PreprintPDF Available

A reconstruction of the mammalian secretory pathway identifies mechanisms regulating antibody production

November 2024

DOI:10.1101/2024.11.14.623668

Authors:

Preprints and early-stage research may not have been peer reviewed yet.

The secretory pathway processes >30% of mammalian proteins, orchestrating their synthesis, modification, trafficking, and quality control. However, its complexity— spanning multiple organelles and dependent on coordinated protein interactions—limits our ability to decipher how protein secretion is controlled in biomedical and biotechnological applications. To advance such research, we present secRecon—a comprehensive reconstruction of the mammalian secretory pathway, comprising 1,127 manually curated genes organized within an ontology of 77 secretory process terms, annotated with functional roles, subcellular localization, protein interactions, and complex composition. Using secRecon to integrate multi-omics data, we identified distinct secretory topologies in antibody-producing plasma cells compared to CHO cells. Genes within proteostasis, translocation, and N-glycosylation are deficient in CHO cells, highlighting them as potential engineering targets to boost secretion capacity. Applying secRecon to single-cell transcriptomics and SEC-seq data, we uncovered secretory pathway signatures underlying secretion diversity among IgG-secreting plasma cells. Different transcriptomic clusters had unique secretory phenotypes characterized by variations in the unfolded protein response (UPR), endoplasmic reticulum-associated degradation (ERAD), and vesicle trafficking pathways. Additionally, we discovered specific secretory machinery genes as new markers for plasma cell differentiation. These findings demonstrate secRecon can identify mechanisms regulating protein secretion and guide diverse studies in biomedical research and biotechnology. Graphical Abstract

Ontological Classification and Gene Involvement in the Secretory 141 Pathway. (A) Sunburst plot showing the hierarchical structure of our secretory pathway 142 ontology. This ontology consists of 77 distinct terms categorized under five primary 143 systems: translocation, protein conformation, post-translational modifications, 144 proteostasis, and vesicle trafficking. Each system is further subdivided into subsystems, 145 processes, and subprocesses, reflecting the nested organization. (B) The UpSet plot 146 demonstrates the overlap and interconnectivity of genes involved in various secretory 147 pathway systems. Each column represents a specific combination of processes, and the 148 height of the bar indicates the number of genes shared among those processes. The first 149

…

Figures - uploaded by Pablo Di Giusto

Content may be subject to copyright.

Content uploaded by Pablo Di Giusto

Content may be subject to copyright.

A reconstruction of the mammalian secretory pathway

identifies mechanisms regulating antibody production

Helen Masson1,*, Jasmine Tat,1,*, Pablo Di Giusto2,*, Athanasios Antonakoudis3, Isaac

Shamie4, Hratch Baghdassarian4, Mojtaba Samoudi2, Caressa M. Robinson1,2, Chih-

Chung Kuo1, Natalia Koga1,2 Sonia Singh1,2, Angel Gezalyan1,2, Zerong Li2, Alexia

Movsessian1,2, Anne Richelle5, Nathan E. Lewis1,2,6,¶

1 Department of Bioengineering, University of California, San Diego

2 Department of Pediatrics, University of California, San Diego

3 Sartorius Corporate Research, Royston, UK

4 Bioinformatics and Systems Biology Program, University of California, San Diego

5 Sartorius Corporate Research, Brussels, Belgium.

6 Center for Molecular Medicine, Complex Carbohydrate Research Center, and

Department of Biochemistry and Molecular Biology, University of Georgia, Athens, GA

* These authors contributed equally

¶ Corresponding author: Nathan E. Lewis

Email: nlewisres@ucsd.edu

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Graphical Abstract

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Abstract

The secretory pathway processes >30% of mammalian proteins, orchestrating

their synthesis, modification, trafficking, and quality control. However, its complexity—

spanning multiple organelles and dependent on coordinated protein interactions—limits

our ability to decipher how protein secretion is controlled in biomedical and

biotechnological applications. To advance such research, we present secRecon—a

comprehensive reconstruction of the mammalian secretory pathway, comprising 1,127

manually curated genes organized within an ontology of 77 secretory process terms,

annotated with functional roles, subcellular localization, protein interactions, and complex

composition. Using secRecon to integrate multi-omics data, we identified distinct

secretory topologies in antibody-producing plasma cells compared to CHO cells. Genes

within proteostasis, translocation, and N-glycosylation are deficient in CHO cells,

highlighting them as potential engineering targets to boost secretion capacity. Applying

secRecon to single-cell transcriptomics and SEC-seq data, we uncovered secretory

pathway signatures underlying secretion diversity among IgG-secreting plasma cells.

