biomolecules plus excipients.

AUC at high concentration

Until this meeting, I would have included AUC with the other dilute solution technologies including SEC, FFF, and light scattering. However, Peter Schuck of the National Institutes of Health reported extension of the useful concentration to the “100-mg/mL range.” The real story is how: One step was to extend the detailed treatment for nonideal solution behavior1 to include long-range hydrodynamic, electrostatic, steric, and other interactions. Plus, Schuck’s group used 3-D printing to make specialized rigid, thin sample cells that reduce optical distortion in their AUCs.

Workarounds

Drug development employs several competing armies of smart, often entrepreneurial, scientists. They are quick to see patterns and test hypotheses, especially in structure–activity relationships. Often, they are using advanced AI technology to see patterns in primary protein structure might lead to efficacy but fail because of instability. Or, they use dilute solution technology to characterize allosteric binding sites important in mechanisms of action (MOAs).

Smart workflows have quickly evolved.

Dr. Carl Jone of Narwhal Sciences in Brussels opened the meeting with an interesting description of the importance of higher-order structure in development of biotherapeutics from “Characterization to QC.” The QC part is obvious, since the critical quality attributes (CQAs) associated with the structure of the desired product are the reason for the entire expensive effort. However, he pointed out that the concern about structure begins with selection of the cell line used to make the product. This may involve comparing the structure of millions of proteins produced by as many individual cells. Experienced developers are able to apply empirical rules-of-thumb to weed out proteins with problematic amino acid sequences, or hot spots that increase the probability of poor performance scenarios.

A day later, Professor Paul Dalby of University College London discussed several technologies that are now combined to predict protein stability at the discovery stage of protein development. One is forced degradation of the proteins at elevated temperature, which shows unfolding of third- and fourth-order structures at 25 °C and small temperature increments up to 55 °C (Figure 1). Protein folding is usually driven by favorable enthalpy (ΔH), which is also available from elevated temperature data. Using the activation energy and estimates of concentration, they can predict the stability profile of proteins with less than a milligram of protein sample. Predictions of solubility behavior are also useful.

ImageFigure 1 – Higher-order structures usually refer to third or fourth levels of the structure shown. The first level is the peptide backbone of the protein. The second level recognizes simple structure elements such as sheet or helix. Third-order structure connects the secondary structures. Quaternary structures show the space-filling molecule. Even more complex is the interaction of two or more proteins in a docking scenario as is proposed for transmembrane signaling.

According to Dalby, hydrogen/deuterium exchange (HDX) was the rage a few years ago. It proved to be useful in characterizing the binding sites, since binding protected the hydrogens from exchange. Now, the excitement of HDX has passed. He notes that it is accepted, routine, and useful in understanding the basics of drug candidates, and confirming the MOA.

Computational microscopy

Genentech is the acknowledged leader in the discovery and engineering of biomolecular therapeutics. With decades of experience, they have databanks that can be mined to support computational microscopy using AI technology. This is combined with external data sets including distillations from 35,000 publications, empowering predictions of protein structure and reactivity. Undesirable hotspots can be identified for mitigation. Often, in-silico predictions avoid experimentally difficult experiments.

Calorimetry

Structural changes in proteins are accompanied by enthalpic changes revealed by calorimetry. Differential scanning calorimetry (DSC) has been used to measure irreversible changes such as the melting point. More recently, DSC has been extended to measure the rate of aggregation or denaturization, especially below the Tm. Ernesto Freire of Johns Hopkins University introduced a new algorithm for kinetic deconvolution of the kinetics of isothermal calorimetry of proteins. He combined this with other techniques, including steric exclusion chromatography, to predict the long-term stability of proteins.

Novel technologies

As with prior meetings in the series, HOS 2019 was an attractive forum for the introduction of new products and advances in older ones.

Ion mobility spectrometry (IMS)

ImageFigure 2 – NIST mAb monomer and dimer. The spatial/temporal solution chemistry of monoclonal antibodies can be complex. The IonDx IMS spectrometer measures the cross-sectional area of all conformations in <2 min with femtomole detection sensitivity. Samples are 0.05 mg/mL and require only 100 nL for a single scan. This technology may have the potential to rapidly study the structure of protein aggregates and antibody–drug conjugates (ADCs) including bi-antibodies.

