Host Cell Protein Profiling by Targeted and Untargeted Analysis of

Apr 12, 2017 - Untargeted data processing with DIA-Umpire provided a means of identifying HCPs not represented in the assay library used for targeted,...
0 downloads 12 Views 997KB Size
Subscriber access provided by OKLAHOMA STATE UNIV

Article

Host Cell Protein Profiling by Targeted and Untargeted Analysis of Data Independent Acquisition Mass Spectrometry Data with Parallel Reaction Monitoring Verification Simion Kreimer, Yuanwei Gao, Somak Ray, Mi Jin, Zhijun Tan, Nesredin A. Mussa, Li Tao, Zhengjian Li, Alexander R. Ivanov, and Barry L. Karger Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.6b04892 • Publication Date (Web): 12 Apr 2017 Downloaded from http://pubs.acs.org on April 17, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Host Cell Protein Profiling by Targeted and Untargeted Analysis of Data Independent Acquisition Mass Spectrometry Data with Parallel Reaction Monitoring Verification Simion Kreimer1, Yuanwei Gao1, Somak Ray1, Mi Jin2,3, Zhijun Tan2, Nesredin A. Mussa2, Li Tao2, Zhengjian Li2, Alexander R. Ivanov1, and Barry L. Karger1* 1) Barnett Institute and Department of Chemistry and Chemical Biology, Northeastern University, Boston, MA, 02115. 2) Bristol-Myers Squibb, Biologics Process and Product Development, 38 Jackson Road, Devens, MA 01434. 3) Present address. TEVA Biopharmaceuticals, 145 Brandywine Highway, West Chester, PA 19380. *Author for inquiries.

[email protected]

ABSTRACT Host cell proteins (HCPs) are process-related impurities of biopharmaceuticals that remain at trace levels despite multiple stages of downstream purification. Currently, there is interest in implementing LC-MS in biopharmaceutical HCP profiling alongside conventional ELISA, because individual species can be identified and quantitated. Conventional datadependent LC-MS is hampered by the low concentration of HCP-derived peptides, which are 56 orders of magnitude less abundant than the biopharmaceutical-derived peptides. In this paper, we present a novel data independent acquisition (DIA)-MS workflow to identify HCP peptides using automatically combined targeted and untargeted data processing, followed by verification and quantitation using parallel reaction monitoring (PRM). Untargeted data processing with DIA-Umpire provided a means of identifying HCPs not represented in the assay library used for targeted, peptide-centric, data analysis. An IgG1 monoclonal antibody (mAb) purified by Protein A column elution, cation exchange chromatography, and ultrafiltration was analyzed using the workflow with 1D-LC. Five protein standards added at 0.5 to 100 ppm concentrations were detected in the background of the purified mAb, demonstrating sensitivity to low ppm levels. A calibration curve was constructed based on the summed peak areas of the three highest intensity fragment ions from the highest intensity peptide of each protein standard. 16 HCPs were identified and quantitated based on the calibration curve over the range of low ppm to over 100 ppm in the purified mAb sample. The developed approach

1

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

achieves rapid HCP profiling using by 1D-LC and specific identification exploiting the high mass accuracy and resolution of the mass spectrometer.

INTRODUCTION Host-cell proteins (HCPs) are ubiquitous process-related trace impurities of purified biologically-derived pharmaceutical products.1,2 Conventional HCP analysis by ELISA generally uses anti-sera raised against the HCP pool (i.e. the host cell proteins expressed by the null cell line).3 ELISA analysis provides a bulk quantitation of the overall HCP abundance. However, there is no information on individual species or their concentrations. Currently, there is interest to identify individual species as specific HCPs could be toxic, immunogenic, or potentially degrade the drug substance.3,4 Furthermore, specific HCP species may not elicit a strong immune response from the donor animal, leading to an underrepresentation of HCP species by ELISA.5 In addition, development of a process-specific ELISA method may require months to generate high quality polyclonal antibodies.3 LC-MS analysis, on the other hand, can identify and quantitate individual protein species at low ppm levels and has the potential to be an orthogonal analytical method to complement or even substitute conventional ELISA.6-9 In comparison to proteomic analysis of high complexity samples, such as cell lysates, HCP analysis involves samples of much lower complexity (fewer than 50 proteins generally, compared to more than 104 proteins in a cell lysate) with the need for a high dynamic range to detect HCPs at the low ppm level in the presence of the therapeutic protein. Previous LC-MS approaches attempted to address the high dynamic range issue by depleting the therapeutic protein10,11 or using multi-dimensional chromatography,12 with ion-mobility mass spectrometry as an additional separation dimension.13 Depletion of the therapeutic protein would seem to be a potential solution to the high concentration range challenge; however, extra steps are required in the analysis, and some HCP species may interact with the therapeutic protein14,15 and thus also be depleted. HCP analysis in the background of the therapeutic protein is, generally, the desired approach both for simplicity and completeness. While some studies have used data-dependent acquisition (DDA) to identify HCPs,16-18 data-independent acquisition (DIA) represents a potentially superior alternative.5,12,13,19,20 DDA suffers from a bias towards sampling and identification of high abundance species, resulting in poor and inconsistent detection of low level HCPs. The key advantage of DIA is that MS2 fragment information from all eluting precursors is acquired, and identification of proteins, even at low levels, is more reproducible.21 Contrary to DDA data, DIA data are multiplexed (each spectrum potentially contains fragments from multiple precursors), and fragment ions are detected multiple times across the elution peak, enabling simultaneous chromatographic tracking of the precursor as well as the associated fragment ions.

