The TPT active pharmaceutical ingredient (API) and FDA-approved impurities were evaluated using immunoinformatics analyses with the EpiMatrix and JanusMatrix algorithms (3, 6, 7). Several TPT impurities were selected for this study because they contained alterations to the promiscuous HLA-DR epitope in frame 5—such as amino acid insertions, deletions, or side-chain modifications—that reduced the epitope’s conservation with human proteome-derived sequences.

In some cases, impurity sequences contained unnatural or otherwise modified amino acid residues. While the HLA binding properties of peptides with unnatural amino acids cannot be directly estimated using standard in silico methods, a three-step approach was used to estimate the HLA binding properties of the modified sequence. Briefly, the modified position is substituted with all 20 naturally occurring amino acids, generating a range of possible scores to estimate the potential impact of the impurity modification on HLA binding. A wide range of scores for substitutions at a given position indicates that position may have a greater impact on predicted HLA binding, whereas a narrow range suggests that a modification at that position is unlikely to significantly impact the HLA binding affinity. Finally, a suitable natural amino acid substitution is selected for further analysis. This three-step approach for estimating HLA binding of sequences containing unnatural amino acids in silico has been described in greater detail previously (8).

Having defined the approach to handling unnatural amino acids in selected impurities, EpiMatrix was used to identify T-cell epitopes in the sequences of the TPT API and 34 impurities (Figure 4). Each sequence was parsed into overlapping 9-mer frames and evaluated against a panel of nine class II supertype alleles (HLA-DRB1*0101, HLA-DRB1*0301, HLA-DRB1*0401, HLA-DRB1*0701, HLA-DRB1*0801, HLA-DRB1*0901, HLA-DRB1*1101, HLA-DRB1*1301, and HLA-DRB1*1501) to assess binding likelihood (3). These “supertype” alleles represent the binding preferences of HLA-DRB1 alleles for approximately 95% of the global population, and their selection has been described in detail elsewhere (7).

Next, the JanusMatrix algorithm was employed to determine whether the API and its impurities contained sequences that could be tolerated or tolerogenic. JanusMatrix identifies putative T-cell epitopes in protein and peptide sequences that, while not necessarily identical in sequence to known human T-cell epitopes, are conserved at their TCR-facing residues and retain the same HLA binding restriction. Many such epitopes have been shown to be tolerated or even tolerogenic in vitro and in vivo when extensive cross-conservation with the human proteome is observed (6, 9). As demonstrated for highly cross-conserved and tolerogenic T-cell epitopes, modification of the TCR-facing residues typically reduces their tolerogenic properties (10, 11). Similarly, a reduction in the cross-conservation of the frame 5 promiscuous and cross-conserved epitope in the API might lead to increased immunogenicity in vitro and in vivo, as discussed in greater detail below.

Finally, the impurity sequences were assessed for their potential to increase the immunogenicity of the drug product due to the presence of new epitopes predicted in the impurity but not in the API (3). Specifically, the analysis prioritized amino acid modifications likely to introduce new HLA ligands and/or new TCR-facing contours compared to the unmodified amino acid sequence. In the context of immunogenicity risk assessment, both types of these newly created epitopes (new ligands and new TCR-facing residues) can be considered “new epitopes” to the host immune system (different from those in the API).

This emphasis on the creation of new epitopes that are not conserved with the API relates to the additive effect of effector T-cell responses to these epitopes. Immune responses to new epitopes could induce or enhance immune responses targeting the active drug ingredient through a process known as “epitope spreading” (12–15). In contrast, T-cell epitopes that are identical in both the unmodified product and the amino acid sequences of product impurities (common epitopes) may engage and activate cognate T cells but are considered unlikely to induce new immune responses and are not expected to contribute to an increase in immunogenicity.

In addition to prioritizing new epitopes, in silico assessments emphasize the identification of promiscuous binding sequences that are likely to bind to multiple HLA-DR alleles. Impurities containing modifications in these promiscuous sequences are likely to introduce novel epitopes capable of inducing CD4⁺ T-cell responses across multiple alleles (16, 17); for this reason, such impurities are prioritized for evaluation both in silico and in vitro.

