Background

There is growing interest in alternative microbial hosts as manufacturing chassis to produce recombinant proteins [1, 2], including ones with therapeutic uses typically manufactured in mammalian cells. The methylotrophic yeast Komagataella phaffii (Pichia pastoris) offers unique advantages compared to the conventional model microorganisms Escherichia coli and Saccharomyces cerevisiae because of its productive secretory pathway [3,4,5]. K. phaffii is routinely used for large-scale manufacture of small therapeutic proteins (< 30 kD) such as insulin [6], vaccine antigens [7], and VHH antibodies [8]. In addition, K. phaffii has now been used for the commercial production of a full length monoclonal antibody (mAb) as well (eptinezumab) [9].

Mammalian cell lines such as Chinese hamster ovary (CHO) and human embryonic kidney (HEK293) have required several decades of empirical selections and process-related optimizations to manufacture mAbs and other large proteins efficiently and reliably [10]. Emerging applications of gene editing in CHO cells have shown that the knockout of up to 14 natively secreted host cell proteins (HCPs) can improve both the secreted titer and purity of a recombinant mAb [11]. In contrast to CHO cells, K. phaffii secretes a limited number of HCPs [12], which can result in high initial purity of recombinant proteins in culture supernatant and facilitate purification and characterization of the product [13]. Proteins secreted at lower titers, however, may compete with HCPs for cellular resources including amino acids, ribosomes, protein folding machinery, and secretory capacity [14,15,16]. We hypothesized that knockout of natively secreted proteins may improve recombinant protein secretion in K. phaffii.

Here, we characterized a set of proteins from K. phaffii identified in the culture fluids after fermentation, and disrupted up to 11 genes that code for these secreted proteins. Several engineered strains, especially one with six disrupted genes, exhibited improved production of multiple large (>50 kDa) human proteins. To facilitate the iterative knockout of more than three secreted proteins, we performed a pooled CRISPR-Cas9 knockout library to measure the essentiality of all secreted proteins. This knowledge of gene essentiality and new capability for pooled screening methods should also inform future efforts to engineer other cellular processes or pathways in K. phaffii for improved production of recombinant proteins.

Materials and methods

Yeast strain cultivation

Strains were grown in 3 mL cultures in 24-well deep well plates (25 °C, 600 rpm) or 100 mL cultures in 500 mL shake flasks (25 °C, 300 rpm). Cells were cultivated in complex media (potassium phosphate buffer pH 6.5, 1.34% nitrogen base w/o amino acids, 1% yeast extract, 2% peptone). Cells were inoculated at 0.1 OD600, outgrown for 24 h with 4% glycerol feed, pelleted, and resuspended in fresh media with 3% methanol for HSA production, or 1% methanol, 40 g/L sorbitol, and 10 mM glutathione for trastuzumab production. Supernatant samples were collected after 24 h of production and analyzed.

For quantitative measurement of secreted protein titer, strains were cultivated in biological triplicate from frozen stocks. For growth assays, strains that produce trastuzumab were seeded at an optical density of 0.01 OD600 in 200 µL cultures (25 °C, 300 rpm) in either outgrowth or production media in biological triplicate from frozen stocks. OD600 was measured every hour for 48 h and plotted on a log axis. Growth rate was calculated manually as the slope of the curve during exponential or stationary growth.

Yeast strain construction

All strains were derived from wild-type Komagataella phaffii (NRRL Y-11430). Strains for recombinant protein production were derived from a modified base strain [AltHost Research Consortium Strain S-63 (RCR2_D196E, RVB1_K8E)] described previously [17]. Genes containing recombinant protein products HSA and trastuzumab were synthesized (Integrated DNA Technologies) and cloned into a custom vector with the methanol-inducible promoter PAOX1. To enable protein secretion, HSA was expressed with the signal peptide from the S. cerevisiae α-mating factor, and trastuzumab was expressed with the signal peptide from the S. cerevisiae α-mating factor for the light chain and the signal peptide from human HSA for the heavy chain. All vector sequences are listed in the Supplemental Materials.

K. phaffii strains were transformed as described previously [18]. After transformation of the HSA or trastuzumab expression vectors, 4–8 clones were selected and grown in 3 mL cultures. Supernatant samples were analyzed by SDS-PAGE, and the clone that exhibited the highest productivity was selected for quantitative growth and titer measurements.