Different transcriptomic clusters had unique secretory phenotypes characterized by

variations in the unfolded protein response (UPR), endoplasmic reticulum-associated

degradation (ERAD), and vesicle trafficking pathways. Additionally, we discovered

specific secretory machinery genes as new markers for plasma cell differentiation. These

findings demonstrate secRecon can identify mechanisms regulating protein secretion and

guide diverse studies in biomedical research and biotechnology.

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Keywords

secRecon; Secretory pathway; Plasma cells; CHO cells; multi-omics; SEC-seq

Introduction

Protein secretion is a fundamental biological process in all living organisms

coordinated via the secretory pathway, a highly coordinated network of processes

responsible for synthesizing, processing, and delivering products of approximately one-

third of all protein coding genes in mammals1,2 including both secreted and membrane-

bound proteins. These proteins are critical for maintaining cellular homeostasis, cell

signaling, and intercellular communication3,4. The secretory pathway in mammalian cells

is hosted in a series of membrane-bound organelles and transport vesicles that facilitate

the trafficking of proteins and lipids between these organelles5. The process starts with

the synthesis of proteins in the endoplasmic reticulum (ER), where they undergo folding,

assembly, and post-translational modifications. Proteins are then transported to the Golgi

apparatus, where they are further modified and sorted6. Finally, the proteins are delivered

to their designated locations, e.g., the plasma membrane, lysosomes, or the extracellular

space, through the trans-Golgi network6.

Understanding the secretory network is critical to the study of diverse diseases and

efforts to produce life-altering biotherapeutics. Specifically, in the biotechnology industry,

the secretory pathway is harnessed for manufacturing diverse recombinant proteins, from

therapeutic monoclonal antibodies to industrial enzymes7. Thus, optimizing protein

secretion in host cells, including mammalian cells, is crucial for enhancing protein

production yield and quality8,9. Furthermore, understanding protein secretion has

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

important implications in immunology, as the immune response is mediated by the

release of cytokines, chemokines, and other signaling molecules 10. Disruptions in protein

secretion pathways can lead to various pathological conditions, such as

neurodegenerative diseases, diabetes, and cancer11–14. Therefore, research into the

mechanisms of protein secretion has important biotechnological applications and

provides essential insights into human health and disease6.

To fully comprehend the intricacies of protein secretion, it helps to account for the

function of each molecular component within the pathway15. This includes considering

complex composition, localization, and interaction partners, which can then be organized

as a map that captures the spatial and functional organization of the secretory landscape1.

This further allows researchers to analyze data in the context of the cell, thus more

effectively diagnosing sources of variation in protein secretion using systems biology

tools16.

To enable systems-level analyses of the secretory pathway, previous studies

aimed to enumerate the proteins involved. For example, previous work presented

secretory pathway reconstructions in yeast, mouse, Chinese Hamster Ovary (CHO) cells,

and human cells17–19. Our lab further reconstructed a mechanistic model of the

mammalian secretory pathway, consisting of 261 proteins in CHO cells and 271 proteins

in human and mouse distributed across 12 subsystems1. We further formulated it for

constraint-based modeling, which allowed us to simulate biosynthetic fluxes and quantify

metabolic resource demands for protein secretion. While these reconstructions provide a

solid foundation of the core components of the secretory pathway, they only accounted

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

for less than ¼ of the pathway; thus, we aimed to expand these models to include more

components and detailed annotations of the secretory pathway.