2

ACS Paragon Plus Environment

Page 2 of 21

Page 3 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Targeted (also referred to as peptide-centric) DIA data analysis requires a comprehensive assay library.22,23 An assay, in this context, includes the normalized retention time of the peptide, the precursor m/z, and the relative intensities of characteristic MS2 fragment ions (typically b and y ions from high energy collision dissociation (HCD) or collision induced dissociation (CID)). Moreover, each peptide charge state constitutes a separate assay.22,23 Thousands of assays can be generated from a DDA experiment by extracting the most intense fragment ions from high confidence peptide spectral matches. However, in DDA, the number of identifying spectra, and consequently, the number of targeted assays, correlates with protein abundance.24 Additionally, low abundance proteins may not be detected by DDA, resulting in an incomplete library with a bias towards high abundance proteins. The assay library can be enhanced through sample pre-fractionation (2D-LC-MS), additional replicates, analysis of related samples (e.g. null-cell line lysate and samples from early stages of purification), and targeted analysis. Nevertheless, there is always potential that some HCPs may not be represented in the assay library. Targeted analysis can be supplemented by an “untargeted” strategy in which the DIA data is converted to pseudo-DDA data with DIA-Umpire,25,26 and this data is searched against the entire Chinese Hamster Ovary (CHO) protein sequence database (or other relevant database). A recent benchmark comparison of DIA data processing strategies found that untargeted analysis with DIA-Umpire identified a significant number of additional peptides beyond targeted analysis in spite of an extensive targeted assay library. Importantly, when the assay library was appended with the unique DIA-Umpire identifications, those peptides were identified by the targeted strategies as well.27 A further challenge of HCP analysis is evaluation of the confidence in the peptide and subsequent protein identifications using standard targetdecoy approaches that were developed for high complexity samples. For purified therapeutics, the number of HCP proteins in the sample will generally be fewer than 50, and the true positive peptide population would be too small to accurately determine a false-discovery rate (FDR) threshold, leading to a compromise between identification sensitivity and specificity.28 In this paper, we describe a novel, automatable workflow for rapid in-depth HCP analysis. This workflow is implemented in the HCP profiling of an IgG1 monoclonal antibody (mAb) sample in the early preclinical stages of development after several stages of purification (Protein A column elution, cation exchange chromatography, and ultrafiltration buffer exchange). First, a targeted assay library containing over 4,000 assays for 632 protein groups was generated by 2D-LC-MS DDA analysis of a Protein A column eluate containing the mAb therapeutic. The generated library can be used for detection of the represented HCPs in similar therapeutic products and appended further with spectra from peptides identified in additional experiments. Then, the processed mAb sample, spiked with protein standards spanning a 0.5 to 100 ppm concentration range, was analyzed by LC-DIA-MS using the combined targeted and

3

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

untargeted data analysis, followed by verification and quantitation by parallel reaction monitoring (PRM). Peptide verification used a maximum of 10-10 MS-GF+ E-score and 6 fragment ion transitions. The methodology identified all of the standards and 16 HCPs down to low ppm level in the purified mAb sample, using 1D-LC-MS. The presented data analysis strategy is readily automatable and allows a rapid transition from DIA to PRM, enabling sensitive and specific identification of HCPs down to the low ppm level. The label-free quantitation based on fragment peak area provides a convenient estimation of HCP abundance suitable for support of downstream process development.

EXPERIMENTAL Materials and equipment Samples of a monoclonal therapeutic IgG1 antibody in the early preclinical stages of development after Protein A column elution and purified further by cation exchange chromatography and ultrafiltration buffer exchange (purified mAb) were generated from a CHO cell line at the Bristol-Myer Squibb bioprocessing facility (Devens, MA). Triethylammonium bicarbonate buffer (TEAB) (1.0 M, pH 8.0), LC-MS grade ammonium hydroxide solution (25% in H2O), cytochrome C (≥ 95%, from horse heart), lysozyme (≥ 95%, from chicken egg white), βcasein (≥ 90%, from bovine milk), myoglobin (≥ 95%, from horse skeletal muscle), lactoferrin (≥ 90%, from bovine milk), dithiothreitol (DTT), iodoacetamide (IAM), formic acid, and LC-MS retention time calibration standards, were obtained from MilliporeSigma (St. Louis, MO). The bicinchoninic acid (BCA) protein assay kit, LC-MS grade water, and LC-MS grade acetonitrile were obtained from Thermo Fisher Scientific (Waltham, MA). Sequencing-grade modified trypsin was purchased from Promega (Madison, WI), and MS grade lysyl endopeptidase (Lys-C) from Wako (Richmond, VA). A set of 10 stable isotope labeled (lysine +8 or arginine +10) tryptic peptides (8 - 13 residues) spanning the elution gradient were purchased from JPT GmbH (Germany); their sequences are presented in Supplement Table 1. High pH (~10) LC fractionation was carried out off-line using an XBridge peptide BEH C18 column (3.5 µm beads, 300Å, 2.1 x 150 mm) (Waters Corporation, Milford, MA) on an Agilent 1200 LC (Agilent Technologies, Santa Clara, CA). An ACQUITY UPLC M-class peptide CSH C18 column (1.7 µm beads, 130Å, 0.3 x 150 mm) (Waters Corporation) on an Ultimate 3500-RS LC (Thermo Fisher Scientific) was used for low pH (~2.7) reversed phase separation. All MS data were acquired on a QExactive Plus mass spectrometer with a heated ESI source (Thermo Fisher Scientific, San Jose, CA). HCP concentration was also measured using the CHO protein ELISA kit from Cygnus Technologies (Southport NC).