Knockout of individual genes was performed with a custom knockout cassette as described previously [19]. Disruption of genes was confirmed by PCR and Sanger sequencing. Design, construction, and screening of the pooled knockout library is described in the Additional file 1.

Analytical assays for protein characterization

SDS-PAGE was carried out as described previously [20]. HSA supernatant titers were measured by reverse phase liquid chromatography. Trastuzumab supernatant titers were measured by Protein A biolayer interferometry. Specific productivity was calculated as titer normalized to cell density by OD600, relative to the original base strain.

LCMS measurement of the K. phaffii secretome

Wild-type K. phaffii was cultivated in 200 mL shake flask cultures in complex media. Cells were inoculated at 0.1 OD600, outgrown for 48 h with 4% glycerol feed, pelleted, and resuspended in fresh media with 1.5% methanol feed to simulate recombinant gene expression. Supernatant samples were collected after each phase of the cultivation.

Supernatant was reduced (10 mM dithiothreitol, 56 °C for 45 min) and alkylated (50 mM iodoacetamide, room temperature in the dark for 1 h). Proteins were subsequently digested with trypsin (sequencing grade, Promega, Madison, WI), at an enzyme/substrate ratio of 1:50, at room temperature overnight in 100 mM ammonium acetate pH 8.9. Trypsin activity was quenched by adding formic acid to a final concentration of 5%. Peptides were desalted using C18 SpinTips (Protea, Morgantown, WV), lyophilized, and stored at − 80 °C.

Peptides were labeled with TMT 6plex (Thermo) per manufacturer’s instructions. Lyophilized samples were dissolved in 70 μL ethanol and 30 μl of 500 mM triethylammonium bicarbonate (pH 8.5), and the TMT reagent was dissolved in 30 μl of anhydrous acetonitrile. The solution containing peptides and TMT reagent was vortexed and incubated at room temperature for 1 h. Samples labeled with the ten different isotopic TMT reagents were combined and concentrated to completion in a vacuum centrifuge. The samples were labeled using the TMT 10plex channels as follows: 126–4/27/16 48 h induction; 127N–4/29/16 48 h harvest; 127C–5/6/16 96 h harvest; 128N–5/4/16 48 h induction; 129N–5/13/16 96 h harvest; 129C–5/20/16 96 h harvest; 130N–5/18/16 48 h induction; 130C–5/27/16 96 h harvest; 131–5/25/16 48 h induction.

Peptides were loaded on a precolumn and separated by reverse phase HPLC (Thermo Easy nLC1000) over a 140 min gradient before nanoelectrospray using a QExactive Plus mass spectrometer (Thermo). The mass spectrometer was operated in a data-dependent mode. The parameters for the full scan MS were: resolution of 70,000 across 350–2000 m/z, AGC 3e6, and maximum IT 50 ms. The full MS scan was followed by MS/MS for the top 10 precursor ions in each cycle with a NCE of 34 and dynamic exclusion of 30 s. Raw mass spectral data files were searched using Proteome Discoverer (Thermo) and Mascot version 2.4.1 (Matrix Science). Mascot search parameters were: 10 ppm mass tolerance for precursor ions; 15 mmu for fragment ion mass tolerance; 2 missed cleavages of trypsin; fixed modification were carbamidomethylation of cysteine and TMT 10-plex modification of lysines and peptide N-termini; variable modification was methionine oxidation. TMT quantification was obtained using Proteome Discoverer and isotopically corrected per manufacturer’s instructions. Only peptides with a Mascot score greater than or equal to 25 and an isolation interference less than or equal to 30 were included in the quantitative data analysis. Relative abundance of each protein was defined by the log10 of the total area under the curve for all peptide counts detected for each protein, summed over four independent replicate cultivations.

Analysis of the K. phaffii secretome

We determined which genes in the K. phaffii genome contained a signal peptide using SignalP 5.0 and filtering for Sec/SPI > 0.5 [21]. Comparison of the K. phaffii secretome to Valli et al. [22] was performed by manual comparison in SnapGene (snapgene.com) of protein coding sequences from both the Love et al. genome [23] and the genome from Pichiagenome.org [24]. Descriptions of protein functions were obtained using BLAST.