Here we introduce secRecon, a comprehensive reconstruction of the mammalian

secretory pathway that considerably enhances previous models by expanding both

coverage and depth. secRecon integrates extensive information from literature and public

databases, encompassing 1,127 manually curated genes organized within a functional

ontology of 77 secretory pathway processes. A key finding from our analysis using

secRecon reveals that the topology of the secretory pathway is predominantly organized

by functional associations rather than subcellular localization, highlighting the central role

of biological processes in structuring the pathway. This knowledgebase facilitates

exploration of disease mechanisms involving altered secretion and supports the targeted

engineering of cell factories for biotherapeutic production. To demonstrate its utility, we

applied secRecon to multi-omics datasets of CHO and human plasma cells, uncovering

conserved and distinct secretory pathway features that influence protein secretion.

Additionally, analysis of single-cell SEC-seq data using secRecon revealed key secretory

processes driving plasma cell differentiation and IgG secretion heterogeneity.

100

Consequently, secRecon emerges as a valuable resource for dissecting complex cellular

101

processes at both bulk and single-cell levels, addressing a wide range of biological and

102

biotechnological questions.

103

Results

104

secRecon contains a comprehensive annotation of 1127 secretory

105

machinery genes in the mammalian secretory pathway

106

Previous reconstructions of the mammalian secretory pathway identified genes

107

involved in Human18,1, Mouse1,17, and Chinese hamster1,17. In this work, we compiled and

108

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

curated these gene sets to generate a consensus reconstruction with expanded scope

109

and depth through extensive literature review and database curation. We meticulously

110

assessed each gene for its role in protein secretion (Supplementary Figure 1). During

111

this process, we excluded certain gene subsets (Supplementary Table S1), such as

112

transcription factors and genes involved in lipid and cholesterol metabolism or cell cycle

113

processes, because they had been previously included solely based on protein-protein

114

interactions identified through STRING17, without direct evidence of involvement in the

115

secretory pathway. Other genes were removed upon confirming limited evidence in the

116

literature supporting their role in secretion. Recognizing the crucial role of glycosylation

117

in protein structure and function, we included an additional 725 genes from the

118

GlycoGene DataBase20. We further enriched our dataset by incorporating information

119

from databases such as CORUM21, UniProt22, STRINGdb23 and the OrganellesDB from

120

the Chan Zuckerberg Biohub24. These resources provided detailed annotations of gene

121

aliases, orthologs, protein complex interactions, subcellular localization, tissue specificity,

122

and transmembrane domains (Supplementary Figure 1C). Our resulting knowledge

123

base, named secRecon, comprises 1127 genes, all validated through scientific literature

124

and comprehensively annotated using well established databases (Supplementary

125

Figure 1, Supplementary Table S1).

126

During the curation of this knowledgebase, we classified each gene into one or

127

more categories within a newly developed ontology specific to the secretory pathway and

128

assigned a relevance score to each assignment (Figure 1A, Supplementary Table S2).

129

This ontology comprises 77 distinct terms organized hierarchically, where subprocesses

130

are nested within processes, which are further nested within subsystems, all

131

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

encompassed by five primary systems: translocation, protein conformation, post-

132

translational modifications, proteostasis, and vesicle trafficking. Among these primary

133

systems, the most gene-rich categories were Vesicle Trafficking (385 genes), Post-

134

translational Modifications (273 genes), and Proteostasis (228 genes). In contrast, Protein

135

Conformation and Translocation encompassed fewer genes, with 50 and 17 genes

136

respectively. These numbers represent the count of genes assigned exclusively to a

137

single system. Additionally, approximately 14% of the genes in secRecon (174 genes)

138

are associated with multiple systems (Figure 1B), underscoring the inherent

139

interconnectivity of the secretory pathway.

140

Figure 1. Ontological Classification and Gene Involvement in the Secretory

141

Pathway. (A) Sunburst plot showing the hierarchical structure of our secretory pathway

142

ontology. This ontology consists of 77 distinct terms categorized under five primary

143

systems: translocation, protein conformation, post-translational modifications,

144

proteostasis, and vesicle trafficking. Each system is further subdivided into subsystems,

145

processes, and subprocesses, reflecting the nested organization. (B) The UpSet plot

146

demonstrates the overlap and interconnectivity of genes involved in various secretory

147

pathway systems. Each column represents a specific combination of processes, and the

148

height of the bar indicates the number of genes shared among those processes. The first

149

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

five columns represent the number of genes involved in unique systems, while the

150

remaining columns represent genes shared by more than one system. The majority of the

151

genes belong to vesicle trafficking, proteostasis and post-translational modifications. The

152

plot also reveals significant gene overlap, underscoring the multifunctional roles of many

153

genes within the secretory pathway.

