TotalLab

High-Risk Host Cell Proteins: The Named Offenders, What They Do, and How to Find Them

A high-risk host cell protein is a residual protein from the production organism that causes harm out of proportion to its concentration. Most host cell proteins (HCPs) are inert at the levels that survive purification. A small number are not. They degrade the product, they degrade the formulation, they act pharmacologically in the patient, or they provoke an immune response, and they do it at concentrations well below any total HCP specification. That is what makes them a separate problem from total HCP control, and why a passing total HCP result is not evidence that you are safe. This page is a catalog. It names the individual proteins the literature has implicated, gives the impact category and mechanism for each, states what kind of evidence stands behind it, and cites the primary source. It then explains how to look for them and what the industry recommends doing when you find one.

Key facts

  • A high-risk host cell protein (HCP) is one that is immunogenic, biologically active, or enzymatically active against the product or an excipient, or that resists removal by purification [1].
  • The BioPhorum Development Group HCP Workstream, a 26-company collaboration, sorts high-risk HCPs into four categories of impact: product quality, formulation, direct biological function in humans, and immunogenicity [1].
  • Proteomic and glycoproteomic analysis of Chinese hamster ovary (CHO) cells identified 6,164 grouped proteins, which is the population a purification process has to clear [28].
  • In the BioPhorum survey, 18 of the 26 member companies responded and around 69% had experienced issues with individual HCPs during drug production [1].
  • Approximately 90% of subjects in lebrikizumab clinical studies developed a specific, measurable immune response to CHO phospholipase B-like 2 (PLBL2) [2].
  • PLBL2 drew the most survey responses of any single HCP, but its specific role in polysorbate degradation is disputed: genetic knockout and immunodepletion of PLBL2 neither diminished nor reduced polysorbate degradation [26].
  • Lysosomal phospholipase A2 (LPLA2) was quantified at less than 1 ppm in three formulated antibodies that showed polysorbate hydrolysis, and was not detected in a fourth that did not [27].

The four impact categories, and why they are the right framework

The industry has already agreed how to classify these proteins, and borrowing that classification is more useful than inventing one. The BioPhorum Development Group HCP Workstream, a collaboration across 26 member companies, published its consensus in Biotechnology and Bioengineering in 2021 as “‘High-risk’ host cell proteins (HCPs): A multi-company collaborative view” [1]. It defines the class as HCPs “that are immunogenic, biologically active, or enzymatically active with the potential to degrade either product molecules or excipients used in formulation”, notes that “some have been shown to be difficult to remove by purification”, and sorts the named proteins into four categories of impact [1].

The basis BioPhorum gives for calling an HCP high risk is its ability to co-purify with the product, how often it is seen in downstream processing, its ability to modify or degrade the drug or the excipient, and its immunogenic potential [1].

Note the third category, because it is the one most often left out. A protein does not have to attack the drug or the formulation to be dangerous. It can simply do in the patient what it does in the hamster. Chemokines and growth factors are potent at picomolar concentrations by design, and a residual one is a pharmacologically active co-administered agent rather than an inert impurity. This is the category that has produced a clinical hold, and it is absent from every competitor page on this topic.

Persistence is the multiplier, not a fifth category

A protein only matters if it survives purification. Persistence is not itself an impact, it is what lets the four impacts reach the patient, and it is why the same short list of names recurs across unrelated programs.

Two routes let an HCP escape clearance. The first is product association: the HCP binds the product through specific or non-specific interaction and is carried through the process with it. Levy and colleagues characterized these interactions directly using cross-interaction chromatography under solution conditions typical of downstream processing, then identified the interacting species by two-dimensional gel electrophoresis and mass spectrometry, showing “both the differences in HCP-mAb interactions among different mAbs, and the relative importance of product association compared to co-elution in protein A affinity chromatography” [29]. The second is interaction with the chromatography medium itself, so the HCP co-elutes with the product from protein A. Clavier and colleagues documented a co-purifying CHO glycosylasparaginase whose “hitchhiking behavior” was supported by in silico surface characterization and which the native digestion protocol failed to detect in final drug substance [30].

The interaction types that drive co-purification are hydrophobic interaction, electrostatic repulsion, hydrogen bonding, Van der Waals forces, ionic interaction, and, for protein A capture specifically, the presence of immunoglobulin-like domains [1].

Two mechanistic results make this concrete. Clusterin binds IgG at both the Fc and the Fab, by a multivalent mechanism, across every human IgG isotype, and aggregated IgG inhibits that binding more strongly than monomer does [31]. And the binding can be driven by the product’s own sequence: an LYY motif in the heavy chain complementarity determining region 2 (CDR2) of an affected antibody was shown by surface plasmon resonance to mediate binding to CHO cathepsin D, and mutating it to AAA abolished the binding [32]. Two of thirteen antibodies tested bound cathepsin D, and only those two carried the motif [32]. Co-purification is therefore partly a property of your molecule, not only of your process.

In rituximab biosimilars, ten HCPs identified as difficult to remove shared “common characteristics of product association, coelution, and age-dependent expression”, including cathepsin D, clusterin, lipoprotein lipase and nidogen-1 [11]. Across seven monoclonal antibody processes from five cell lines, a common set survived into polishing pools: “peroxiredoxin-1, elongation factor 1-alpha 1, actin cytoplasmic 1, putative phospholipase B-like 2, serine protease HTRA1” [13]. For products made in Escherichia coli and purified by immobilized metal affinity chromatography (IMAC), a recurring set of host proteins binds the nickel column directly; of 17 described E. coli IMAC contaminants, 15 were reported to elute above 55 mM imidazole [12].