Transcriptome analysis

Cells were cultivated at 3 mL plate scale and harvested after 18 h of production in methanol medium. RNA was extracted and purified according to the Qiagen RNeasy 96 kit. RNA quality was analyzed on an Agilent BioAnalyzer to ensure RNA Quality Number > 6.5. RNA was reverse transcribed with Superscript III (ThermoFisher) and amplified with KAPA HiFi HotStart ReadyMix (Roche). RNA libraries were prepared using the Nextera XT DNA Library Preparation Kit with the Illumina DNA/RNA UD Indexes Set A. sequenced on an Illumina Nextseq to generate paired reads of 50 (read 1) and 50 bp (read 2). Sequenced mRNA transcripts were demultiplexed using sample barcodes, aligned to the WT.fa Komagataella phaffii genome (strain Y11430) and exogenous transgenes, and quantified using Salmon version 1.6.0 [25]. Gene level summaries were prepared using tximport version 1.24.0 [26] running under R version 4.2.1 [27]. Gene set enrichment analysis (GSEA) was performed with GSEA 4.1.0 using Wald statistics calculated by DESeq2 [28] and gene sets from yeast GO Slim [29].

Results

Identification and knockout of secreted proteins

We sought to identify the set of proteins manifest in the culture fluids during fermentation (the secretome) in K. phaffii. We computationally identified 257 coding sequences in the K. phaffii genome with putative secretory signal peptides (Methods, Table S1). We then cultured wild type K. phaffii (NRRL Y-11430) and analyzed the proteins found in the extracellular fluid by mass spectrometry (Fig. S1) [23]. We detected 134 proteins (Table S1). The relative abundance of most proteins was similar between the glycerol and methanol medium (R = 0.93), respectively (Fig. 1A). (These two sources of carbon are commonly used to accumulate biomass and induce recombinant protein production, respectively.) Interestingly, only 30 of the proteins identified in the cell cultivation fluids by mass spectrometry had computationally predicted signal peptides. A significant number of these proteins were previously found to be enriched in microsomes (endoplasmic reticulum) (p < 0.0001 in glucose medium, p < 0.001 in methanol medium) and in the very early Golgi (p < 0.01 in methanol medium) (Fig. S2), suggesting these proteins may be secreted by the canonical yeast protein secretory pathway [22]. We did not detect 227 additional proteins predicted to contain signal peptides. These proteins may be either (1) incorrectly annotated, (2) present at concentrations too low to detect, or (3) targeted to other cellular organelles (and therefore are not secreted). A significant number of these proteins were previously found to be enriched in organelles such as the very early Golgi, early Golgi, microsome, vacuole, and mitochondria, all of which use signal peptides for protein localization (Fig. S2) [30, 31]. Finally, we experimentally detected 104 proteins not predicted to contain secretory signal peptides. A significant number of these proteins were associated with organelles such as the cytosol, mitochondria, and peroxisome (Fig. S2). These proteins were also previously found in most organelles and cell fractions, even if not statistically enriched (Fig. S3). The genes that code for these proteins are also highly expressed (p < 0.0001) (Fig. S4) [23]. We postulate that these proteins are abundant in the cell and may escape into the extracellular space by cell lysis or non-specific packaging into vesicles. These abundant intracellular proteins may also compete with the recombinant protein for cellular resources during transcription and translation [14]. Based on this analysis, we defined the secretome of K. phaffii, therefore, as the collection of 361 proteins that were predicted to contain a signal peptide or that were detected in culture supernatants (Table S1).

Fig. 1
figure 1

Identification and knockout of secreted proteins. A Relative abundance of proteins detected in culture supernatant harvested from cultures in glycerol medium (outgrowth) or methanol medium (production). B Knockout efficiencies of sequential disruption of the most abundant proteins in the K. phaffii secretome

Full size image

To improve the production of recombinant proteins, we next sought to disrupt the most abundant secreted proteins based on a rank-ordering of our initial analysis and using a previously reported, host-informed strategy for CRISPR-Cas9 genome editing in K. phaffii [18, 19]. We used this tool to serially disrupt several of the most abundant secreted proteins. We encountered engineering challenges in this approach, however, including in-frame deletions in disrupted genes, and were unable to disrupt two of the first five targeted genes (Fig. 1B). (GQ67_01286 is a homolog of the gene ayr1 in S. cerevisiae—ayr1 and tdh3 are both non-essential in S. cerevisiae (Saccharomyces Genome Database).) We posited that identification of essential genes—particularly unannotated essential genes—would streamline further engineering of K. phaffii. We therefore sought to identify which genes among the identified secretome of K. phaffii are essential in laboratory conditions (under standard conditions for fermentation).