A new algorithm was developed for this case study that can be used to explore multiple potential impurities that may impact the immunogenicity of a generic peptide. This algorithm, called the What-If-Machine (WhIM), can be used to create a ranked list of potential impurities and prospectively identify impurities with sequences that would have high (or low) immunogenic risk potential (18). WhIM is an iterative algorithm that produces lists of theoretical impurity peptides, each containing a single modification from a baseline sequence (generally the API sequence). WhIM records possible sequence-related impurities created through known failures in the synthesis process (19). WhIM mimics the solid-phase peptide synthesis process, in which each addition of an amino acid from the C-terminus to the N-terminus is considered a cycle. At each cycle, WhIM documents theoretical impurities based on possible single modifications to the baseline sequence. These modifications include amino acid deletions, insertions, duplications, and side-chain modifications. Ultimately, the WhIM algorithm generates a list of theoretical impurities for a given API and then assesses each for its potential immunogenicity using EpiMatrix, ClustiMer, and JanusMatrix, and creates scores for each of the modified peptides.

Thousands of theoretical synthetic peptide impurities of TPT, resulting from sequence changes (insertions, deletions, duplications, and others), were designed using the tool. Following the generation of this extensive list of theoretical peptide-related impurities by WhIM, existing algorithms (EpiMatrix and JanusMatrix) were used to rank each member of the list for its potential to generate immune responses in global patient populations. These algorithms are described in greater detail in the Supplementary Material and in De Groot et al. (3) and Mattei et al. (5). The results of the WhIM analysis can be plotted on a quadrant plot showing their risk potential relative to defined EpiMatrix and JanusMatrix thresholds (and to the “observed” impurities tested in this study) (see Figure 5).

Following the in silico analysis of the 34 observed impurities (from the FDA) and thousands of WhIM-identified potential impurities, nine impurities were selected for further testing. These included two impurities selected to represent low-risk impurities (for impurity-induced immunogenicity) and five selected to represent high-risk impurities (based on their estimated immunogenicity) by in silico analysis. In addition, two high-scoring theoretical impurities designed by the WhIM were selected for evaluation in vitro.

As class II HLA binding peptides are usually 12–25-amino acid long, the peptides produced for in vitro assays were designed to capture the modified region and center any predicted binding epitopes in the middle of the peptide to be tested, while remaining consistent with standard class II peptide length limits. In general, peptides used in HLA-DR binding assays are shortened to improve the potential for the assay to capture the effect of the centered epitope, whereas peptides evaluated in T-cell assays are tested as full-length peptides (see Table 1) (20).

Impurity peptides used in the in vitro studies were synthesized by 21st Century Biochemicals (Marlborough, MA, USA). Molecular weight was verified by mass spectrometry, and all peptides were determined to be greater than 95% pure by HPLC. Forteo^® (RLD) drug product was purchased from Pharmaceutical Buyers Inc. (New Hyde Park, NY, USA).

Class II HLA binding assays used in these studies measure the binding affinity of a target peptide to HLA-DRB1*0101, HLA-DRB1*0301, HLA-DRB1*0401, HLA-DRB1*0701, HLA-DRB1*0901, HLA-DRB1*1101, HLA-DRB1*1301, and HLA-DRB1*1501. The HLA binding assay was originally described by Steere et al. (21) and adapted by EpiVax (22, 23). Briefly, unlabeled test peptides were incubated overnight to equilibrium with a soluble HLA-DR molecule (Benaroya Research Institute, Seattle, WA, USA) and a biotinylated, allele-specific competitor peptide of known binding affinity. The binding reaction was then neutralized with a buffered pH change, and the peptide-HLA complexes were transferred to a 96-well plate coated with a pan-HLA-DR antibody, clone L243 (BioLegend, San Diego, CA, United States), and incubated overnight. The following day, plates were resolved with the addition of europium-labeled streptavidin (Perkin-Elmer, Waltham, MA, USA). An indirect measure of binding was determined by time-resolved fluorescence. Each peptide was evaluated in triplicate over a range of seven concentrations. The percent inhibition values for each experimental peptide across this concentration range were used to calculate an IC₅₀, the concentration at which the peptide inhibits 50% of the labeled competitor peptide.

PBMCs used in these assays were freshly isolated from leukocyte reduction filters purchased from the Rhode Island Blood Center (RIBC) in Providence, RI. High-resolution (four-digit) class II HLA haplotyping of donors was performed at the Transplant Immunology Laboratory at Hartford Hospital (Hartford, CT, USA) using the sequence-specific oligonucleotide method. Donor HLA-DRB1 types, age, and sex are provided in Supplementary Data (see Supplementary Figure S1). All Donor PBMC samples had greater than 85% viability, postisolation.