Scale: why a short list is the only workable approach

Proteomic analysis of CHO cells using the CHO genome identified 6,164 grouped proteins across the proteome and glycoproteome, an eight-fold increase on what was known before, from 120 mass spectrometry analyses generating 682,097 MS/MS spectra [28]. You cannot risk-assess six thousand proteins. You assess the few dozen with a documented reason to worry, which is what the rest of this page is.

The concentrations involved are small

Enzymatic harm does not need much enzyme. Gao and colleagues reported a high-purity human IgG1 from CHO cells losing roughly 60% of its monomer peak after 20 days at 40 degrees Celsius and pH 4, attributed to residual proteolytic activity, with the rate falling to roughly 13% at pH 5 [33]. That is a highly purified drug substance degrading from enzyme activity that no total HCP number flagged. On the formulation side, Li and colleagues state plainly that “high-risk HCPs, such as PSDEs, could have a direct impact on PS degradation at ppb levels” [3].

Named high-risk host cell proteins, by impact category

This is the reference table. Every row carries its BioPhorum impact category and, in its own column, the type of evidence behind it, because “a published case history” and “an in silico prediction” are not the same claim and a risk assessment should not treat them as one. All rows are Chinese hamster ovary (CHO) proteins unless stated. UniProt accessions were taken from live UniProtKB queries restricted to Cricetulus griseus [19]; cells that could not be verified are marked.

Evidence types used: Case history (a published account of harm in a real product), Mechanism study (a published experiment demonstrating the activity, not necessarily in a product), Persistence study (published process data showing the protein survives purification), In silico prediction (a computational immunogenicity score, with no documented clinical consequence), and Industry list (named as high risk by the BioPhorum multi-company collaboration [1]).

Product Quality

Formulation

Direct Biological Function in humans

Immunogenicity

The prediction tools behind those three rows are CHOPPI, a web tool that integrates the likelihood a CHO protein is expressed and secreted with characterizations of its immunogenicity (T cell epitope count and density, and conservation with human counterparts) [24], and the Immune Epitope Database. Read them for what they are. CHOPPI’s authors describe it as “a valuable computational complement to existing experimental approaches for CHOP risk assessment” that “can focus experimental efforts in the most important directions” [24]. A high score is a reason to look, not a finding of harm. Saying so is not a weakness in the list; it is the difference between a risk assessment and a rumor.

Persistence: proteins listed mainly because they will not leave

Also named by BioPhorum, with no primary harm citation located

These proteins appear on the BioPhorum high-risk list [1] and this page includes them for completeness, but no primary publication documenting product harm was found for any of them. They are monitoring candidates, not documented offenders, and the table says so rather than padding the rows above.

Named often, but with weaker primary evidence

One more name appears on nearly every problematic-HCP list and deserves monitoring without having the published case history the rows above have. It is worth saying so.

PLBL2 and polysorbate: the disagreement worth knowing about

PLBL2 is the most reported problem host cell protein in the industry, and part of the story commonly told about it is contested in the published literature. Both halves matter, and pages that give only the first half are giving you half a risk assessment.

What is not in dispute. PLBL2 co-purifies with monoclonal antibodies and is difficult to remove. It drew the most responses in the BioPhorum survey, both for detection in final product and for demonstrated clearance [1]. And its immunogenicity in a real clinical program is documented: after initial phase III studies, lebrikizumab material was found to contain CHO PLBL2, approximately 90% of subjects developed a specific and measurable immune response to it, and material with substantially reduced PLBL2 produced significantly less frequent, dose-dependent responses [2]. That is a case history, not a prediction.

What is disputed. PLBL2 is frequently described as a cause of polysorbate degradation, on the basis of a 2016 proposal that it was the residual host cell protein degrading polysorbate 20 in a drug formulation. Zhang and colleagues tested that directly and could not reproduce it. Their paper spells the protein PLBD2, which is the same protein under a different gene-name convention. Using genetic knockout and immunodepletion in recombinant CHO material, they report that “when PLBD2 was removed or depleted, the degradation of PS20 or PS80 was neither diminished nor reduced”, and that quantitative analysis across multiple formulated monoclonal antibody products “did not establish a correlation between the amount of PLBD2 and the level of PS20 degradation”. Their conclusion is that PLBL2 “is not the primary cause of polysorbate degradation in formulated drug products purified using standard Protein A and ion exchange chromatography” [26].

What that means for your risk assessment. Keep PLBL2 on the list, because the immunogenicity evidence and the clearance difficulty are both solid. Do not assume that clearing PLBL2 protects your polysorbate. The lipases with the clearest polysorbate evidence are different proteins:

  • Lipoprotein lipase (LPL). Recombinant LPL degrades polysorbate 80 and polysorbate 20 under formulation-relevant conditions, and knocking Lpl out of CHO with CRISPR and TALEN gave harvest fluid with significantly reduced polysorbate degradation and no significant viability cost [14]. That is a causal experiment in both directions.
  • Lysosomal phospholipase A2 (LPLA2). Endogenous LPLA2 was quantitated at less than 1 ppm in three formulated antibodies, and all three showed polysorbate hydrolysis; a fourth antibody with LPLA2 below 0.1 ppm did not [27]. Recombinant LPLA2 hydrolyzed polysorbate 20 and polysorbate 80 monoester and produced degradation profiles matching those from the formulated product [27].