Identification and knockout of non-essential secreted proteins

We aimed to evaluate the essentiality of the genes encoding the secretome of K. phaffii in parallel. One key innovation in our previously reported CRISPR-Cas9 tool was the reduction of the number of nucleotides that must be replaced in a single guide RNA (sgRNA) cassette to retarget cleavage of DNA by Cas9, enabling pooled synthesis of sgRNA libraries [18]. We created a pooled library of sgRNAs for CRISPR-Cas9-mediated disruption of all genes in the secretome (Fig. 2A) and from the resulting screen, calculated an “essentiality” score for each one (see Additional file 1, Table S1). Genes that had the highest essentiality scores included ribosome subunits (rpl16a), translation factors (tef1), and essential enzymes (tdh3/gapdh, pgk1). We performed weighted gene set enrichment analysis (GSEA) on the essentiality scores of all 361 genes and identified several gene sets enriched with essential secretome genes including carbohydrate metabolism, translation, and membrane transport (Fig. 2B). These observations suggested that the screen with the secretome-directed library was successful for scoring the essentiality of those genes in K. phaffii.

Fig. 2
figure 2

Identification and knockout of non-essential secreted proteins. A Schematic of CRISPR-Cas9 knockout screen to determine essentiality. B Gene set enrichment analysis of genes in the K. phaffii secretome weighted by their essentiality score. C Plot of gene essentiality and relative abundance of proteins in production phase culture supernatant. Genes with an essentiality score of > 0.5 were considered likely to be essential. D Knockout efficiencies of sequential disruption of non-essential genes in the K. phaffii secretome. Knockout genotypes were determined by Sanger sequencing of the targeted locus for up to 16 colonies per transformation

Full size image

We next prioritized a short list of non-essential genes from the secretome as engineering targets. We filtered the secretome for 61 genes with an essentiality score less than 0.5 and with gene products experimentally detected in the culture supernatant (Fig. 2C). Then, we removed seven genes from the potential target pool known to contribute to methanol metabolism since disruption of these genes may affect cellular function during methanol-induced protein secretion (though appear non-essential when cultured on solid glucose medium). Finally, we constructed and separately transformed CRISPR-Cas9 vectors with multiplexed sgRNAs targeting up to four genes per vector (Fig. S5). These 14 vectors targeted all 54 gene targets at least once. On the first attempt, we observed disruption of 20 of the 54 gene targets. We hypothesized that certain combinations of multiplexed knockouts may cause unforeseen synthetic lethality. It may be feasible to disrupt more of these genes with further optimization. We chose to proceed, however, with engineering strains based on the 20 gene disruptions observed.

We next attempted to combine many disruptions to create strains with a reduced secretome. We performed gene disruptions sequentially in two lineages (S∆3a (∆tos1, ∆crh1, ∆exg1) and S∆3b (∆tfs1, ∆lsc2, ∆gq67_05326)), and then combined these sets to construct a new strain S∆6 (∆tos1, ∆crh1, ∆exg1, ∆tfs1, ∆lsc2, ∆gq67_05326). We also extended the S∆3a lineage to construct S∆9 with six additional disruptions (∆gq67_02848, ∆tef4, ∆cts1, ∆gq67_01368, ∆gq67_03274, ∆msc1). We added ∆lsc2 and ∆gq67_05326 to S∆9 to create S∆11, but we were unable to disrupt tfs1 in S∆11. This observation suggested that tfs1 may confer synthetic lethality with another disrupted gene in S∆11.

Throughout this engineering process, we noticed high disruption efficiencies (typically 80–100% of colonies were disrupted). During construction of S∆11, for example, we combined all 11 knockouts without screening more than 16 colonies at each step (Fig. 2D). We attributed this engineering efficiency to the additional knowledge of gene essentiality used to guide the selection of the targeted genes. We next sought to assess the utility of these engineered strains for production of recombinant proteins.