4

ACS Paragon Plus Environment

Page 4 of 21

Page 5 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Sample preparation Aliquots containing 0.6 mg of the IgG1 mAb sample (determined by BCA assay) were denatured in 150 µL of 8 M urea and 100 mM TEAB (pH 8.0). The sample was reduced in 10 mM DTT at 37 ˚C for 1 hour and then alkylated with 10 mM IAM in the dark and at room temperature for 45 minutes. 900 µL of cold acetone (pre-chilled to -20 °C) were added and the proteins precipitated at -20 ˚C overnight. After centrifugation at 12,000 x g for 15 minutes, the supernatant was discarded. The precipitated proteins were reconstituted in 250 µL of the digestion buffer, 25 mM TEAB (pH 8.0) in 10% acetonitrile and 90% water, and Lys-C was then added at a 1:100 (w/w) ratio. The samples were digested for 5 hours at 37 ˚C, and trypsin was added at 1:50 (w/w) to continue digestion overnight (18 hours) at 40 ˚C. After digestion, the samples were lyophilized to dryness and stored at -80 ˚C. LC-MS The digested purified mAb samples and high pH fractions of the Protein A eluate were injected directly onto an ACQUITY UPLC M-class peptide CSH C18 column and separated at 10 µL/min flow rate. Mobile phase A was 0.1% formic acid in water, and mobile phase B was 0.1% formic acid in acetonitrile. Gradient elution consisted of 10-minute loading and desalting in 2% acetonitrile, followed by a linear ramp to 32% acetonitrile over 120 minutes, then a linear ramp to 90% acetonitrile over 20 minutes, isocratic hold at 90% acetonitrile for 8 minutes, return to 2% acetonitrile in 1 minute, and a 20-minute hold at this mobile phase. The eluent was sprayed at +4.5 kV through the heated ESI source with nitrogen sheath gas. The mass spectrometer precursor scan was acquired at 70,000 resolution (at m/z =200), with the AGC set to 1,000,000 and 110 ms maximum injection time. The 455.12002 m/z polydimethylcyclosiloxane ion was set as the internal lock-mass calibrant. In DDA analysis, the MS1 scan was acquired over 400 to 1600 Th, whereas in DIA and PRM analysis, the MS1 scan was acquired over 390 to 1100 Th. In all data acquisition strategies, MS2 scans were acquired by HCD at a normalized collision energy set to 28, with the AGC set to 50,000 and a maximum ion injection time of 110 ms. Targeted assay library generation Three aliquots of digested Protein A column eluate (0.6 mg each, 1.8 mg total) containing 500 ppm HCP based on the Cygnus ELISA kit were combined and reconstituted in 40 µL of aqueous 20 mM ammonium formate (pH 10). The sample was injected directly onto the XBridge peptide BEH column and separated using a water and acetonitrile gradient with a 20 mM ammonium formate (pH 10) modifier at 0.2 mL/min. The gradient was delivered as follows: 2% acetonitrile for 30 minutes for loading and desalting, 46-minute linear ramp to 90% acetonitrile, 2-minute linear drop to 2% acetonitrile, and 9-minute hold at 2% acetonitrile. Fractions were collected every 2.5 minutes after the 30-minute desalting period. The 23

5

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

collected fractions were pooled by combining fractions 1, 11 and 21, fractions 2, 12 and 22, fractions 3, 13, and 23, and the remaining fractions were pooled as fractions 4 and 14, 5 and 15, 6 and 16, etc. The 10 resulting fractions were dried in vacuum and reconstituted in 120 µL of 0.1% formic acid (roughly 1.5 µg/µL concentrations), and spiked with 100 fmol of LC-MS retention time calibrant peptides. Each fraction was injected directly onto the ACQUITY UPLC M-class peptide CSH C18 column, separated using the described LC-MS protocol, and analyzed by Top 15 DDA. The MS2 scans were acquired using 1.5 m/z wide isolation windows centered on the monoisotopic peak at 35,000 resolution (m/z = 200). Each fraction was analyzed in triplicate. In the first replicate, 30 µg of sample were injected onto the column to maximize HCP loading. 15 µg were injected in the second replicate to avoid column over-loading. 15 µg were also injected in the third replicate, but the mass spectrometer was set to exclude precursors that were identified in the second replicate at high confidence (FDR < 0.5%) by an automated Morpheus29 post-acquisition search. The acquired data were processed in SearchGUI30 using the Myrimatch31 and MS-GF+32 algorithms and the CHO-K1 consortium database of 24,044 protein sequences.33 Maximum mass errors of 6 ppm were allowed for the precursor and fragment m/z, and carbidomethylation of cysteine was set as a fixed modification. Heavy lysine and arginine (for detection of RT standards) and oxidation of methionine were added as variable modifications. The search allowed for up to 2 missed cleavages and one tryptic miscleavage (semi-specific). The results from both search engines were combined and re-scored by PeptideShaker34 and exported in the mzIdentML format. An in-house script was used to extract a maximum of 10 highest intensity b and/or y ion transitions from the top scoring spectrum for each peptide in each charge state identified at 500 protein groups) for the FDR to be accurately set at 1% using the target-decoy model (see below). The retention time was normalized across the 30 runs using spiked-in retention time standards; a requirement since the peptide populations differed between fractions. Once the retention time normalized assay library was assembled, a specific set of peptides in the library could be used for retention time normalization during DIA analysis of the purified mAb. In our case, 20 high intensity mAb derived peptides spanning the gradient time were selected. The assembled library is intended to be implemented in subsequent analyses of the monoclonal antibody and related samples, without the need to repeat Stage 1. With the developed workflow the library can be continuously appended with peptides identified with high confidence in the untargeted search. Figure 2 presents the number of assays on a base 2 logarithmic scale incorporated into the targeted assay library for each protein (Y-axis) plotted against the protein intensity determined by MaxQuant37 on a base 10 logarithmic scale (X-axis). The figure demonstrates a concentration range spanning more than 6 orders of magnitude between the lowest level HCPs and the mAb. Contrary to an assay library generated from the null cell line, the Protein A eluate library is focused on the HCPs that could potentially be in the further purified samples. However, analysis of the HCPs in the Protein A eluate is affected by the high level of the mAb. As evidence of this, Figure 2 shows a positive correlation between protein abundance and the number of targeted assays. A more complete library could be constructed using more fractions, additional technical replicates, or by supplemental analysis of a null-cell line,38 but this would still not guarantee a complete assay library. Untargeted analysis by DIA-Umpire has been shown to identify proteins that are not represented in the assay library in complex samples,27 and this principle is exploited in this workflow.