That sub-1 ppm figure is the single most useful number on this page. It is roughly two orders of magnitude below a typical total HCP acceptance limit, and it comes with a matched negative control in the same study.

What 18 companies actually reported

Case reports tell you what can happen. A survey tells you what does happen, and the BioPhorum workstream ran one across its membership. Of 26 member companies, 18 responded [1]. The findings below are industry-wide rather than vendor claims, which is what makes them worth quoting in a risk assessment.

Two numbers deserve a second look. The first is the 40% clearance rate for the write-in HCPs: when companies met something outside the usual list, most of the time they did not get rid of it. The second is that the counts for identifying individual HCPs exceed 18, which means companies are running mass spectrometry and protein-specific ELISA together rather than choosing between them. That is the correct reading of the two methods, and it is the subject of the companion page on why ELISA and LC-MS HCP results disagree.

Independent corroboration: the nine HCPs USP makes standards for

A list of high-risk host cell proteins is only as good as its sources, and most published lists share authors. One check on this table comes from a body with no product to sell and no case to make: the United States Pharmacopeia decides which host cell proteins are worth the cost of producing a characterized measurement standard, and it has now done so for nine named Chinese hamster ovary (CHO) proteins.

USP catalogs stable isotope labeled (SIL) heavy peptide analytical reference materials, the reagents used for targeted quantitation under USP General Chapter <1132.1> Method C, for the following CHO host cell proteins. Our page on running USP <1132.1> Methods A, B and C explains what that method involves.

USP also supplies purified recombinant CHO PLBL2 protein (catalog number 1582716) and CHO Null Cell Harvest Cell Culture Fluid (1544913) as a spiking matrix [20]. Read the category carefully: USP states that analytical reference materials “are different from USP Reference Standards and not required for compendial compliance” [20]. Buying one does not make a method compendial. It does give you a characterized, catalog-numbered reagent instead of a custom synthesis, which is the practical difference that decides whether a targeted method gets built.

Seven of the nine sit somewhere in the main table above, which is a useful independent agreement between what the literature documents, what 26 companies agreed on, and what the pharmacopeia thinks is worth standardizing. Two observations follow.

Protein disulfide isomerase. An earlier version of this page placed PDI in a weak-evidence table on the grounds that no published harm case existed. That was wrong, and it is corrected here. BioPhorum lists PDI as high risk under Product Quality with the impact given as reduction of disulfide bonds [1], and there is a primary mechanism study behind it: Maeda and colleagues showed that PDI-catalyzed reduction of insulin disulfide bonds promotes aggregation, correlating the kinetics of reduction with the kinetics of aggregate growth and characterizing the aggregates as antiparallel beta-sheet [22]. That is not a case report from a licensed biologic, and the table’s evidence column says “mechanism study” rather than “case history” for exactly that reason. But it is published, mechanistic evidence of the activity BioPhorum names, which is a different thing from no evidence at all.

Glutathione-S-transferase P1 and peroxiredoxin 1. Both are now in the main table, under Immunogenicity, and both carry an explicit evidence type of in silico prediction with no documented clinical consequence. That wording is deliberate. The reason they are on the high-risk list is that sequence-based algorithms flagged them, along with PLBL2 and procollagen-lysine 2-oxoglutarate 5-dioxygenase 1, as having high immunogenicity scores [23][24]. The same study then tested the prediction in an in vitro peripheral blood mononuclear cell assay and found that samples with up to 4,000 ppm HCP content, 200 times the level in drug substance, did not pose a higher immunogenicity risk than highly purified material [23]. A page that promoted these proteins to “documented offender” would be misrepresenting their own source. A page that dropped them would be hiding names the industry list carries. Naming them and labeling the evidence type is the only honest option, and it is the one taken here.

The wider point is the one to take into a risk assessment. The proteins worth targeting are a short, stable, named list, and three independent sources now converge on much the same names: the BioPhorum multi-company list, the published case literature, and the reagents USP chose to produce.

Polysorbate degradation: the case study

Polysorbate degradation is the mechanism that best explains why total HCP control is not enough, so it is worth working through.

Polysorbate 20 and polysorbate 80 are non-ionic surfactants built from fatty acid esters of polyoxyethylene sorbitan. They keep the protein off air-liquid and solid-liquid interfaces. A residual lipase or esterase hydrolyzes the ester bond, releasing free fatty acids such as lauric acid from polysorbate 20 and oleic acid from polysorbate 80. Two things then happen at once. The surfactant concentration falls below the level needed to protect the protein, and the liberated free fatty acids, which are poorly soluble, nucleate into visible or subvisible particulates. The product can therefore fail both a polysorbate content assay and a particulate matter test without the protein itself having changed.

Three features make this hard to catch.

It appears late. Enzymatic hydrolysis is slow at 2 to 8 degrees Celsius. A formulation can pass release, pass three months, and fail at twelve or twenty-four. Work on predicting long-term polysorbate degradation from short-term kinetics exists precisely because the problem shows up at the end of a stability program rather than the start [16].

The enzyme concentration is trivially small. A typical total HCP acceptance limit sits in the region of 100 ng of HCP per mg of product. Polysorbate-degrading enzymes act far below that. Li and colleagues state plainly that “high-risk HCPs, such as PSDEs, could have a direct impact on PS degradation at ppb levels” [3]. In the primary study, LPLA2 was quantitated at less than 1 ppm in three antibodies that showed polysorbate hydrolysis and was below 0.1 ppm in a fourth that did not [27]; a later review puts the endogenous levels in affected material at roughly 0.3 to 0.6 ppm, and reports sialate O-acetylesterase showing strong activity below 5 ppm [3]. A batch at 20 ppm total HCP, comfortably inside specification, can contain enough lipase to strip the surfactant.