Productivity and growth of secretome-deficient strains

We evaluated the secreted productivity of S∆3a, S∆3b, S∆6, and S∆11 compared to the parent strain (Table 1). We first transformed strains with a vector enabling the secreted expression of human serum albumin (HSA), a 67 kDa protein using the methanol-responsive promoter PAOX1. We cultivated cells in glycerol-containing medium to build biomass, induced expression of the recombinant gene by replacing the medium with a methanol-containing one for 24 h, and evaluated the extracellular protein titer (Fig. 3A-B). Surprisingly, while neither strain with three knockouts exhibited a significant change in specific productivity, the strain with all six knockouts (S∆6) exhibited a ~20-fold increase in protein titer and specific productivity (protein titer normalized to the biomass of the culture based on measured optical density).

Table 1 Engineered strains with reduced secreted proteins
Full size table
Fig. 3
figure 3

Productivity and growth of engineered knockout strains. A Secreted titer and specific productivity of HSA from 3 mL microplate cultures. B SDS-PAGE of 3 mL microplate culture supernatant. HSA protein is visible at ~70 kDa. C Enriched gene sets between S∆6 and the base strain during production of albumin. D Secreted titer and specific productivity of trastuzumab in 3 mL microplate cultures. E Secreted titer and specific productivity of trastuzumab in 100 mL shake flask cultures. F Enriched gene sets between S∆6 and the base strain during production of trastuzumab. G Growth rates of strains in 200 µL cultures in glycerol outgrowth medium. H Growth rates of strains in 200 µL cultures in methanol production medium. In all bar plots, significance of specific productivity or exponential growth rate compared to the base strain was determined by unpaired Welch’s t-test

Full size image

To assess the differences between the S∆6 and wildtype (control) strains producing HSA, we analyzed the transcriptional states of the cells by RNA-seq during recombinant protein expression. Genes related to ribosomal processing and translation were upregulated in the base strain, while genes related to cell division and genome replication were upregulated in S∆6 (Fig. 3C, Table S2). We hypothesize that the engineering changes in S∆6 may alleviate the translational burden experienced by the base strain, either by reduction of the overall translational load from knockout of abundant secreted proteins, or by specific functions of the disrupted genes. We also evaluated the expression of three chaperones that are commonly used as markers for endoplasmic reticulum stress (pdi1, ero1, kar2). We observed higher expression of pdi1 and kar2 in S∆6 compared to wild type during expression of HSA (Fig. S6). We hypothesize that translational capacity in S∆6 has been subsumed by the recombinant HSA, which may lead to secretion-related stress.

Given the improved production of HSA, we hypothesized that S∆6 may also have benefits for producing other large proteins (such as mAbs). We evaluated the secreted titer of trastuzumab, a mAb used to treat HER2 + breast cancer. S∆6 exhibited a ~ 30% increase in specific productivity compared to the wildtype strain in 3 mL cultures (p = 0.002, unpaired Welch’s t-test) (Fig. 3D). To evaluate the performance of engineered strains at higher cell densities, we cultivated the base strain and all four engineered strains producing trastuzumab in 100 mL cultures in shake flasks. At this scale, all strains reached an optical density of 40–50 OD600 after one day of production. In these growth conditions, all the engineered strains secreted two-fold more trastuzumab than the base strain, particularly S∆6 and S∆3a (Fig. 2E).

We also assessed the gene expression of S∆6 and the base strain during expression of trastuzumab. Like the strains that produced HSA, we observed that genes related to ribosomal and RNA synthesis were upregulated in the base strain, while genes related to cell and genome replication were upregulated in S∆6 (Fig. 2F, Table S2). An overall reduction of the translational load may also improve production of trastuzumab, similar to the results for producing HSA. We again observed higher expression of pdi1 and kar2 in S∆6 compared to wild type when expressing trastuzumab (Fig. S6). Interestingly, strains expressing trastuzumab had overall higher expression of all three chaperones pdi1, ero1, and kar2. We hypothesize that secretion of trastuzumab is also limited by other secretory process such as protein folding.