9

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 21

Figure 2. Comparison of protein intensity and number of assays in the targeted assay library. The targeted assay library generated from high/low pH RP-LC-DDA-MS analysis of the Protein A eluate presented as a comparison between the number of assays and protein intensity. The positive correlation suggests that the targeted DIA data processing search alone is inefficient in detecting low level HCPs. The proteins that were detected in the processed mAb sample are circled in red, and the mAb proteins and digestion enzymes are labeled for reference.

Stages 2 and 3: DIA analysis of the purified mAb With the assay library established, DIA data could be acquired and analyzed in the second and third stages of the workflow (Figure 1). As detailed in the Experimental Section, the purified mAb was rapidly analyzed by 1D-LC-DIA-MS, and the acquired data were interrogated by the targeted assay library using, in this case, 20 mAb peptides for retention time normalization in the targeted search. In addition, and automatically, an untargeted search was conducted in which the data were converted to pseudo-MS2 spectra using DIA-Umpire, followed by a search against the NCBI CK1 CHO protein sequence database. The data processing procedure was developed to utilize open access algorithms that were weaved together using an in-house script, as detailed in the Supplementary Material. Figure 3 presents the number of putative peptide identifications from this combined (targeted and untargeted) analysis of 3 separate runs. Figure 3, which will be discussed further below, demonstrates that the two strategies identify overlapping, as well as different peptide populations supporting the necessity for performing both analyses.

10

ACS Paragon Plus Environment

Page 11 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Figure 3. Putative and verified peptide identifications from 3 DIA runs using the developed strategy. Blue - untargeted analysis; yellow - targeted analysis; green - overlap between the two strategies.

Contrary to high complexity samples where a target-decoy model is utilized to evaluate the confidence of the peptide and protein identifications, the assessment of false discovery rate

11

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 21

(FDR) is not straight-forward in a purified sample. 28,39 A purified mAb is low in complexity (fewer than 50 proteins), and the target-decoy model is problematic because the true positive population (peptides present in the sample) is much smaller than the decoy population consisting of peptides from a scrambled CHO protein sequence database. In targeted DIA data analysis, the FDR is established using a machine learning algorithm such as mProphet to assess attributes of library matches (co-elution, dot product, etc.) and generate a weighted combined score that best distinguishes between the target and mass shifted or shuffled decoy assays.40 When the true positive population is small, however, the decoy percentage is inflated, and the score threshold cannot be accurately set. If the score is set at the traditional 1% FDR, low scoring but true identifications, which would be accepted with a larger true positive population, will be discarded. Our workflow circumvents this risk by using less stringent FDR filters: 5% FDR for the untargeted search (DIA-Umpire) and only the internal OpenSWATH quality filters without an FDR threshold for the targeted search. This strategy minimized the chance of missing low level HCPs, but consequently, roughly 40% false positive identifications were putatively accepted. The subsequent PRM analysis step is required to verify the true positive peptides and discard false positive identifications. Stage 4A: PRM verification After generating a list of 154 putative HCP and protein standard peptides from the targeted and untargeted analysis of the 1D-LC-MS DIA data, PRM analysis was employed to verify true positive identifications and filter out false positive matches. The PRM approach isolated precursors in a narrow m/z window (1.4 Th), and the high mass accuracy (5 ppm maximum mass error) and resolution (70,000 at m/z 200) of the Orbitrap for both the precursor and fragment ions were used to identify peptides with high specificity. The retention time overlap of the precursor and characteristic fragment ions was the basis for accepting a peptide as a true positive. Using Theoretical PRM,41 we confirmed that 3 characteristic fragments at 5 ppm mass error along with a 1.4 Da precursor isolation window could be used to specifically identify a peptide in the CHO protein sequence database. Confidence in identification by targeted analysis was increased by peptide elution at the reference retention time. However, the untargeted analysis had no retention time reference, thus the PRM spectra verifying untargeted peptides were searched by MS-GF+ and filtered to a maximum E-value of 10-10 which is a high confidence threshold in DDA identifications. This threshold resulted in at least 6 co-eluting transitions verifying peptide identifications by PRM. Furthermore, each peptide sequence was checked for homology to the therapeutic antibody to ensure it was not a peptide from the antibody. The numbers of verified peptides determined from the targeted and untargeted approaches are listed in Figure 3 for 3 separate technical replicates. There is a high degree of

12

ACS Paragon Plus Environment

Page 13 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

agreement between the targeted and untargeted search strategies, with 41 to42 overlapped PRM verified peptides (from a total of 154 putative peptides). At the same time, 16 to 18 peptides were exclusively identified by the targeted strategy. Twenty peptides below 10-10 MSGF+ E-score (with at least 6 co-eluting fragment ions) were identified exclusively by the untargeted search, for a total of 75 peptides. This list includes peptides from the spiked-in standards and contaminant proteins. With an expanded targeted library, the number of identifications from the targeted search will be increased, but the untargeted search will still, at a minimum, provide a measure of the completeness of the analysis at no additional time or cost. The 16 to 18 peptides observed in the targeted search but not detected by the untargeted approach were likely due to the limitations of the search engines (MSGF+ and Myrimatch) in the low complexity samples. The peptides identified in the purified mAb sample were mapped to 16 CHO proteins, which are listed in Table 1.