How HCP ppm is calculated from LC-MS data

It is hard to see by mass spectrometry. The same paper notes that “despite their high biological activity, the concentrations of most PSDEs are so low that they are practically undetectable by MS without fractionation and enrichment” [3]. This is a real constraint and it shapes the method design discussed below.

The catalog of polysorbate-degrading enzymes now includes lipoprotein lipase, lysosomal phospholipase A2, liver carboxylesterase, lysosomal acid lipase, palmitoyl-protein thioesterase 1, sialate O-acetylesterase and phospholipase A2 group VII [3]. Recombinant expression of CHO hydrolases has been used to test which of them actually degrade polysorbate rather than merely being annotated as lipases [15]. Cell line engineering is the definitive fix: knocking out Lpl in CHO improved polysorbate stability in monoclonal antibody formulations [14].

The wider regulatory picture

The BioPhorum workstream is ongoing rather than finished. Its stated goal is “industry alignment; building a common understanding of agency requirements for HCPs through benchmarking and gap analysis of guidance” [17], so the 2021 paper [1] is a snapshot of a moving consensus, not a closed list. Treat it as the reference to cite in a risk assessment and expect it to grow.

A more recent survey extends the picture. Graham and colleagues, writing for the IQ DruSafe Impurities Safety Working Group, published “Assessment and Control of Host Cell Proteins in Biologics: Survey of Industry Practices and a Vision for Harmonization” in Biotechnology and Bioengineering in January 2026 [18]. It reports current default limits for total and individual HCP impurities, approaches to safety and immunogenicity risk assessment, and feedback from global health authorities. The authors describe “both perceived risks and experienced impact from HCP impurities” and conclude that a harmonized approach may be justified in specific areas [18].

Why HCP ELISA usually misses these

An HCP ELISA reports total immunoreactivity against a polyclonal anti-HCP reagent. It returns one number for the whole mixture. Three properties of that design work against high-risk HCP detection.

First, the reagent only sees what the immunized animal responded to. If a protein was scarce or weakly immunogenic in the null-cell-line immunogen, there is little or no antibody against it, and it contributes little or no signal regardless of how much is present.

Second, even a protein that is well covered contributes in proportion to its mass, not its activity. A lipase at 0.5 ppm inside a 30 ppm total is arithmetically invisible.

Third, the assay does not name anything. A result of 30 ppm tells you nothing about which proteins make up the 30 ppm, so it cannot trigger a risk decision on its own.

None of this makes ELISA a bad assay. It remains the lot release and process consistency method. It is the wrong tool for this specific question. The two methods answer different questions and routinely give different numbers, which is covered in detail in the companion page on why ELISA and LC-MS HCP results disagree.

How you find them: risk-based monitoring

The workable approach is identify first, then assess. Do not try to build a method that only detects dangerous proteins. Build a method that detects as many proteins as possible, then filter the output against a curated risk list.

That ordering matters because the two goals pull in opposite directions. A conservative identification threshold suppresses false positives, but it also suppresses the low-abundance enzymes that are the entire point of the exercise. A sensitive threshold surfaces the enzymes and also surfaces noise. If you are going to filter afterward against a list of known offenders, you can afford to run sensitive, because a spurious identification of a ribosomal protein is harmless while a missed lipase is not.

In practice:

  1. Run a sensitive identification pass on drug substance using LC-MS, searching the full host organism proteome rather than a reduced list.
  2. Filter the identified proteins against a curated list of HCPs with documented harm, such as the BioPhorum high-risk list [1] and the polysorbate-degrading enzyme list [3].
  3. Rank the hits by impact category and by relevance to your molecule. A lipase matters if you formulate in polysorbate. A protease matters if your molecule has an exposed hinge or linker. A cytokine matters regardless of what you are making.
  4. Confirm any hit that would trigger action using an orthogonal method before acting on it.
  5. Carry the same filtered list forward as the monitoring panel for subsequent batches and process changes.

This is the design SpotMap MS, TotalLab’s LC-MS HCP analysis software, implements: it can “reduce the output of an analysis to feature only those HCPs” identified in “peer-reviewed publications” as “potentially dangerous” and assign “a threat level to them”, using a spectral library that carries an associated HCP threat level. The general point stands whatever software you use. Identification and risk assessment are two steps, and collapsing them into one costs you sensitivity.

Where an enzyme is too dilute for identification even at a sensitive threshold, enrichment or activity-based detection is the route. Li and colleagues are explicit that most polysorbate-degrading enzymes are “practically undetectable by MS without fractionation and enrichment” [3].

Impact categories mapped to monitoring

What to do once you find one: BioPhorum’s recommended sequence

Finding a named high-risk HCP is not a failure. It is information, and the industry has published a sequence for acting on it. BioPhorum’s recommendation is a case-by-case risk assessment for any HCP that appears on the high-risk list, run in a specific order [1]. The order is the point: cheap and general first, expensive and specific last.