154

155

The variations in gene counts among the primary systems reflect the complexity

156

and breadth of the subsystems, processes, and subprocesses they encompass (See

157

Supplementary Table S2). For instance, Vesicle Trafficking, the most gene-rich system,

158

includes numerous subsystems and processes such as pre-Golgi and post-Golgi

159

trafficking, vesicle budding, membrane fusion, and cytoskeletal remodeling. Similarly,

160

Post-Translational Modifications covers a wide array of modifications including

161

glycosylation (with multiple types and pathways), lipidation, phosphorylation, and disulfide

162

bond formation. Proteostasis involves diverse processes including autophagy, UPR

163

signaling pathways, ER-associated degradation (ERAD), and calcium homeostasis. In

164

contrast, Protein Conformation and Translocation involve more specialized and narrowly

165

focused processes.

166

167

Topological analysis of secRecon shows the secretory pathway is first

168

organized by function, followed by localization

169

One might assume that the primary determinant of the secretory pathway topology

170

is subcellular localization—given that proteins localize to specific membrane-bound

171

compartments (e.g., ER, Golgi apparatus, lysosomes, and vesicles). However, we

172

hypothesized that functional associations might play a more significant role in organizing

173

the topology of the secretory pathway. To explore this, we utilized the extensive

174

annotation in secRecon to integrate functional annotations, subcellular localization,

175

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

protein-protein interactions (PPIs), and protein complex information (Figure 2,

176

Supplementary Table S1, Supplementary Table S3).

177

Using the Fruchterman-Reingold force-directed algorithm25, which positions nodes

178

(genes) based on their connections, we constructed networks where edges represent

179

shared processes and protein complexes (Figure 2A). As expected, the algorithm

180

naturally clustered the genes into well-defined groups according to specific secRecon

181

systems, such as vesicle trafficking, proteostasis, and others (Figure 2A, left panel).

182

Within these major clusters defined by secRecon systems, genes further exhibited

183

organization based on their subcellular localization, as visible in the plot. This pattern is

184

particularly evident for genes associated with the Golgi apparatus, proteasome, and ER-

185

Golgi intermediate compartment (ERGIC), among others (Figure 2A, right panel). The

186

co-localization of functionally related genes within the same subcellular compartment

187

allows for spatial organization and optimization of metabolic and signaling pathways26,27.

188

We next visualized the protein-protein interaction (PPI) topology of the network

189

obtained from STRINGdb (Figure 2B), which captures physical interactions underlying

190

virtually every cellular process, from signal transduction to metabolic and secretory

191

pathways28,29. The PPI network for secRecon shows genes cluster by shared functions in

192

specific biological processes. Indeed, the Fruchterman-Reingold force-directed algorithm,

193

applied to the PPI data, clusters genes into the major secRecon systems (Figure 2B, left

194

panel), suggesting protein interactions are tightly correlated with secretory function.

195

Similarly, when the network is annotated by subcellular localization,

196

197

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

198

Figure 2. Network-Based Representation of secRecon Highlights Functional and

199

Interaction Topologies in the Mammalian Secretory Pathway: Each one of the four

200

panels contains an undirected node-link graph representation of secRecon with edges

201

depicting shared processes (A) or protein-protein interactions (B). Left panels display

202

nodes colored by the systems in secRecon and right panels display nodes colored by

203

subcellular localization. The normalized mutual information (NMI) score displayed at the

204

bottom left of each network indicates the degree of alignment between the network's

205

community structure and either the secRecon systems (left panels) or subcellular

206

localization categories (right panels). A higher NMI score signifies a stronger correlation

207

between the network clustering and the respective annotation.