To measure the naïve CD4⁺ T-cell response to TPT and impurities, freshly isolated PBMCs were plated at a density of 2.5 × 10⁵ cells per well in 96-well U-bottom cell plates in RPMI-1640 cell culture media supplemented with IL-2 (10 ng/mL) and IL-7 (20 ng/mL) (Gibco, Grand Island, NY, United States). For each test article and control, nine replicate wells of 2.5 × 10⁵ cells (2.25 × 10⁶ cells total) were plated. Cells were incubated with Forteo^® (the approved TPT RLD product) or individual impurities at 20 µg/mL for 14 days at 37°C/5% CO₂. Micromolar (µM) concentrations for peptides evaluated in the in vitro immunogenicity protocol (IVIP) assay are provided in Table 2, below (data for impurities evaluated at 0.2 µg/mL is provided in Supplementary Figure S4). Media exchanges, including cytokine support (IL-2 and IL-7) but no antigen, were performed on days 4, 7, and 11. The in vitro concentration of 20 µg/mL was chosen based on a dose-optimization study performed on Forteo^® to determine the maximum dose at which a T-cell response could be observed with no toxicity to the cells.

Each set of donor PBMC was plated with the test peptides as well as the following sets of controls: positive control: keyhole limpet hemocyanin (KLH; Thermo Fisher, Waltham, MA, United States) and the antigenic memory peptide pool Cytomegalovirus, Epstein–Barr virus, and influenza (CEFT; ImmunoSpot, Cleveland, OH, United States); negative control: human serum albumin (HSA; Sigma-Aldrich, Saint Louis, MO, United States); and Functional Control: Phytohemagglutinin (PHA; Thermo Fisher), a well-known T-cell mitogen. Each test article and control peptide was tested in nine replicate wells. Only donor PBMC that generated a positive response to KLH, CEFT, and PHA, and a negative response to HSA were included in the study.

To compare the impact of the Forteo^® formulation on cell viability and immunogenicity, impurity peptides evaluated in these assays were reconstituted in either supplemented RPMI-1640 cell culture media or a diluent formulated to replicate the Forteo^® product formula (1), which contained: 0.41 mg/mL glacial acetic acid, 0.1 mg/mL sodium acetate anhydrous, 45.4 mg/mL mannitol, 3.0 mg/mL metacresol, and pH 4.0, as specified in the Forteo^® package insert.

Please see the discussion for an update on this approach that resulted from this and other studies conducted in this EpiVax-FDA collaboration.

Following the 14-day incubation, cells were harvested and plated in triplicate on a precoated antihuman interferon gamma (IFN-γ) Fluorospot plate (Mabtech, Cincinnati, OH, United States) at a concentration of 1.0 × 10⁵ cells per well in the presence of the appropriate test article in RPMI-1640 culture media. Fluorospot plates were incubated for 48 h at 37°C/5% CO₂. After 48 h, Fluorospot plates were developed according to the manufacturer’s instructions. Fluorospot plate counting was performed by unbiased experts at ZellNet Consulting Inc. (Fort Lee, NJ, USA) as previously described by Roberts et al. (24). A response was considered positive if (1) the number of spots was at least twice the background (SI ≥ 2), and (2) there were greater than 50 spot-forming cells (SPC) per 1,000,000 PBMCs.

To evaluate the regulatory potential of predicted regulatory T-cell (Treg) epitope peptides identified in TPT, a previously published tetanus toxoid bystander suppression assay (25) was adapted to determine if predicted Treg epitopes could suppress the memory CD4⁺ T-cell response to the common tetanus toxoid antigen. Cryopreserved PBMCs were thawed, the viability was measured postthaw using AO/PI viability dye on a Nexelcom Cellometer (Nexelcom, Lawrence, MA, United States), and then labeled with carboxyfluorescein succinimidyl ester (CFSE, eBioscience, San Diego, CA, United States). The CFSE-labeled cells were plated in 96-well U-bottom culture plates at a concentration of 3.0 × 10⁵ cells per well in RPMI-1640 cell culture medium supplemented with 2 nM l-glutamine, 50 µg/mL gentamicin (Life Technologies, Beverly, MA, United States), 10% human AB serum (Sigma), and MEM nonessential amino acids and 55 µM β-mercaptoethanol (Gibco), known as complete RPMI-1640 (cRPMI-1640) and rested overnight at 37°C/5% CO₂. The putative regulatory TPT peptide was solubilized in dimethyl sulfoxide (DMSO) and further diluted to assay concentrations in cRPMI-1640 culture medium. At day 1, cells were stimulated with tetanus toxoid (TT; Astarte Biologics, Redmond, WA, United States) at 0.5 µg/mL alone for positive control of the experiment or in combination with putative peptide TPT Treg epitope, in a series of concentrations at 0, 10, 20, and 40 µg/mL and incubated for 7 days at 37°C/5% CO₂. After 7 days, cells were stained for the expression of cell surface markers and analyzed by flow cytometry (see below). Wells treated with TT alone and TT⁺ Treg epitope-treated cells were compared after 7 days. TPT was compared with two well-characterized Treg epitopes (Tregitope 289 and FV621) that had been previously identified and extensively validated (26–30). Negative control peptides have been included in previous assays and compared with the positive controls used in this study (see, for example (31)).