  1. Literature survey. Start with what is already known about that protein. Has it been reported in a product before, at what level, with what consequence, and was it cleared? The tables above are a starting set; the primary citations are where the detail is.
  2. LC-MS/MS HCP profiling. Find out what is actually in your drug substance, by name, without pre-committing to a target list. This is the step that turns an unbounded question into a bounded one.
  3. Individual HCP quantification. Once a protein is named, measure it. A targeted mass spectrometry assay, multiple reaction monitoring or parallel reaction monitoring, monitors two or three peptides specific to that protein and gives far better sensitivity and precision than a survey scan.
  4. Enzymatic activity assays, where the category is Product Quality or Formulation. Presence and activity are different questions and only the second one degrades anything. For polysorbate-degrading enzymes especially, activity is what matters.
  5. In silico immunogenicity assessment, where the category is Direct Biological Function in humans or Immunogenicity. Tools such as CHOPPI score a sequence for foreign T cell epitope content and for conservation with the human counterpart [24].
  6. In vitro comparative immunogenicity assessment (IVCIA), to test what the in silico step predicted. This is the step that separates a high score from a real response, and in the published example it changed the answer: predicted-high HCPs did not stimulate a response at the concentrations present in in-process material [23].

A point from the same paper that most vendor pages leave out: individual HCP ELISAs are powerful, but few are commercially available, and developing one is a significant investment of time and resource [1]. That is a real constraint on a control strategy, not a footnote. BioPhorum’s suggested resolution is the sequence above run in the order given, developing individual HCP ELISAs for the HCPs frequently seen in drug substance based on LC-MS/MS analysis [1]. In other words, let the mass spectrometry decide which two or three assays are worth building, rather than guessing which kits to buy.

That is also the argument this page has been making: identify first, then target. It is worth saying plainly that the sequence is not ours. It is BioPhorum’s, published in 2021 across 26 companies, and it is the one to cite in a control strategy document [1].

Then act on the result

  1. Confirm it. A single identification is a hypothesis. Confirm by an orthogonal route: a second injection with different chromatography, a protein-specific ELISA if one exists, or an activity assay that shows the enzyme is functional and not merely present.
  2. Establish whether it is doing harm. Spiking is the direct test. Cathepsin D’s role in particle formation was confirmed by spiking experiments [4]. If a spike of the purified enzyme into your formulation reproduces the defect, you have causation rather than correlation.
  3. Change the process first. Removal beats specification. Cathepsin D was cleared by a high-pH wash with salt and caprylate during protein A chromatography, which disrupted the protein-protease interaction and gave material with no detectable protease activity and no particles [4]. Cathepsin L was cleared by mixed-mode chromatography across a wide parameter range [6]. For persistent lipases, cell line engineering is available: the Lpl knockout is the published precedent [14], and the carboxypeptidase D knockout is the precedent for a product quality attribute rather than a formulation one [25].
  4. Set a specification only where a process change cannot close the gap. An individual HCP limit needs a justified basis: the level at which harm was observed, the level your process delivers, and the analytical capability to measure it. Industry default limits for individual HCPs are surveyed in Graham et al. [18].
  5. Trend it. One batch is a data point. The question a regulator will ask is whether the level is stable across batches, scales and campaigns, which requires longitudinal tracking rather than repeated one-off reports. TrendLab is TotalLab’s longitudinal trending application for exactly that job.

Getting this into your own lab

Risk-based HCP monitoring is not an exotic capability. It needs an LC-MS method with adequate depth, a curated list of proteins worth worrying about, and software that applies the second to the first without hiding the low-abundance hits.

SpotMap MS is TotalLab’s LC-MS HCP analysis software. It runs data-independent acquisition data against your FASTA file, flags HCPs that peer-reviewed publications identify as potentially dangerous, and assigns a threat level. It is compatible with TotalLab’s AuditSafe software for 21 CFR Part 11 and GMP compliance. Request a free trial or talk to us about your molecule.

Frequently asked questions

Q: What is the most problematic host cell protein?
A: Phospholipase B-like 2 (PLBL2) drew the most responses in the BioPhorum survey of 18 companies, and it is the most documented: approximately 90% of lebrikizumab subjects developed a specific immune response to it. Lipoprotein lipase and lysosomal phospholipase A2 carry the clearest polysorbate evidence, and cathepsin D has caused more product degradation. There is no single worst HCP. The one that matters is the one your process fails to remove.

Q: Which HCPs degrade polysorbate?
A: The named polysorbate-degrading enzymes are lipoprotein lipase (LPL), lysosomal phospholipase A2 group XV (LPLA2, PLA2G15), liver carboxylesterase (CES1 family), lysosomal acid lipase (LIPA), palmitoyl-protein thioesterase 1 (PPT1), sialate O-acetylesterase (SIAE) and phospholipase A2 group VII (PLA2G7). They hydrolyze the fatty acid ester bonds in polysorbate 20 and 80, releasing free fatty acids that form particulates.

Q: Does HCP ELISA detect PLBL2?
A: Not reliably in a total HCP ELISA. A polyclonal anti-HCP reagent only responds to proteins that were immunogenic in the null-cell-line immunogen, and PLBL2 contributes to the total signal by mass, not by risk. Once PLBL2 is identified as a concern, PLBL2-specific ELISA kits are commercially available and are the appropriate monitoring tool. Detection and identification are separate jobs.

Q: What does high risk mean for a host cell protein?
A: It means the protein is immunogenic, biologically active, or enzymatically active against the product or an excipient, or that it resists removal by purification. That definition comes from the BioPhorum multi-company collaboration published in Biotechnology and Bioengineering in 2021, which sorts high-risk HCPs into four impact categories: product quality, formulation, direct biological function in humans, and immunogenicity. Risk is a property of the protein and of your product, not of the concentration alone.