Ion Dx introduced a new ion mobility spectrometer that is custom-designed to separate mono charged molecules in the gas phase. The authors believe that the ion mobility in the gas phase mimics the solution conformation more accurately than the multiply charged ions obtained from an electrospray ionization (ESI) mass spectrometer.The key is a stage in the inlet where polonium irradiates the collision gas to provide low-energy electrons that quickly neutralize the positive charge on the adducts. The IMS spectrum takes less than two minutes to scan (Figures 2 and 3).

ImageFigure 3 – mAb oligomers and homo-adducts. Ion mobility provides a rapid survey of intact mAb and their nonspecific aggregates. Common applications include characterization of ADCs and forced degradation pathways. This plot reveals conformations of singly charged NIST mAb monomers, dimers, and trimers with increased concentration. The mass range of these singly charged ions extends beyond 450,000 Da, which is larger than most mass spectrometers.

Hydroxide radical protein profiling (HRPF)

In a variation of HDX, Scott Weinberger of GenNext Technologies described the use of hydroxyl radical (OH) to map the amino acids on the exterior of large proteins in solution. The OHis very reactive. Because of its high reactivity, it does not travel much after its photon-induced formation. Specifically, it seldom lives long enough to travel into the interior of the proteins. Instead, it bonds covalently to the first site it encounters. This facilitates location labeling with OH at sites exposed to the exterior protein surface. If the site is blocked by another molecule, as in binding, the OH is effectively blocked from bonding at that site. In contrast, HDX locates exchangeable hydrogens. But these are dynamic and subject to back-exchange. The reversible labeling introduces time-dependent ambiguities in the HDX protocol.

Professor Joshua S. Sharp of the University of Mississippi presented a lecture showing that the location specificity of hydroxyl radical protein footprinting (HRPF) can be improved by adding a free radical quencher such as guanine. This decreases the solution lifetime of the OH so that it only has time to react with the protein when it is immediately adjacent to the target protein. If it is a bit away, it will encounter a guanine, which deactivates the OH. The ability to covalently label proteins via HRPF should develop into an important tool in HOS studies.

Cryo-electron microscopy

Cryo-electron microscopy (cryo-EM) was often mentioned as a way to image HOS of proteins. Although it is new, the technique seems to be well accepted and generally used. Some images provided resolution better than 2.5 Å. The images of many particles showed similar morphological features such as a hole, bends, tubes, and nose. There were subtle references to cryo-EM tomography as being the next step to improved resolution.

Claudio Ciferri of Genentech described the complementary results from cryo-EM and nuclear magnetic resonance (NMR) for elucidating the details of HOS of proteins, including interaction with small and large molecules. He said that Genentech has invested more than $20 million in its facility. Ciferri noted that Genentech has used the facility to provide high-resolution structures of membrane–protein complexes bound to antibodies. Throughput is about two structures/week. Currently, he is working on several projects with protein complexes smaller than 100 kDa.

I was impressed with two groups that provide transmission electron microscopy as a service. Since TEM instruments are expensive and productive, this seems to fit the market need very well. Nanoimaging Services displayed a range of images on postcards. One showed homoaggregates and adventitious agents. Another showed a 3-D structure of an adenovirus, while another had an image of antibodies compared to longer, worm-like proteins.

One example showed the structure of apoferritin at 2.4-Å resolution; this is a 480-KDa protein with 24 subunits in octahedral symmetry. The company collected ~3,000 images using a Titan/Krios TEM (Thermo Fisher Scientific). The example showed the overall image and facilitated homing-in on aa 17–38 plus the surrounding water molecules in a gray background.

NMR

High-field NMR, especially 2-D, was mentioned very favorably as useful for showing similarity or differences. Resolution is a function of frequency and isotopic abundance. Proton and natural abundance of 13C provide useful signals for antibodies and other large proteins. The natural abundance of 15N is too low for routine work with large molecules. Perhaps Bruker’s recently announced 1.2-GHz spectrometers will help.