Cells were washed with 1× PBS and stained with live/dead viability stain (Life Technologies, Beverly, MA, United States) according to standard procedures. Cells were then washed with 1× FACs buffer (1× PBS+5% FBS), and surface-stained for CD3 (clone OKT3), CD4 (clone OKT4), and CD25 (clone BC96) (BioLegend) in 1× PBS+5% FBS at 4°C for 30 min. Stained cells were washed twice with FACS buffer and acquired on an Attune NxT Acoustic Cytometer with three laser capacity (violet: 405 nm; blue: 488 nm; and red: 637 nm; Life Technologies) and analyzed for proliferation (dilution of CFSE) and surface antigen expression using FlowJo Software (Treestar Inc., Mesa, AZ, United States).

A detailed in silico analysis of the TPT API sequence is presented in Figure 2. TPT contains 19 predicted HLA-DRB1 ligands, or EpiMatrix “hits”, which is more than one would expect to find in a random peptide of equivalent length, and consequently, TPT falls in the elevated range on the immunogenicity scale (greater than 10) based on total predicted T-cell epitope content. Eight of the 19 EpiMatrix hits were concentrated within a promiscuous T-cell epitope (highlighted in yellow), located at frame 5. Due to the extensive HLA binding potential in frame 5, the overall score of the sequence was high, and the promiscuous binding region in the frame 5 epitope was predicted to be a relatively high-affinity HLA ligand. Evaluation for cross-conservation with the human genome using the JanusMatrix algorithm, however, revealed extensive human genome epitope homology at the TCR-facing residues of this peptide (see below). Additional EpiMatrix hits that may be immunogenic are in frames 4, 7, 8, 11, 21, and 23. While these additional hits may be T-cell epitopes, their individual effects are expected to be more HLA restricted (as they are not within a promiscuous T-cell epitope).

The JanusMatrix algorithm was employed to screen the putative T-cell epitopes identified within TPT against the human proteome (Figure 3). The TPT sequence shares a high number of TCR-facing contours with peptides derived from self, resulting in a JanusMatrix human homology score of 4.74 (a high score, indicating that the peptide is likely to be tolerogenic). Specifically, the 9-mer peptide in frame 5 shares a TCR-facing pattern with epitopes derived from 12 human proteins, including peptides from parathyroid hormone (as expected) and numerous subtypes of β-tubulin (a ubiquitous structural protein in human cells), integrator complex, and insulin receptor-related protein. The HLA ligand in frame 11 was related by JanusMatrix to three human proteins at its TCR face, including peptides derived from from parathyroid hormone and annexin. The HLA ligands in frame 23 could be related to three human proteins at its TCR face, including peptides derived from parathyroid hormone, WASH complex, and desmocollin-1. The HLA ligands in frames 4, 7, 8, and 21 share some homology with parathyroid hormone at the TCR face.

As a result of the extensive homology with human genome epitopes uncovered in this analysis, it was postulated that the TPT epitope (frame 5) sequence-specific T cells may have either been deleted during thymic development, rendered anergic, or adopted a Treg phenotype. Consequently, changes to this 9-mer region, particularly those that abrogate homology with other peptides in the human proteome, might be detrimental to homology-induced tolerance and lead to increased immunogenicity risk (31).

The 34 observed impurities evaluated in silico included impurities resulting from amino acid deletions, amino acid duplications, side-chain modifications (including oxidation, succinimide formation, dehydration, acetylation), N- and C-terminal truncations, and N-terminal extension and modification (Figure 4). These peptide-related impurity sequences were specifically evaluated for new immunogenic (nonhuman-like) epitope content relative to the TPT API sequence (Supplementary Table S1).