208

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

clusters of genes that co-localize within the same cellular compartments are observed

209

(Figure 2B, right panel).

210

To assess how much of the clustering can be attributed to system-level

211

organization versus subcellular localization, we performed Louvain community detection

212

across a range of resolution parameters. We evaluated the resulting community

213

structures using three key metrics: modularity, normalized mutual information (NMI) with

214

subcellular localization, and NMI with system categories. By analyzing these metrics

215

through elbow plots, we identified optimal resolutions that revealed a clear trade-off

216

between modularity and the NMI scores for systems and subcellular localization

217

(Supplementary Figure 2).

218

In the functional topology network, the NMI score with system categories reached

219

0.76, while the NMI with subcellular localization remained relatively low at 0.21 (Figure

220

2A, Supplementary Figure 2A). This outcome aligns with our expectations since the

221

network was constructed primarily based on functional annotations from secRecon. As

222

for the PPI topology network, the NMI score with system categories was 0.43, slightly

223

higher than the NMI with subcellular localization at 0.35 (Figure 2B, Supplementary

224

Figure 2B). These results further support the notion that system-level organization is a

225

stronger determinant of network structure than subcellular localization.

226

These findings indicate that the community structures of both networks are more

227

closely aligned with system-level organization than with subcellular localization,

228

reinforcing our hypothesis that functional associations play a more significant role in

229

organizing the secretory pathway. The stronger correlation between community structure

230

and system categories suggests that the functional roles of proteins—such as their

231

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

involvement in vesicle trafficking or proteostasis—are more predictive of the network's

232

organization than their spatial localization. This supports the idea that biological

233

processes, rather than compartmental localization alone, drive the structural and

234

functional coherence of the secretory pathway. Both network plots representing functional

235

and PPI topologies highlight how secRecon encapsulates our current understanding of

236

the mammalian secretory pathway in a biologically relevant format, offering a valuable

237

resource for systemic analysis of mammalian protein secretion.

238

239

Analysis of multi-omic data with secRecon identifies secretory pathway

240

signatures linked to antibody secretion in CHO and Plasma cells

241

The biotherapeutics industry is rapidly expanding, with drugs spanning a wide

242

array of therapeutic modalities30. However, increasing demands for complex biologics

243

(e.g., multispecific antibodies, fusion proteins, etc.), presents a considerable challenge

244

for efficient and scalable production. CHO cells remain the gold standard mammalian host

245

system for manufacturing most therapeutic proteins, since they can secrete large

246

molecules with human-compatible post-translational modifications31. However, their

247

epithelial-like origin does not inherently equip them for high secretion. By contrast, plasma

248

cells are specialized immune cells that secrete vast quantities of antibodies32, offering a

249

natural model to advance cell line engineering to maximize protein secretion in CHO cells.

250

To identify important features of plasma cells missing in CHO, industry-standard

251

monoclonal antibody-producing CHO cell lines (CHO-DG44-mAb1 and CHO-K1-mAb2)

252

were compared against four plasma cell-derived (PCD) lines of murine origin (MPC-11,

253

P3X63Ag8) and human origin (JK-6L, Karpas-25) in a multi-omics analysis33. Given the

254

focal point of the secretory pathway in these comparisons, we explored if secRecon could

255

offer insight on the global secretory topology and facilitate process-level network analyses

256

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

of differentially expressed secretory machinery, thus, revealing avenues for future CHO

257

cell line optimization.

258

The expression of the secretory pathway topologically deviates between

259

human, murine, and CHO cells.

260

We first questioned if the topology of the secretory transcriptome and proteome

261

differed between plasma cells and CHO cells. Pairwise correlations of secRecon gene

262

expression between the cell lines revealed high intra-species correlation (r > 0.9) and

263

moderate correlation (r ~ 0.7) between CHO and plasma cell lines (Figure 3A). At the

264

proteome scale, a similar trend was observed, though with lower correlations (intra-

265

species r > 0.7; inter-species r ~ 0.5) (Figure 3C). When comparing hierarchical clustering

266

of secRecon gene expression, we found that CHO cell gene-level dendrograms

267

moderately correlated with those of murine and human plasma cells, while the murine

268

and human plasma cell topologies surprisingly exhibited lower similarity (Figure 3B).