Q: Can an HCP cause a problem below the total HCP specification?
A: Yes, routinely. Polysorbate-degrading enzymes act at parts-per-billion levels. Lysosomal phospholipase A2 was quantitated at less than 1 ppm in three antibodies that showed polysorbate hydrolysis, and below 0.1 ppm in a fourth that did not. A batch reporting 20 ppm total HCP, well inside a typical 100 ppm limit, can still contain enough lipase to strip the surfactant over shelf life.

Q: Why does polysorbate degradation only show up late in stability?
A: Enzymatic hydrolysis at 2 to 8 degrees Celsius is slow. The surfactant concentration falls gradually and the released free fatty acids need time to nucleate into detectable particles. A formulation can pass release and three-month testing and fail at twelve or twenty-four months. This is why short-term degradation kinetics are used to forecast long-term behavior rather than waiting for the timepoint.

Q: Does PLBL2 degrade polysorbate?
A: It is disputed. PLBL2 was proposed as a polysorbate 20 degrader in 2016, but Zhang and colleagues reported in 2020 that genetic knockout and immunodepletion of PLBL2 left polysorbate 20 and 80 degradation neither diminished nor reduced, with no correlation between PLBL2 amount and degradation across formulated products. PLBL2 remains a documented immunogenicity and clearance concern. For polysorbate specifically, lipoprotein lipase and lysosomal phospholipase A2 have the stronger evidence.

Q: How do I know which high-risk HCPs to look for in my process?
A: Start from the published catalogs: the BioPhorum high-risk HCP list and the polysorbate-degrading enzyme list. Filter by your expression system, then by impact category. Lipases and esterases matter if you formulate in polysorbate. Proteases and carboxypeptidases matter if your molecule has an exposed hinge, linker, tag or a C-terminal lysine. Bioactive HCPs and immunogenic HCPs matter most for chronic dosing and subcutaneous administration.

Q: What should I do if LC-MS identifies a high-risk HCP in my drug substance?
A: Follow BioPhorum’s sequence: literature survey, LC-MS/MS profiling, individual quantification, then activity assays for product quality and formulation impact, or in silico and in vitro comparative immunogenicity assessment for biological function and immunogenicity. If harm is confirmed, change the process before you change the specification, then move to a targeted MRM assay for routine monitoring and trend the result across batches.