The EpiMatrix immunogenicity scores of the submitted impurity sequences (Supplementary Table S1) varied from low- to high-risk potential (higher EpiMatrix scores denote higher T-cell epitope content). Additionally, the JanusMatrix (human homology) analysis of the impurity sequences resulted in scores ranging from 1.42 (potentially immunogenic) to 5.25 (potentially tolerated or tolerogenic). Some impurity modifications, therefore, resulted in sequences with higher EpiMatrix scores and reduced JanusMatrix scores, and consequently, potentially higher immunogenicity risk.

Relative to the TPT API, we hypothesized that some modifications had the potential to reduce tolerogenic epitope content, which could drive an impurity-induced immune response. In these cases, the peptide impurities assessed had (1) equivalent, (2) elevated epitope content, and/or (3) reduced human homology relative to the TPT API. Impurities with elevated epitope content (higher EpiMatrix scores) and low human homology (low JanusMatrix scores) were identified as the most likely to be immunogenic.

Figure 5A shows the immunogenicity quadrant plot and relevant benchmarks. TPT falls into the top right quadrant, indicating that, while it contains a high amount of predicted epitope content, its epitopes were highly conserved with epitopes derived from the human proteome. Peptides residing in this quadrant include known tolerogenic control peptides, suggesting that the API may also be tolerogenic. In contrast, peptides known to be immunogenic can be observed in the top left quadrant, which carries the highest immunogenicity risk. Human-derived peptides with clinically low immunogenicity can sometimes fall in the left quadrants, indicating low JanusMatrix human homology scores due to their lack of cross-conservation with epitopes derived from additional autologous proteins cataloged in the JanusMatrix (UniProt, Geneva, Switzerland) database of human proteins, beyond that from which the database was derived (see glucagon, lira/semaglutide). Figure 5B shows the results of the in silico analysis of theoretical TPT impurities identified with the WhIM algorithm (small gray circles), as well as the potential immunogenicity of the 34 observed TPT impurity sequences (medium gray circles), as defined by their EpiMatrix score (y-axis) and their JanusMatrix score (x-axis), relative to the TPT API (large black circle). Those impurities that maintain the tolerogenic sequence in frame 5 are co-located with the parent peptide on this plot and are also considered lower risk for immunogenicity. Note that other TPT impurity sequences that have increased or equivalent T-cell epitope content and reduced human homology compared to the TPT API move up and left on the plot into the higher-risk quadrants (upper left). Several of these impurities (diamond-shaped on the quadrant plot) were found to have an enhanced immunogenicity risk potential when compared to the API peptide.

Following the in silico analysis, epitope binding predictions by EpiMatrix were evaluated in vitro by measuring the binding affinity of each impurity peptide relative to the corresponding N- or C-terminal portion of the TPT API, in vitro. The class II HLA-DRB1 binding assay was adapted from an assay originally described by Steere et al. (21) that provides an indirect measurement of peptide-HLA affinity. Class II HLA binding peptides typically range in length from 12–20 amino acids; however, the core HLA binding region is nine amino acids. The class II HLA binding groove is open-ended, and the additional flanking residues on either end of the core binding motif stabilize the peptide: HLA binding interaction. Targeted binding motifs were properly centered within the shorter length peptides synthesized for these binding studies (Table 3).

The HLA binding results demonstrate that the N-terminus of TPT (which contains a promiscuous HLA binding motif in frame 5) bound to seven of the eight class II Supertype Alleles evaluated in these assays with negligible to high affinity (Figure 6A). These results support the in silico EpiMatrix assessment of the N-terminal promiscuous T-cell epitope (frame 5), as the assessment suggested promiscuous binding behavior. The HLA-DR promiscuity of this peptide may contribute to its activity as a Treg epitope (because the peptide may be tolerogenic for most subjects). These assays also independently confirmed that the C-terminus of TPT (which was assessed by EpiMatrix as having very few T-cell epitopes) bound to only three of the eight class II alleles tested, supporting the hypothesis that any immunogenic risk these impurities may pose would be HLA-restricted (Figure 6B).