13

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 21

Table 1. Quantitation of the identified HCPs. Estimated HCP Concentration in Purified mAb Accession number

Experiment 1 (ppm) (n = 3)a

Host cell protein

Average

XP_007636365.1 Verified peptides XP_007621224.1

35

Verified peptides XP_007613399.1

40

Verified peptides XP_007639237.1

60 62

Verified peptides: XP_007610885.1

N/A

45%

43

55%

4

2%

38

7%

30

42%

32

50%

N/A

62%

27

26%

1

QVIQDGVLHGLCHQMPPEK Lysosomal acid lipase

7

35%

8

15%

12

17%

6

12%

7

85%

4

22%

18

LCTNVFFLICGFNEK Peptidyl-prolyl cis-trans isomerase B

2 DTNGSQFFITTVK

Protein disulfide-isomerase

4

ILEFFGLK, FFPATADR, THILLFLPK, NFEEVAFDEK, VHSFPTLK Uncharacterized protein LOC100755734

3

35%

4

38%

N/A

22%

0.5

23%

4

0.4

46%

16

FLTSVIPR Ubiquitin

Verified peptides NP_001233694.1

13%

DSAIGFLR, NLLFNDNTECLAK

Verified peptides XP_007622992.1

56

78 kDa glucose-regulated protein precursor 27 27% 26 6% 32 ELEEIVQPIISK, TFAPEEISAMVLTK, NQLTSNPENTVFDAK, TWNDPSVQQDIK, LVQAFQFTDK, IINEPTAAAIAYGLDK, VEIIANDQGNR DNA Repair Protein RAD52 28 13% 17 15% N/A

Verified peptides XP_007634117.1

11%

EEVALDLSVK Lactotransferrin

Verified peptides XP_007608107.1

N/A

SLLNSLEEAK, LFDSDPITVVLPEEVSK, LTQQYNELLHSLQTK Uncharacterized protein C15orf39

Verified peptides XP_007633232.1

32 40

Verified peptides NP_001233668.1

35%

QIVYCIGGENLSVAK Clusterin

Verified peptides XP_007619666.1

57

FFTQPDKNFSNTK 26S proteasome non-ATPase regulatory 5

Verified peptides XP_007614428.1

158%

AVGEVTNSEGTWVQLDK Sister chromatid cohesion protein PDS5

Verified peptides XP_007623559.1

%CV

SLLFESAWKK E3-Ubiquitin Ligase

Verified peptides XP_007609958.1

Average

Peptides in Targeted Library

Putative phospholipase B-like 2 178 12% 134 8% 40 c SVLLDAASGQLR, LALDGATWADIFK, YVQPQGCVLEWIR, VLTILEQIPGMVVVADK, LSLGSGSCSAIIK, VLTILEQIPGMVVVADKTEDLYK, DLLVAHNTWNSYQNMLR, AFIPNGPSPGSR, VLTILEQIPGMVVVADK TRAF3-interacting protein 1 isoform X2 47 68% 64 16% N/A

Verified peptides: XP_007607154.1

%CV

Experiment 2 (ppm) (n = 10)b

0.4

TITLEVEPSDTIENVK, TLSDYNIQK Peroxiredoxin-1

0.5

Verified peptides

37%

LVQAFQFTDK

a

Experiment 1 was used to verify putative peptides using 10 minute retention time windows to ensure detection. (3 replicates) b Experiment 2 was used to evaluate the label-free quantitative strategy using different concentration of spiked peptide standards. (five levels, duplicate analysis = 10 runs) c HCP quantitation was based on the peptide in bold.

14

ACS Paragon Plus Environment

Page 15 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

The ability to conduct HCP analysis using 1D-LC has an obvious time advantage compared to 2D-LC approaches. To assess potential losses in sensitivity using only the 1D approach, the purified mAb sample was analyzed by 2D-high/low pH reversed phase LC-DIA-MS with 5 high-pH fractions run in duplicate in the second dimension. A total of 600 mg of digested purified mAb were injected unto the high pH column and an equivalent of 15 µg were injected unto the analytical column with identical settings used for 1D-LC analysis. No additional HCP proteins were found, although several additional peptides were identified for the higher concentration HCPs. The ability to use 1D-LC-MS without significant loss of information is an important advantage for frequent utilization of the workflow. The 16 verified HCPs are presented in Table 1. Experiment 1 was focused on verification of putative HCP peptides. Thus, a 10 minute retention time window was used in the PRM analysis to ensure detection of HCP despite potentially large shifts in retention time. Three PRM runs were carried out to provide a first estimate of concentration of each HCP with %CV for technical replicates. Quantitation is discussed in the next section. Many of the identified HCPs have been reported in previous studies. For example, clusterin,12,19,42-44 putative phospholipase B-like 2,42,45 78 kDa glucose-regulated protein,12,19,42,44 and protein disulfide-isomerase16,19,42 have been observed in mAb and Fc fusion protein samples after Protein A purification and CEX chromatography. Other HCPs such as Sister chromatid cohesion protein PDS5, and TRAF3 interacting protein have not been previously reported. Notably, the HCP profile can be affected with slight changes in the mAb structure and upstream and downstream processes.14,46 Stage 4B: PRM Quantitation The PRM stage also targeted peptides from the five spiked-in protein standards (β−casein, lysozyme, myoglobin, lactoferrin, and cytochrome C) listed in Table 2 along with their concentrations to evaluate the sensitivity and linearity of the workflow and for construction of a calibration curve to estimate the abundance of the identified HCPs. The signal for each peptide was determined as the sum of the 3 highest intensity b or y ion transitions, and the protein concentration was estimated from the highest intensity peptide signal (signals from multiple charge states were combined). Single peptide-based quantitation was used to estimate the concentration of low level HCPs for which only one peptide was detected. Table 2 presents the average intensity measurements and standard deviation for the 5 protein standards. The protein standards at the 0.5 ppm and 2.5 ppm levels were identified by one peptide and the remaining protein standards were detected by several peptides (but quantitated using the highest intensity peptide). The exclusion of myoglobin produced a linear (R2 > 0.990) calibration curve that was used for assessment of the HCP concentration. Myoglobin was detected by large, low responding peptides, and its concentration was underestimated by a factor of 2 based on the linear calibration plot.