References

  • Jones, M., Palackal, N., Wang, F., Gaza-Bulseco, G., Hurkmans, K., Zhao, Y., Chitikila, C., Clavier, S., Liu, S., Menesale, E., Schonenbach, N.S., Sharma, S., Valax, P., Waerner, T., Zhang, L., Connolly, T. “‘High-risk’ host cell proteins (HCPs): A multi-company collaborative view.” Biotechnology and Bioengineering, 118(8):2870-2885, 2021. https://doi.org/10.1002/bit.27808 (PubMed: https://pubmed.ncbi.nlm.nih.gov/33930190/)
  • Fischer, S.K., Cheu, M., Peng, K., Lowe, J., Araujo, J., Murray, E., McClintock, D., Matthews, J., Siguenza, P., Song, A. “Specific Immune Response to Phospholipase B-Like 2 Protein, a Host Cell Impurity in Lebrikizumab Clinical Material.” The AAPS Journal, 19(1):254-263, 2017. https://link.springer.com/article/10.1208/s12248-016-9998-7
  • Li, X., Wang, F., Li, H., Richardson, D.D., Roush, D.J. “The measurement and control of high-risk host cell proteins for polysorbate degradation in biologics formulation.” Antibody Therapeutics, 5(1):42-54, 2022. https://academic.oup.com/abt/article/5/1/42/6508676
  • Bee, J.S., Tie, L., Johnson, D., Dimitrova, M.N., Jusino, K.C., Afdahl, C.D. “Trace levels of the CHO host cell protease cathepsin D caused particle formation in a monoclonal antibody product.” Biotechnology Progress, 31(5):1360-1369, 2015. https://doi.org/10.1002/btpr.2150
  • Robert, F., Bierau, H., Rossi, M., Agugiaro, D., Soranzo, T., Broly, H., Mitchell-Logean, C. “Degradation of an Fc-fusion recombinant protein by host cell proteases: Identification of a CHO cathepsin D protease.” Biotechnology and Bioengineering, 104(6):1132-1141, 2009. https://doi.org/10.1002/bit.22494
  • Luo, H., Tie, L., Cao, M., Hunter, A.K., Pabst, T.M., Du, J., Field, R., Li, Y., Wang, W.K. “Cathepsin L Causes Proteolytic Cleavage of Chinese-Hamster-Ovary Cell Expressed Proteins During Processing and Storage: Identification, Characterization, and Mitigation.” Biotechnology Progress, 35(1):e2732, 2019. https://doi.org/10.1002/btpr.2732
  • Kao, Y.-H., Hewitt, D.P., Trexler-Schmidt, M., Laird, M.W. “Mechanism of antibody reduction in cell culture production processes.” Biotechnology and Bioengineering, 107(4):622-632, 2010. https://doi.org/10.1002/bit.22848
  • Handlogten, M.W., Zhu, M., Ahuja, S. “Glutathione and thioredoxin systems contribute to recombinant monoclonal antibody interchain disulfide bond reduction during bioprocessing.” Biotechnology and Bioengineering, 114(7):1469-1477, 2017. https://doi.org/10.1002/bit.26278
  • Rane, S.S., Dearman, R.J., Kimber, I., Uddin, S., Bishop, S., Shah, M., Podmore, A., Pluen, A., Derrick, J.P. “Impact of a Heat Shock Protein Impurity on the Immunogenicity of Biotherapeutic Monoclonal Antibodies.” Pharmaceutical Research, 36(4):51, 2019. https://link.springer.com/article/10.1007/s11095-019-2586-7
  • Hajam, I.A., Dar, P.A., Shahnawaz, I., Jaume, J.C., Lee, J.H. “Bacterial flagellin, a potent immunomodulatory agent.” Experimental & Molecular Medicine, 49:e373, 2017. https://www.nature.com/articles/emm2017172
  • Singh, S.K., Mishra, A., Yadav, D., Budholiya, N., Rathore, A.S. “Understanding the mechanism of copurification of ‘difficult to remove’ host cell proteins in rituximab biosimilar products.” Biotechnology Progress, 36(2):e2936, 2020. https://doi.org/10.1002/btpr.2936
  • Robichon, C., Luo, J., Causey, T.B., Benner, J.S., Samuelson, J.C. “Engineering Escherichia coli BL21(DE3) Derivative Strains To Minimize E. coli Protein Contamination after Purification by Immobilized Metal Affinity Chromatography.” Applied and Environmental Microbiology, 77(13):4634-4646, 2011. https://journals.asm.org/doi/10.1128/aem.00119-11
  • Oh, Y.H., Mendola, K.M., Choe, L.H., Min, L., Lavoie, A.R., Sripada, S.A., Williams, T.I., Lee, K.H., Yigzaw, Y., Seay, A., Bill, J., Li, X., Roush, D.J., Cramer, S.M., Menegatti, S., Lenhoff, A.M. “Identification and characterization of CHO host-cell proteins in monoclonal antibody bioprocessing.” Biotechnology and Bioengineering, 121(1):291-305, 2024. https://pmc.ncbi.nlm.nih.gov/articles/PMC10842603/
  • Chiu, J., Valente, K.N., Levy, N.E., Min, L., Lenhoff, A.M., Lee, K.H. “Knockout of a difficult-to-remove CHO host cell protein, lipoprotein lipase, for improved polysorbate stability in monoclonal antibody formulations.” Biotechnology and Bioengineering, 114(5):1006-1015, 2017. https://doi.org/10.1002/bit.26237
  • Kovner, D., Yuk, I.H., Shen, A., Li, H., Graf, T., Gupta, S.K., Liu, W., Tomlinson, A.A.G. “Characterization of Recombinantly-Expressed Hydrolytic Enzymes from Chinese Hamster Ovary Cells: Identification of Host Cell Proteins that Degrade Polysorbate.” Journal of Pharmaceutical Sciences, 112(5):1351-1363, 2023. https://doi.org/10.1016/j.xphs.2023.01.003
  • Zhang, S., Riccardi, C., Carlson, D., Kamen, D., Graham, K.S., Shameem, M., Bak, H., Xiao, H., Li, N. “Prediction of long-term polysorbate degradation according to short-term degradation kinetics.” mAbs, 15(1):2232486, 2023. https://doi.org/10.1080/19420862.2023.2232486
  • BioPhorum. “Host cell proteins workstream.” Accessed September 2026. https://www.biophorum.com/workstream/host-cell-proteins/
  • Graham, J.C., Anand, S.S., Bercu, J., Besenhofer, L., de Zafra, C., Feng, Y., Glowienke, S., Hillegass, J., Hutchinson, R., Jolly, R., Masuda-Herrera, M.J., Nicholas, T., Olszova, D., Schmitz, M., Semmelmann, F., Tien, E. “Assessment and Control of Host Cell Proteins in Biologics: Survey of Industry Practices and a Vision for Harmonization.” Biotechnology and Bioengineering, 123:1050-1063, 2026 (published 9 January 2026). https://doi.org/10.1002/bit.70154