To determine if the impurity modifications impacted class II HLA-DRB1 binding, impurity peptides were evaluated alongside their respective TPT API peptide (N-term or C-term) in HLAs. The resulting data is summarized in Figure 6B below. (Detailed results for each peptide-allele assessment are available in Supplementary Figure S2). Overall, most impurities showed equivalent or increased affinity as compared to the corresponding region of the API. This was consistent with the EpiMatrix predictions of equivalent or increased epitope content in these impurities relative to the API. By contrast, impurity Des-Leu7 was predicted by EpiMatrix to significantly reduce overall epitope content and likewise showed decreased or equivalent binding affinity compared to the corresponding region of the API.

As previously mentioned, the low incidence of observed clinical immunogenicity associated with TPT may be attributed to (a) TPT being derived from the N-terminus of human PTH (an endogenous peptide) and/or (b) due to the presence of a potential Treg epitope in the N-terminus that is cross-conserved with β-tubulin and other self-proteins. Of significant concern is whether or not impurities that make the peptide appear more foreign (by changing the tolerogenic epitope) might impact the immunogenicity of the drug product. To assess the immunogenicity of the impurities relative to the RLD product, Forteo^®, PBMCs from healthy donors were incubated with individual impurity peptides or the Forteo^® drug product at an equal concentration to the API.

Data collected during a previous analysis of salmon calcitonin (24) and innate immunogenicity studies conducted by Holley et al. (32) indicated that excipients within the formulation could impact cell health and viability in culture. Thus, synthetic peptide impurities were reconstituted in either cell culture media, or a buffer made in our laboratory that was made to mimic the Forteo^® product buffer (0.41 mg/mL of glacial acetic acid, 0.1 mg/mL of sodium acetate anhydrous, 45.4 mg/mL of mannitol, 3.0 mg/mL of metacresol, pH 4.0 as specified in the Forteo^® package insert) (1).

As shown in Supplementary Figure S5, when compared to impurity peptides reconstituted in RPMI-1640 cell culture media, no difference in the T-cell response, measured by IFN-γ secretion was observed when compared to impurity peptides from the same manufacturing lots reconstituted in our laboratory-made product formulation buffer. For this reason, the data presented were from studies where impurities have been reconstituted in RPMI-1640 culture medium. The immunogenicity of individual impurities was then compared to Forteo^®, the TPT RLD, as shown in Figure 7.

In this case study, donor PBMC responses to the RLD were assessed to formulate the product (Forteo^®, in this case) rather than the API peptide. Differences in immunogenicity between Forteo^® and the individual peptide impurities were assessed with two different approaches: (1) the frequency of PBMC donor that elicited T-cell responses (IFN-γ secretion) that were positive (SFC > 50 and SI > 2) within the study cohort (Figure 7, hashed grey bars; and (2) IFN-γ SFC distribution plot, which compared the mean distribution of the IFN-γ-secreting cells per million PBMCs of Forteo^® to the impurities for the donor PBMC cohort (Figure 7, colored bars showing the mean SFC count for each donor PBMC response, with individual donor samples represented by circles).

Each of the selected impurities elicited an IFN-γ response in a greater number of donors when compared to Forteo^® (Figure 7, hashed grey bars). Consistent with the increase in PBMC donor response rate, the mean expression of IFN-γ SFCs per million cells increased for each of the individual impurities when compared to the RLD product (Figure 7, colored bars and circles). These results suggested that eight of the nine selected impurities (the exception being Des-Leu28, the peptide with the highest JanusMatrix score) have a higher immunogenic risk potential compared to Forteo^®, either due to less “human-like” content (lower JanusMatrix score) or the introduction of new T-cell epitopes (minimally different JanusMatrix score but higher epitope content). These impurities have the potential to enhance the immunogenicity risk of a generic TPT drug product if present in sufficient quantities in the final product. A summary of the in silico analysis is compared to the results of the naïve donor T-cell assay to make an overall assessment of risk in Table 4 below. Additional examples of another set of TPT impurities with a loss of “human-like” content from independent studies can be found in Supplementary Figure S3.

We note that impurity Lys-Ac26 was considered a nonbinder when evaluated in the HLA binding assays. Peptides evaluated in the IVIP T-cell assay are full-length peptides, while peptides tested in the HLA binding assay are shorter peptides, 12–20 amino acids in length (consistent with the length of Class II HLA presented peptides), with the impurity modification centered within the peptide. From the in silico analysis, impurity Lys-Ac26 has both higher predicted epitope content and reduced human homology compared to the TPT API, suggesting that this impurity would have greater immunogenic potential when compared to TPT (Forteo^®), consistent with the results of the T-cell assay. Additional studies, such as three-dimensional peptide-MHC modeling, may clarify the discordance between these two results. The limited number of flanking amino acids (3) that are present at the N terminus could influence the peptide core binding resulting in the “no binding” classification Alternatively, this peptide could contain other process-related impurities that increase the response of a 14-day T-cell activation assay, although they would only be present in minute quantities in a synthetic peptide that is 95% pure.