Christian Fischer of Bruker BioSpin presented a vendor lecture on comparing 1-D and 2-D NMR using high-field (500–800 MHz instruments). Amgen has developed a 1-D approach that is quick and useful to see if a sample is similar to a reference material. It is probably quicker than peptide mapping. However, 2-D approaches are much more revealing. Fischer discussed improved computer-assisted workflows including CCSD (combined chemical shift deviation) and ECHOS (easy comparability of higher-order structure).

Several lectures and posters provided case histories of NMR to elucidate structural details. For example, a few amino acids have methylene (CH2) or methyl groups that have unusually clear signals. Adjacent atoms provide more details in the fine structure.

Fabio Baroni and colleagues from EMD Serono described their success using 1-D and 2-D NMR to characterize HOS of proteins. Their throughput is now about 1–2 protein products per week. Baroni noted that NMR provides detail that is not obtainable from legacy techniques (CD, FTIR, fluorescence) for development of biosimilar therapeutics in formulation design and QC of products. An interlab comparison of results is significantly better than with other techniques. Comparison was enabled with the help of Bruker’s ProfileNMR.

For high-contrast images, get rid of the solvent

Halo Labs has found a better way to image subvisible solids in formulation. One starts with a 96-well plate with filter discs. The image of each disc is recorded and stored as background with the reader. Next, each filter disc is loaded with a 25-µL aliquot of sample. A slight vacuum is applied, which pulls the liquid through the filter, leaving the subvisible particles on top of the filter. The background image is subtracted from the sample image. The result is a high-contrast image of particles. Since the solvent is removed, colorless particles are clearly seen. This is true even if the solvent and particles have the same refractive index, which would usually make them invisible.

Epilog

Higher Order Structures 2019 attracted about 110 scientists to the San Mateo Marriott Hotel to discuss the state-of-the-art and future needs. The basic premise is captured in the meeting tagline, “Because Structure Matters.” It matters a great deal: 7 of the top 10 drugs by sales volume in 2017 were biotherapeutics. Six were antibodies. Sales totaled $85 billion. Today, anecdotal reports claim that about two-thirds of the active NDAs are for protein-based candidates.

About a decade ago, the biotech community was in near panic over protein aggregation. Everyone was talking about it, but few knew what to do about it. The FDA was aware that proteins could aggregate with potentially deadly consequences due to immunogenicity or physical blocking of the host’s circulatory capillaries. Some leaders of CASSS recognized the sorry state of affairs and organized the first scientific meeting of the Higher Order Structure series, which was held in 2011.

Three years ago, I attended HOS 2016 in Long Beach, CA. I sensed that things were beginning to gel. The FDA offered useful guidance2:

Later in development, full characterization of HOS using orthogonal methods is expected. HOS is an integral part of product characterization.

Assessment of aggregates is expected at all stages of development along with a risk assessment of their potential impact on safety and efficacy.

Plus, tools were starting to appear. The meeting focused on promising applications of developing technology. There was acceptance that more were needed, especially to meet the orthogonality requirement above.

Fast forward only three years to 2019. Speakers were notably more confident that the tools were working, leading to improved product flow. Plus, the feedback loop of information from late-stage development to early-stage structural optimization including host selection and protein engineering seems to be working to assure efficacy with safety.

However, the highly concentrated world of 100 mg/mL is still a relevant and difficult experimental space that begs for better technology.

So, plan on attending HOS 2020, scheduled for April 20–22, 2020 in Gaithersburg, MD. I hope to meet you there.

References

  1. https://www.nature.com/articles/s41467-018-06902-x
  2. Gutierrez Lugo, M. Regulatory Consideration for the Characterizaion of HOS in Biotechnology Products, slide 7. 5th International Symposium on Higher Order Structure of Protein Therapeutics, April 11–13, 2016, Long Beach, CA.

Robert L. Stevenson, Ph.D., is Editor Emeritus, American Laboratory/Labcompare; e-mail: [email protected]