15

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 21

As noted above, Experiment 1 used 10 minute retention time windows in PRM to ensure detection of putative peptides for verification. Three technical replicates were used to obtain an initial estimate of the concentration levels and check repeatability. Experiment 2, which quantitated HCPs in a second digestion of the purified mAb was conducted to assess the reproducibility, accuracy, and linearity of the label-free PRM quantitative strategy. In Experiment 2, 6 minute retention time windows were used to increase the number of data points per analyte peak, resulting in more reproducible measurements for most HCPs (Table 1). In future implementations of this workflow, narrow retention time windows should be used, as in Experiment 2, once the reproducibility of the chromatographic separations is ensured. Importantly, there is good agreement in HCP concentration between Experiments 1 and 2. Spiked Quantity ppm fmol

Measured Quantity

Verified peptides

41 ± 2.2

100

92

92.0 fmol

8

17 kDa

9.5 ± 1.9

50

44

21.3 fmol

3

Lactoferrin (bovine)

78 kDa

2.7 ± 0.2

25

4.8

6.1 fmol

14

Cytochrome C (horse)

12 kDa

0.3 ± 0.2

2.5

3.2

0.7 fmol

1

β-casein (bovine)

25 kDa

0.005 ± 0.003

0.5

0.3

0.0 fmol

1

Protein standard

M.W.

Top peptide peak area (million intensity units)

Lysozyme (chicken)

16 kDa

Myoglobin (horse)

Table 2. Quantitation of spiked-in protein standards.

The accuracy of quantitation was examined in Experiment 2 using ten stable isotope labeled (SIL) peptides that spanned the retention range of the gradient, but did not match any HCP sequences (Supplement Table S-1). The peptides were spiked into five aliquots of the digested purified mAb at 0, 2.5, 5, 10, or 20 fmol/injection and analyzed alongside the HCP peptides by PRM in duplicate. The spiked peptide concentrations were plotted against the concentrations measured using the calibration curve (generated from the protein standards). The peptides showed linear (R2 > 0.990) response in the 2.5 to 20 fmol range. The linear regression slope (response factor) for each peptide ranged from 0.4 to 2.4, indicating that the PRM quantitation should be accurate within approximately 2 to 3 fold (Table S-1). This range of error is likely the result of differences in peptide response factors and/or ion suppression from co-eluting mAb peptides. To further assess quantitative accuracy, 78-kDa glucose regulated protein, disulfide isomerase, phospholipase B, and clusterin were quantitated using SIL peptide analogues synthesized with heavy lysine or arginine residues spiked at 10 fmol. Comparison of peak areas between the endogenous and the SIL peptides found 78 kDa glucose regulated protein

16

ACS Paragon Plus Environment

Page 17 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

concentration at 50 ppm (CV 22%), phospholipase B at 54 ppm (CV 56%), clusterin at 20 ppm (CV 42%), and disulfide isomerase at 3 ppm (CV 12%). Compared to the stable isotope strategy, the label-free measurements were within the estimated 2 - 3 fold error. While the SIL strategy is generally preferred to overcome ion suppression effects, in a practical setting, the SIL standards may not be readily available, and the described label-free approximation should be sufficient for many applications. LC-MS measures protein concentration, but conventionally the HCP content is reported as ppm relative to the mAb. Assuming that the identified peptides originated from the full (untruncated) HCP sequences, the total HCP content of approximately 500 ppm is detected in Experiment 2. Phospholipase B at over 100 ppm ppm corresponds to 30 % of the total content. The other HCPs are under 70 ppm, with 6 below 10 ppm, and several as low as 1 ppm. The total of 500 ppm differs from the 20 ppm determined by ELISA. Such large differences have been observed by others44 and attributed to low or absent immune response to some HCP species during the generation of ELISA antibodies. In addition, the ELISA method used in this study was not optimized for the specific sample.47 With further optimization of the ELISA method and accurate LC-MS quantitation using SIL analogues, we could expect closer agreement between the two measurements. However the inability of ELISA to detect immune inert HCPs will remain a discrepancy between the two approaches.

CONCLUSION This paper has presented a rapid and automatable DIA to PRM workflow for HCP identification and quantitation in biopharmaceutical products. While DIA analysis has previously been implemented in HCP characterization using the MSE 20 and SWATH48 strategies, our workflow exploits the high resolution and mass accuracy of the Orbitrap mass analyzer and a novel data analysis strategy to identify HCPs at low ppm levels in the background of the mAb using 1D-LC. While HCP abundance is dependent on the downstream purification process and the specific therapeutic protein, previous studies identified a similar number of HCP species, in processed mAbs.13 Using spiked-in protein standards as internal calibrants, the PRM verification stage allowed quantitative assessment of HCP concentration. Overall the workflow demonstrated potential in biopharmaceutical production, especially during downstream process development. The combined targeted and untargeted workflow with PRM verification and quantitation is not limited to HCP analysis and can be implemented for more complex proteomic samples by identifying most species by DIA and using PRM to verify potentially ambiguous identifications.