Immune responses to the peptide impurities were lower when evaluated at a concentration of 0.2 µg/mL (1.0% of the API) compared to impurities evaluated at 20.0 µg/mL (see Supplementary Figure S4). Although this might suggest a lower risk of immunogenicity when impurities are present at the thresholds specified in the ANDA guidance, given that the TPT product is usually administered daily for 24 months or more, even an impurity with lower risk potential at specified limits could potentially exhibit a higher risk potential over months of repeat daily exposure.

Since in vitro immune responses may be reduced when in vitro assays are performed in the presence of some drug formulations (see Roberts et al. (24)), we performed an in vitro T-cell assay using two separate TPT API peptides in cell culture medium (from two different API manufacturers), rather than the RLD. Donor PBMC responded similarly to the two API peptides (Figure 8). Minimal immune responses to the TPT API peptide were observed in T-cell assays performed with multiple donor PBMC samples, when considering both the percentage of responsive PBMC donors and the magnitude of the IFN-γ cytokine response, as was also observed for the Forteo^® drug product.

In summary, the in vitro T-cell assay results were consistent with the above (independent) in silico assessments in that they suggested that some impurities have an overall higher immunogenic risk potential compared to TPT. Three of the nine impurities (Des-Gly12, Endo-Leu11, and Des-Leu11) had higher EpiMatrix Scores than the TPT API (>16.03) and low JanusMatrix Scores (< 3.00), indicating the highest risk for immunogenicity (dark red shading in Figure 5b). Forty-three percent of donors responded to these highest risk impurities at equal concentration to the API. Three of the nine impurities (Des-His14, Des-Lys13, and Lys-Ac26) had higher EpiMatrix scores (> 16.03) and lower JanusMatrix Scores (3.00–4.74) than the TPT API, indicating a high risk for immunogenicity (light red shading in Figure 5b). Between 38% and 48% of donors responded to these high-risk impurities at equal concentration to the API. One impurity (Des-His9) had a high EpiMatrix score (10.00–16.03) and a low JanusMatrix score (< 3.00), indicating a high risk for immunogenicity (light red shading in Figure 5b). Forty-eight percent of donors responded to this high-risk impurity at equal concentration to the API. One impurity (Des-Leu7) had a low EpiMatrix score (< 10.00) and a low JanusMatrix score (< 3.00), indicating a medium risk for immunogenicity (white shading in Figure 5b). A total of 43% of donors responded to this medium risk impurity at equal concentration to the API. One impurity (Des-Leu28) had a high EpiMatrix score (> 10.00) and a higher JanusMatrix score (> 4.74) than TPT API, indicating a high potential for immunogenicity, which is likely to be offset by homology-induced tolerance (light green shading in Figure 5b). Only 24% of donors responded to this low-risk impurity at equal concentration to the API.

The assay results (T-cell assay, bystander assay) supported the observation that the promiscuous epitope identified in frame 5 (that is conserved with PTH and β-tubulin) may be tolerogenic. This potential tolerogenicity is based on the observation that any impurity peptide containing modifications in this region that changed the TCR face resulted in an increase in the number of donor PBMC samples that responded to the peptide, and increased the magnitude of the IFN-γ response as well. For example, immunogenic impurities Des-Leu7_TPT, Des-His9_TPT, and Des-Leu11_TPT all have modifications to TCR-facing residues in the purported Treg epitope sequence (frame 5), which reduced the JanusMatrix human homology relative to the API (see in silico scores listed in Figure 9). These results support the hypothesis that reduced cross-conservation with the human proteome or an increase in predicted T-cell epitope content may increase the immunogenicity of synthetic peptide impurities.

In addition to the seven synthetic TPT impurities selected for in vitro testing from the 34 identified by the FDA through internal studies, two theoretical impurities identified by the novel WhIM algorithm were evaluated in the ex vivo T-cell assay. These impurities, WhIM_ENDO-Leu11_TPT and WhIM_DES-Gly12_TPT, which were selected for their elevated EpiMatrix scores and reduced JanusMatrix scores relative to TPT, elicited a response in twofold more PBMC donors than the number that responded to TPT (Figure 7). Additionally, these two peptides elicited a greater number of IFN-γ spot-forming cells when compared to Forteo^® and the other TPT-derived impurities (Figure 7). These data also suggest that the WhIM algorithm could be used to prospectively identify peptide impurities with an increased risk potential in generic drug products. Results from WhIM could be used to flag potentially immunogenic peptides that would be likely to induce an immune response, for removal by improvements to the manufacturing process. TPT exhibits suppressive capabilities in vitro.