  • UniProt Consortium. UniProtKB entries for Cricetulus griseus (Chinese hamster), accessions G3I6T1, A0A3L7IKX6, G3HKV9, G3IIB1, G3HN89, G3HQY6, G3I4W7, G3INC5, A0A8C2MY60, G3HNJ3. Accessed September 2026. https://www.uniprot.org/
  • United States Pharmacopeial Convention. Monoclonal Antibody (mAb)-related USP-NF Standards and Analytical Reference Materials. 7 October 2025. Source for the nine CHO HCP heavy peptide catalog numbers, the recombinant PLBL2 protein (1582716) and the CHO Null Cell Harvest Cell Culture Fluid (1544913). https://www.usp.org/sites/default/files/usp-bio_monoclonalantibodies-standards-materials_07oct2025.pdf
  • United States Pharmacopeial Convention. General Chapter <1132> Residual Host Cell Protein Measurement in Biopharmaceuticals. USP 39 and NF 34, official 1 May 2016. Free full-text PDF posted by USP. Note that this free copy is the 2016 version and that no Notice of Intent to Revise has been published for <1132> since it became official. https://www.usp.org/sites/default/files/usp/document/our-work/biologics/USPNF810G-GC-1132-2017-01.pdf
  • Maeda, R., Ado, K., Takeda, N., Taniguchi, Y. “Promotion of insulin aggregation by protein disulfide isomerase.” Biochimica et Biophysica Acta, 1774(12):1619-1627, 2007. DOI 10.1016/j.bbapap.2007.08.016, PMID 17920002. https://doi.org/10.1016/j.bbapap.2007.08.016
  • Jawa, V., Joubert, M.K., Zhang, Q., Deshpande, M., Hapuarachchi, S., Hall, M.P., Flynn, G.C. “Evaluating Immunogenicity Risk Due to Host Cell Protein Impurities in Antibody-Based Biotherapeutics.” The AAPS Journal, 18(6):1439-1452, 2016. DOI 10.1208/s12248-016-9948-4, PMID 27450229. https://doi.org/10.1208/s12248-016-9948-4
  • Bailey-Kellogg, C., Gutierrez, A.H., Moise, L., Terry, F., Martin, W.D., De Groot, A.S. “CHOPPI: a web tool for the analysis of immunogenicity risk from host cell proteins in CHO-based protein production.” Biotechnology and Bioengineering, 111(11):2170-2182, 2014. DOI 10.1002/bit.25286, PMID 24888712. https://doi.org/10.1002/bit.25286
  • Hu, Z., Zhang, H., Haley, B., Macchi, F., Yang, F., Misaghi, S., Elich, J., Yang, R., Tang, Y., Joly, J.C., Snedecor, B.R., Shen, A. “Carboxypeptidase D is the only enzyme responsible for antibody C-terminal lysine cleavage in Chinese hamster ovary (CHO) cells.” Biotechnology and Bioengineering, 113(10):2100-2106, 2016. DOI 10.1002/bit.25977, PMID 26989081. https://doi.org/10.1002/bit.25977
  • Zhang, S., Xiao, H., Goren, M., Burakov, D., Chen, G., Li, N., Tustian, A., Adams, B., Mattila, J., Bak, H. “Putative Phospholipase B-Like 2 is Not Responsible for Polysorbate Degradation in Monoclonal Antibody Drug Products.” Journal of Pharmaceutical Sciences, 109(9):2710-2718, 2020. DOI 10.1016/j.xphs.2020.05.028, PMID 32534029. https://doi.org/10.1016/j.xphs.2020.05.028
  • Hall, T., Sandefur, S.L., Frye, C.C., Tuley, T.L., Huang, L. “Polysorbates 20 and 80 Degradation by Group XV Lysosomal Phospholipase A2 Isomer X1 in Monoclonal Antibody Formulations.” Journal of Pharmaceutical Sciences, 105(5):1633-1642, 2016. DOI 10.1016/j.xphs.2016.02.022, PMID 27056628. https://doi.org/10.1016/j.xphs.2016.02.022
  • Baycin-Hizal, D., Tabb, D.L., Chaerkady, R., Chen, L., Lewis, N.E., Nagarajan, H., Sarkaria, V., Kumar, A., Wolozny, D., Colao, J., Jacobson, E., Tian, Y., O’Meally, R.N., Krag, S.S., Cole, R.N., Palsson, B.O., Zhang, H., Betenbaugh, M. “Proteomic analysis of Chinese hamster ovary cells.” Journal of Proteome Research, 11(11):5265-5276, 2012. DOI 10.1021/pr300476w, PMID 22971049. https://doi.org/10.1021/pr300476w
  • Levy, N.E., Valente, K.N., Choe, L.H., Lee, K.H., Lenhoff, A.M. “Identification and characterization of host cell protein product-associated impurities in monoclonal antibody bioprocessing.” Biotechnology and Bioengineering, 111(5):904-912, 2014. DOI 10.1002/bit.25158, PMID 24254318. https://doi.org/10.1002/bit.25158
  • Clavier, S., Fougeron, D., Petrovic, S., Elmaleh, H., Fourneaux, C., Bugnazet, D., Duffieux, F., Masiero, A., Mitra-Kaushik, S., Genet, B., Fromentin, Y., Kreiss, P., Laborderie, B., Brault, D., Menet, J.-M. “Improving the analytical toolbox to investigate copurifying host cell proteins presence: N-(4)-(beta-acetylglucosaminyl)-L-asparaginase case study.” Biotechnology and Bioengineering, 117(11):3368-3378, 2020. DOI 10.1002/bit.27514, PMID 32706388. https://doi.org/10.1002/bit.27514
  • Wilson, M.R., Easterbrook-Smith, S.B. “Clusterin binds by a multivalent mechanism to the Fc and Fab regions of IgG.” Biochimica et Biophysica Acta, 1159(3):319-326, 1992. DOI 10.1016/0167-4838(92)90062-I, PMID 1390937. https://doi.org/10.1016/0167-4838(92)90062-i
  • Bee, J.S., Machiesky, L.M., Peng, L., Jusino, K.C., Dickson, M., Gill, J., Johnson, D., Lin, H.-Y., Miller, K., Heidbrink Thompson, J., Remmele, R.L. “Identification of an IgG CDR sequence contributing to co-purification of the host cell protease cathepsin D.” Biotechnology Progress, 33(1):140-145, 2017. DOI 10.1002/btpr.2397, PMID 27798957. https://doi.org/10.1002/btpr.2397
  • Gao, S.X., Zhang, Y., Stansberry-Perkins, K., Buko, A., Bai, S., Nguyen, V., Brader, M.L. “Fragmentation of a highly purified monoclonal antibody attributed to residual CHO cell protease activity.” Biotechnology and Bioengineering, 108(4):977-982, 2011. DOI 10.1002/bit.22982, PMID 21404269. https://doi.org/10.1002/bit.22982
  • Beatson, R., Sproviero, D., Maher, J., Wilkie, S., Taylor-Papadimitriou, J., Burchell, J.M. “Transforming growth factor-beta 1 is constitutively secreted by Chinese hamster ovary cells and is functional in human cells.” Biotechnology and Bioengineering, 108(11):2759-2764, 2011. DOI 10.1002/bit.23217, PMID 21618471. https://doi.org/10.1002/bit.23217