269

Protein-level dendrograms between CHO and murine plasma cells, however, showed

270

considerable divergence (Figure 3D). These observations suggest that while the

271

expression ranges of secretory machinery are largely conserved within and across

272

species, the wiring and regulation of specific components may have evolved differently

273

between species.

274

The expression of the secretory machinery is more similar at the pathway

275

level across species.

276

To contextualize how differential wiring of individual machinery genes may impact

277

overall secretory pathway activity and topology, we applied the secRecon ontology to

278

score secretory geneset activity using gene set variability analysis35 (GSVA). As

279

expected, correlations at the pathway level using transcriptomic and proteomic GSVA

280

scores were overall stronger than at individual gene- or protein-level correlations, with

281

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

intra-species correlations exhibiting the strongest correlations (Figure 3E, 3G). Pathway-

282

level topology derived from GSVA score clustering also revealed greater interspecies

283

similarity compared to individual gene- or protein-level clustering (Figure 3F, 3H).

284

Notably, the proteome pathway-level topology between CHO and murine plasma cells

285

were highly similar, despite their greater dissimilarity at the protein level (Figure 3D, 3H).

286

Figure 3. Pairwise correlation and clustering topology of secRecon transcriptome

287

and proteome for antibody secreting CHO and plasma cell lines. Spearman rank

288

correlation of individual secRecon (A) gene expression and (C) protein abundance was

289

performed between each cell line, colored by CHO, murine, or human cell line origin.

290

Dendrograms derived from hierarchical clustering of secRecon (B) gene expression and

291

(D) protein abundance were correlated using Baker’s gamma index34. Pairwise spearman

292

rank correlation (E,G) and dendrogram correlation (F,H) were similarly conducted at the

293

pathway-level using GSVA35 scoring of secRecon genesets.

294

295

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Altogether, these topological analyses suggest that despite variations in individual

296

secretory machinery components across mammalian species, their activities and

297

regulatory mechanisms tend to converge at the pathway level. This highlights the

298

translational potential of functional and pathway-level knowledgebases, such as

299

secRecon, to study cross-species biological systems. However, when targeting specific

300

genes or resolving subtle variations in secretory regulation, it may be crucial to consider

301

species- and host-specific differences in the wiring of secretory machinery, as indicated

302

by diverging gene and protein-level topologies (Figure 3B, 3D).

303

Secretory pathway signatures primarily localized to ER and Golgi

304

compartments are upregulated in plasma cells.

305

To further study the conservation and divergence of secretory mechanisms in

306

plasma and CHO cells, we overlaid the differential secretory transcriptome and proteome

307

onto secRecon's Functional and PPI networks (Figure 4, Supplementary Figure 3). We

308

found that plasma cells exhibit significant upregulation of the global secretory

309

transcriptome relative to CHO cells, in particular for machinery localized to the ER and

310

Golgi apparatus (Figure 4A). By overlaying the data on PPI networks, we found that

311

upregulated genes clustered around "Proteostasis" and "Protein Conformation" with

312

many genes localized to the ER (Figure 4C). Similarly, the secretory proteome was

313

substantially upregulated in plasma cells compared to CHO cells, with dominant

314

upregulated functional clusters mirroring those observed in the transcriptome

315

(Supplementary Figure 3A). However, in the PPI network analysis, a pronounced

316

central cluster of interactions showed increased enrichment of "Protein Conformation"

317

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

318

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Figure 4. Network-based Transcriptomic Analysis of Plasma Cells vs CHO Cells:

319

Log2 fold change in gene expression between plasma and CHO cells were overlaid on

320

secRecon. Differentially expressed genes were visualized in separate network plots for

321

upregulated (A and C) and downregulated (B and D) genes in plasma cells. Panels A

322

and B display these genes in a network layout based on the secRecon system ontology,

323

where nodes are colored according to the major secretory pathway systems (left), or their

324

subcellular localization (right). Panels C and D show the same sets of genes plotted in a

325

network representation based on protein-protein interactions, with node colors according

326

to the major secretory pathway systems (left), or their subcellular localization (right). The

327

size of each node corresponds to the magnitude of fold change in expression.