Having confirmed the promiscuous HLA binding of the epitope in frame 5 to multiple HLA-DR molecules using the class II HLA binding assay, we hypothesized that the suppressive capabilities of this epitope in the API could help explain the overall low clinical immunogenicity observed for TPT, as well as the reduced response seen in the in vitro assessment of T-cell responses. To examine this, we performed a tetanus toxin bystander suppression assay (TTBSA), which was adapted from a publication by Barbey et al. (25). This assay has been used to validate Treg epitopes identified in IgG and other human proteins (33).

Here, the assay was used to determine if the identified promiscuous epitope in frame 5 of TPT functions as a regulatory T-cell epitope. The core 9-mer sequence of the API with flanking residues, TPT_API_2–16 (TPT_2–16), was synthesized and evaluated in the TTBSA assay. The TTBSA assay was used to evaluate the putative Treg epitope TPT_2–16 for its ability to suppress recall response to TT in PBMCs from TT-immune donors. Thus, to assess the suppressive capacity of the TPT_2–16, CFSE-labeled human PBMCs were incubated with either media, TT alone, or TT+TPT_2–16 at increasing concentrations for 7 days. Following 7-day incubation, cells were evaluated by flow cytometry for CD4 T-cell proliferation (CFSE dilution) and for activation status (CD4⁺CD25^high). The immunosuppressive capacity of TPT_2–16 was compared to the response to two well-characterized Tregitopes (289 and FV621) using the same donor PBMC in the assay, which is described in detail in De Groot et al. (33) and Miah et al. (34). Information on negative controls for the TTBSA is included in De Groot et al. (33). The detailed gating strategy for this assay is provided in Supplementary Figure S6.

Figure 10 provides the results of in vitro incubations and flow cytometry studies. This figure shows the results of experiments including (a) CD4 T-cell proliferation measurements and (b) CD4 T effector cell activation values in PBMC treated with TPT_2–16 as compared to PBMC treated with either media alone or tetanus toxoid alone. CD4⁺ T cells treated with media alone (Figure 10A) showed minimal proliferation after 7 days of culture. Also shown in Figure 10A, CD4 T cells displayed a 10-fold increase in proliferation when stimulated with TT antigen alone. In contrast, proliferation to TT decreased in a dose-dependent manner when the cells were coincubated with increasing amounts of TPT_2–16. Similarly, when the expression of CD25 (an activation marker) (35) on CD4⁺ T cells was measured in the same assay, cells treated with TT showed a fivefold increase in CD25⁺ expression compared to media alone (Figure 10B). CD4⁺ T cells exhibited a dose-dependent decrease in CD25 expression when cells were costimulated with TT plus the TPT_2–16 peptide.

A normalized comparison of the TPT_2–16-mediated suppression of CD4⁺ T-cell proliferation and activation for three donors is presented in Figure 11. Results for each individual donor are provided in Supplementary Figure S7. The TPT_2–16 peptide suppressed memory response to TT at all doses, including at the highest peptide concentration of 40 µg/mL (p ≤ 0.01) in PBMC samples from all donors. The suppressive nature of TPT_2–16 is comparable to that elicited by two previously identified and characterized Tregitopes, 289 (derived from human IgG) and FV621 (derived from factor V) (33) (Figure 11).

Collectively, these assays suggested that TPT contained a putative regulatory T-cell epitope in frame 5. This Treg epitope appeared to suppress T-effector (memory) response to a known antigen (TT in this case), so the presence of this tolerogenic epitope in the product may possess the capability to modulate the overall immunogenicity of TPT in clinical settings. The finding illustrates the importance of evaluating the type of T-cell that might respond to a given peptide drug, because disruption of a key regulatory T-cell epitope in impurities that were produced during the synthesis of the API might contribute to an increase in immunogenicity of the drug product in the patient population. Moreover, as the Treg epitope in TPT is a promiscuous HLA binder, impurities that impact this tolerogenic epitope may increase immunogenicity risk for a broad range of individuals in the treated population.