SUPPORTING INFORMATION 17

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 18 of 21

The supporting information contains detailed descriptions of the targeted library assembly and DIA data processing. The descriptions provide step-by-step procedures for generating a targeted assay library for OpenSWATH analysis from raw DDA data and for processing DIA data using a combined targeted and untargeted search. Additionally, the supporting information contains the sequences and intensity measurements of the stable isotope labeled peptides used to check the linearity and accuracy of the label-free concentration.

ACKNOWLEDGMENTS The authors thank Bristol Myers Squibb for support of this research. The authors would also like to thank Dr. Johan Teleman (Lund University, Sweden), Dr. Alexey I Nesvizhskii and Dr. ChihChiang Tsou (University of Michigan) for valuable discussions on DIA data processing.

CONFLICT OF INTEREST STATEMENT The authors declare no conflict of interest.

18

ACS Paragon Plus Environment

Page 19 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

References (1) Leser, E. W.; Asenjo, J. A. J Chromatogr-Biomed 1992, 584, 43-57. (2) Shukla, A. A.; Thommes, J. Trends Biotechnol 2010, 28, 253-261. (3) Wang, X.; Hunter, A. K.; Mozier, N. M. Biotechnol Bioeng 2009, 103, 446-458. (4) Gao, S. X.; Zhang, Y.; Stansberry-Perkins, K.; Buko, A.; Bai, S. J.; Nguyen, V.; Brader, M. L. Biotechnol Bioeng 2011, 108, 977-982. (5) Schenauer, M. R.; Flynn, G. C.; Goetze, A. M. Anal. Biochem. 2012, 428, 150-157. (6) Kaltashov, I. A.; Bobst, C. E.; Abzalimov, R. R.; Wang, G.; Baykal, B.; Wang, S. Biotechnol Adv 2012, 30, 210-222. (7) Marcus, K. Quantitative methods in proteomics; Humana Press ; Springer: New York, 2012, p xv, 539 p. (8) Aebersold, R.; Mann, M. Nature 2003, 422, 198-207. (9) Tscheliessnig, A. L.; Konrath, J.; Bates, R.; Jungbauer, A. Biotechnol J 2013, 8, 655-670. (10) Bomans, K.; Lang, A.; Roedl, V.; Adolf, L.; Kyriosoglou, K.; Diepold, K.; Eberl, G.; Molhoj, M.; Strauss, U.; Schmalz, C.; Vogel, R.; Reusch, D.; Wegele, H.; Wiedmann, M.; Bulau, P. PLoS One 2013, 8, 11. (11) Capila, J. M. J. A. M. V. F. Y. Y. S. S. J. A. I. In Proceedings of the 64th ASMS Conference on Mass Spectrometry and Allied Topics: San Antonio TX, 2016. (12) Doneanu, C. E.; Xenopoulos, A.; Fadgen, K.; Murphy, J.; Skilton, S. J.; Prentice, H.; Stapels, M.; Chen, W. B. Mabs-Austin 2012, 4, 24-44. (13) Doneanu, C. E.; Anderson, M.; Williams, B. J.; Lauber, M. A.; Chakraborty, A.; Chen, W. B. Anal Chem 2015, 87, 10283-10291. (14) Levy, N. E.; Valente, K. N.; Choe, L. H.; Lee, K. H.; Lenhoff, A. M. Biotechnol. Bioeng. 2014, 111, 904912. (15) Nogal, B.; Chhiba, K.; Emery, J. C. Biotechnol Progr 2012, 28, 454-458. (16) Tait, A. S.; Hogwood, C. E. M.; Smales, C. M.; Bracewell, D. G. Biotechnol. Bioeng. 2012, 109, 971982. (17) Hogwood, C. E. M.; Tait, A. S.; Koloteva-Levine, N.; Bracewell, D. G.; Smales, C. M. Biotechnol. Bioeng. 2013, 110, 240-251. (18) Krawitz, D. C.; Forrest, W.; Moreno, G. T.; Kittleson, J.; Champion, K. M. Proteomics 2006, 6, 94-110. (19) Zhang, Q. C.; Goetze, A. M.; Cui, H. C.; Wylie, J.; Trimble, S.; Hewig, A.; Flynn, G. C. mAbs 2014, 6, 659-670. (20) Farrell, A.; Mittermayr, S.; Morrissey, B.; Mc Loughlin, N.; Iglesias, N. N.; Marison, I. W.; Bones, J. Analytical Chemistry 2015, 87, 9186-9193. (21) Wang, J.; Tucholska, M.; Knight, J. D. R.; Lambert, J. P.; Tate, S.; Larsen, B.; Gingras, A. C.; Bandeira, N. Nature Methods 2015, 12, 1106-1108. (22) Rost, H. L.; Rosenberger, G.; Navarro, P.; Gillet, L.; Miladinovic, S. M.; Schubert, O. T.; Wolskit, W.; Collins, B. C.; Malmstrom, J.; Malmstrom, L.; Aebersold, R. Nat Biotechnol 2014, 32, 219-223. (23) Colangelo, C. M.; Chung, L. S.; Bruce, C.; Cheung, K. H. Methods 2013, 61, 287-298. (24) Liu, H. B.; Sadygov, R. G.; Yates, J. R. Anal Chem 2004, 76, 4193-4201. (25) Tsou, C. C.; Avtonomov, D.; Larsen, B.; Tucholska, M.; Choi, H.; Gingras, A. C.; Nesvizhskii, A. I. Nat Methods 2015, 12, 258-+. (26) Tsou, C. C.; Tsai, C. F.; Teo, G. C.; Chen, Y. J.; Nesvizhskii, A. I. Proteomics 2016, 16, 2257-2271. (27) Navarro, P.; Kuharev, J.; Gillet, L. C.; Bernhardt, O. M.; MacLean, B.; Rost, H. L.; Tate, S. A.; Tsou, C. C.; Reiter, L.; Distler, U.; Rosenberger, G.; Perez-Riverol, Y.; Nesvizhskii, A. I.; Aebersold, R.; Tenzer, S. Nat Biotechnol 2016, 34, 1130-1136. 19