328

329

330

331

proteins and decreased enrichment of "Proteostasis" proteins compared to the

332

transcriptome (Supplementary Figure 3C). Additionally, the proteins associated with this

333

central cluster were predominantly localized in the ER, with a smaller connected

334

subcluster involving genes associated with the ERGIC.

335

Global network analysis, however, may mask the specific secretory pathway

336

signatures of these cells. To provide a more detailed view of the secretory pathway

337

architecture unique to each cell type, we leveraged secRecon GSVA scores to identify

338

differentially enriched secretory gene sets between these cell types. This analysis

339

revealed processes within the "Proteostasis" system (e.g., the “PERK pathway”, “ER

340

stress-induced pre-emptive quality control” (ERpQC)) and “N-glycosylation” within the

341

"Post-translational modification" system to be significantly upregulated in plasma cells

342

(Figure 5A). These processes were similarly upregulated in plasma cells at the protein

343

level, with notable increases in enrichment for “Vesicle budding” within the "Vesicle

344

Trafficking" system and “Co-translational translocation” within the “Translocation” system

345

(Figure 5B).

346

347

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Figure 5. Differentially enriched secRecon processes between plasma and CHO

348

cells: Differential testing of secRecon GSVA scores identified differentially enriched

349

secretory processes and subprocesses between plasma and CHO cells at both the

350

transcript (A) and protein (B) levels. In these volcano plots, the x-axis represents the

351

difference in GSVA enrichment scores between plasma cells and CHO cells (positive

352

values indicate higher activity of the geneset in plasma cells), while the y-axis represents

353

the statistical significance of the enrichment (-log10 adjusted p-value). Processes within

354

systems differentially enriched in both transcript and protein analyses are labeled. Log

355

fold-change of secretory machinery annotated under identified processes significantly

356

upregulated in both the transcript or protein enrichments were overlaid on PPI networks

357

for (C) transcripts or (D) protein. The size of each node corresponds to the magnitude of

358

fold change in expression in plasma vs CHO cells (bigger nodes indicate higher

359

expression in plasma cells), while the color of each node indicates the process to which

360

it belongs. Nodes associated with processes not identified in the GSVA scoring analysis

361

are colored in gray.

362

363

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

For processes upregulated in both the transcriptome and proteome we performed

364

PPI-based topological network analyses to study the relationships between the

365

components involved in these processes. Analyzing PPIs can reveal how proteins

366

physically and functionally interact within the cell, and can identify key clusters within the

367

network. The PPI network revealed a tight cluster of components annotated under the

368

“PERK pathway”, “Co-translational translocation” and additional nodes assigned

369

to“ERpQC” at both the transcript and protein levels (Figure 5C, 5D). This suggests that

370

the coordinated activity of these proteins is a key feature distinguishing the secretory

371

machinery of plasma cells from that of CHO cells. Interestingly, both networks also

372

revealed several components related to “N-glycosylation”, with an increased number of

373

these nodes present in the transcriptome (Figure 5C). However, this cluster is not as tight

374

as the previous one, highlighting much less interaction between these proteins. This

375

suggests that the proteins involved in N-glycosylation may function more independently

376

or sequentially, rather than forming tightly interacting complexes. N-glycosylation involves

377

a series of enzymatic steps wherein different glycosyltransferases act on substrates

378

without necessarily forming stable complexes with one another36,37. This contrasts with

379

the “ER stress response”, “UPR”, and “translocation” processes (Figure 5,

380

Supplementary Figure 4), where proteins often assemble into larger complexes to

381

facilitate coordinated actions38,39. These results demonstrate that secRecon can not only

382

identify unique secretory pathway signatures but also effectively contextualize the

383

processes underlying different cellular conditions. Within these enriched processes, we

384

further identified genes differentially expressed at the transcript and protein level

385

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

(Supplementary Figure 5) which could be further studied and engineered to maximize

386

the secretory capacity of industrial CHO cell lines.