Finally, we measured the rate of growth by seeding engineered antibody-producing strains at low density in 200 µL cultures. Interestingly, in the glycerol-containing media used to accumulate biomass, S∆6 exhibited a higher growth rate than the base strain while S∆11 exhibited a lower growth rate (Fig. 2G). Similarly, in the methanol-containing media used to induce expression of the recombinant protein, we observed higher growth rates for S∆3a, S∆3b, and S∆6, and a lower growth rate for S∆11 (Fig. 2H). These results, together with the observed improvement in recombinant protein titers, demonstrate that strains of K. phaffii with a reduced secretome can improve the secreted productivity of multiple proteins relevant for biopharmaceutical and vaccine products without a decrease in growth rate compared to the base strain.

Discussion

We observed the largest increase in protein production in the strain S∆6. This strain showed reduced expression of genes related to translation and synthesis of ribosomes. S∆6 may have an increased cellular capacity for translation of the recombinant protein due to less translational demand from the native proteome. We also observed increased growth rates by S∆3a, S∆3b, and S∆6 during production of trastuzumab. Translational capacity or the availability of amino acids may represent a general limitation for yeasts during recombinant protein production, therefore [32,33,34]. This hypothesis is corroborated by another engineered strain with an upregulated translation factor that exhibited improved secreted productivity by expanding the cellular capacity for translation [35]. Strategies to further redirect translational capacity towards the recombinant product of interest warrant further investigation.

The improved productivity observed with S∆6 may also result from the functions of specific disrupted genes or combinations of disrupted genes [36]. We did not perform comprehensive combinatorial studies to determine how each individual disrupted gene affects the secretion of recombinant proteins. The gene tfs1, disrupted in S∆3b, S∆6, and S∆11, may improve secretion of recombinant proteins by reducing the amount of protein directed towards the vacuole—a common degradation pathway for heterologous proteins [37, 38]. Similarly, three genes disrupted in S∆6 are involved in construction of the yeast cell wall. Cell wall proteins are abundantly secreted and may consume a large fraction of amino acid, translational, and secretory resources [39]. Disruption of the physical cell wall may also facilitate diffusion of large proteins through the cell wall and into the extracellular space [40, 41]. Deeper understanding of the impact of vacuolar and cell wall-related genes on recombinant protein secretion may inform further engineering.

In K. phaffii, 108 of the 361 proteins in the secretome are described as hypothetical proteins (Table S1), and only 218 proteins in the secretome have homologs in the model yeast S. cerevisiae [23, 42]. We performed a pooled CRISPR-Cas9 knockout screen to predict the essentiality of the secretome. With knowledge of essentiality, we successfully disrupted seven unannotated genes without additional effort or screening (Fig. S5). The predicted gene essentiality documented here will facilitate engineering of other pathways and functions in K. phaffii without the need for further pooled screening.

The sgRNA library used here targeted only one gene per cell and thus was unable to predict synthetic interactions between disrupted genes. Indeed, the S∆11 strain exhibited reduced growth rates during production of trastuzumab, and we encountered synthetic lethality after disruption of 11 genes in sequence. We previously demonstrated that the sgRNA library design used here is compatible with multiplexed gene editing, which will enable pairwise or higher multiplexed knockout libraries in the future.

High-throughput functional genomics tools such as transposon libraries, oligo-mediated recombineering, and Cas9-mediated knockout or upregulation libraries are widely applied to model hosts such as E. coli and S. cerevisiae, including multiplexed libraries [43]. When paired with high-throughput screens or selections, pooled genetic libraries enable identification of genes and pathways that may be tractably engineered to impact the desired phenotype [44]. Pooled screening would be especially useful in non-model microbial hosts in which the functions of many genes are unknown [23, 45, 46]. The sgRNA library design described here leverages native host tRNAs, which makes this approach a general strategy for pooled screening in non-model microbial hosts for production of recombinant proteins such as K. phaffii, Trichoderma reesei, Hansenula polymorpha, and Aspergillus oryzae [18, 47].

Conclusion

In this study, we engineered four new strains of K. phaffii with improved productivity of recombinant proteins. One strain in particular, S∆6, exhibited large improvements in extracellular titer of HSA (~ 20x) and trastuzumab (~ 2x) without a reduction in growth rate. To enable this engineering, we performed the first pooled CRISPR-Cas9 screen in K. phaffii to predict the essentiality of all secreted proteins. This knowledge of essential genes will facilitate future engineering efforts in K. phaffii and will enable pooled functional screening.