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 20 of 21

(28) Elias, J. E.; Gygi, S. R. In Proteome Bioinformatics, Hubbard, S. J.; Jones, A. R., Eds.; Humana Press Inc: Totowa, 2010, pp 55-71. (29) Wenger, C. D.; Coon, J. J. J. Proteome Res. 2013, 12, 1377-1386. (30) Vaudel, M.; Barsnes, H.; Berven, F. S.; Sickmann, A.; Martens, L. Proteomics 2011, 11, 996-999. (31) Tabb, D. L.; Fernando, C. G.; Chambers, M. C. J Proteome Res 2007, 6, 654-661. (32) Kim, S.; Pevzner, P. A. Nat. Commun. 2014, 5, 10. (33) Lewis, N. E.; Liu, X.; Li, Y. X.; Nagarajan, H.; Yerganian, G.; O'Brien, E.; Bordbar, A.; Roth, A. M.; Rosenbloom, J.; Bian, C.; Xie, M.; Chen, W. B.; Li, N.; Baycin-Hizal, D.; Latif, H.; Forster, J.; Betenbaugh, M. J.; Famili, I.; Xu, X.; Wang, J.; Palsson, B. O. Nat. Biotechnol. 2013, 31, 759-765. (34) Vaudel, M.; Burkhart, J. M.; Zahedi, R. P.; Oveland, E.; Berven, F. S.; Sickmann, A.; Martens, L.; Barsnes, H. Nat Biotechnol 2015, 33, 22-24. (35) Schubert, O. T.; Gillet, L. C.; Collins, B. C.; Navarro, P.; Rosenberger, G.; Wolski, W. E.; Lam, H.; Amodei, D.; Mallick, P.; MacLean, B.; Aebersold, R. Nat. Protoc. 2015, 10, 16. (36) Teleman, J.; Rost, H. L.; Rosenberger, G.; Schmitt, U.; Malmstrom, L.; Malmstrom, J.; Levander, F. Bioinformatics 2015, 31, 555-562. (37) Cox, J.; Mann, M. Nat. Biotechnol. 2008, 26, 1367-1372. (38) Yu, f. y. d. E. W. J. C. D. A. M. C. In Proceedings of the 64th ASMS Conference on Mass Spectrometry and Allied Topics: San Antonio, TX, 2016. (39) Choi, H.; Nesvizhskii, A. I. J Proteome Res 2008, 7, 47-50. (40) Reiter, L.; Rinner, O.; Picotti, P.; Huttenhain, R.; Beck, M.; Brusniak, M. Y.; Hengartner, M. O.; Aebersold, R. Nat. Methods 2011, 8, 430-435. (41) Peterson, A. C.; Russell, J. D.; Bailey, D. J.; Westphall, M. S.; Coon, J. J. Mol Cell Proteomics 2012, 11, 1475-1488. (42) Joucla, G.; Le Senechal, C.; Begorre, M.; Garbay, B.; Santarelli, X.; Cabanne, C. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci. 2013, 942, 126-133. (43) Levy, N. E.; Valente, K. N.; Lee, K. H.; Lenhoff, A. M. Biotechnol. Bioeng. 2016, 113, 1260-1272. (44) Farrell, A.; Mittermayr, S.; Maorrissey, B.; McLoughlin, N.; Iglesias, N. N.; Marison, I. W.; Bones, J. Analytical Chemistry 2015, 87, 9186-9193. (45) Vanderlaan, M.; Sandoval, W.; Liu, P.; Nishihara, J.; Tsui, G.; Lin, M.; Gunawan, F.; Parker, S.; Wong, R. M.; Low, J.; Wang, X.; Yang, J.; Veeravalli, K. K.; McKay, P.; Yu, C.; O'Connell, L.; Tran, B.; Vij, R.; Fong, C.; Francissen, K.; Zhu-Shinmoni, J.; Quarmby, V.; Krawitz, D. Bioprocess. Int. 2015, 13, 18-55. (46) Aboulaich, N.; Chung, W. K.; Thompson, J. H.; Larkin, C.; Robbins, D.; Zhu, M. Biotechnol. Prog. 2014, 30, 1114-1124. (47) de Zafra, C. L.; Quarmby, V.; Francissen, K.; Vanderlaan, M.; Zhu-Shimoni, J. Biotechnol Bioeng 2015, 112, 2284-2291. (48) Rost, H. L.; Rosenberger, G.; Navarro, P.; Gillet, L.; Miladinovic, S. M.; Schubert, O. T.; Wolski, W.; Collins, B. C.; Malmstrom, J.; Malmstrom, L.; Aebersold, R. Nat Biotechnol 2014, 32, 219-223.

20

ACS Paragon Plus Environment

Page 21 of 21

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

For TOC only

21

ACS Paragon Plus Environment