387

388

Analysis of single cell omics with secRecon identifies pathways driving

389

plasma cell differentiation and IgG secretion

390

Activated B cells undergo transcriptional, epigenetic, and morphological changes

391

to differentiate into antibody-secreting plasma cells, resulting in a highly heterogeneous

392

cell population40–44. Single-cell omics studies and CRISPR screens are identifying

393

regulators and hallmarks of this process, elucidating how the secretory pathway impacts

394

specific antibody-secreting fates, such as the production of distinct heavy and light chain

395

immunoglobulin subtypes43–46. Thus, here we wondered if secRecon could provide more

396

targeted insights into the secretory machinery’s involvement in plasma cell differentiation

397

and antibody secretion.

398

IgG secreting plasma cells exhibit distinct clusters characterized by unique

399

secretory phenotypes.

400

To answer this, we used SEC-seq data that quantified the single cell IgG secretion

401

and mRNA of activated B cells as they differentiated into antibody-secreting plasma

402

cells47. Interestingly, variation in immunoglobulin heavy and light chain subtype

403

expression, major sources of transcriptomic heterogeneity, were not strongly coupled with

404

the relative abundance of secreted IgG (Figure 6A). To identify markers and pathways

405

predictive of high antibody secretion, high and low IgG secreting plasma cells were

406

aggregated across clusters and pseudotime for differential expression analysis. Unfolded

407

protein response, glycosylation, and mitochondria-associated metabolic processes were

408

among the top pathways enriched in high IgG secretors47. Here, we used our secRecon

409

ontology to calculate average expression48 and relative PPI activity49 per geneset for each

410

cell. The relative contribution of each secRecon geneset in predicting single-cell IgG

411

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

secretion was subsequently quantified using Dominance Analysis50,51. We found 10.9%

412

of single-cell IgG secretion variation was explained by secretory processes (Figure 6B).

413

The most important secretory subsystems contributing to this variation were “UPR”,

414

“ERAD”, “vesicle trafficking”, and “protein folding” systems (Figure 6B, Supplementary

415

Figure 6D). This could be unraveled to more granular processes, specifically the “PERK”

416

and “IRE1” pathways, “ubiquitination”, and “autophagy”, which contributed the most

417

towards explaining IgG secretion variability. Given that metabolic processes were

418

previously enriched in high versus low IgG secreting cells in this dataset47, we found that

419

including pathways beyond secRecon such as oxidative phosphorylation GO-BP52,53

420

geneset explained 1.9 fold more variation than secRecon terms alone (Supplementary

421

Figure 6B, 6C), suggesting that maintaining ATP levels is essential to support energy-

422

intensive processes such as antibody secretion.

423

The IgG-secreting plasma cell population is composed of five distinct Leiden

424

clusters47 (Figure 6C). We speculated that analyzing all IgG secreting cells in aggregate

425

may mask further intra-population secretory phenotypes. To explore this, we applied

426

Dominance Analysis to predict IgG secretion for each cluster individually. Doing this, we

427

resolve “secretory fingerprints” per cluster (Figure 6D), which represent distinct and

428

shared secretory machinery that contribute most to the variation in IgG production in each

429

cluster. For example, IgG secretion for cluster 10 was more driven by “vesicle trafficking”,

430

“post-translational modifications”, and differentiation state (as indicated by pseudotime).

431

In contrast, processes under “UPR”, “ERAD”, and “protein conformation” contribute more

432

towards the other clusters’ secretory phenotypes. Further characterization revealed that

433

while overall canonical expression of secRecon genes is similar across clusters (aligned

434

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

with our findings from bulk RNA-Seq analysis of various plasma and CHO cells Figure

435

3A, 3C), 30-50% of these secRecon genes are differentially expressed between these

436

five IgG secreting Leiden clusters with notable differences in co-expression topologies

437

(Figure 6E, 6F), suggesting that fine tuning of secRecon expression and topological

438

distinctions likely contribute to the respective unique secretory phenotypes.

439

440

Figure 6. Using secRecon ontology to identify secRecon processes and machinery

441

correlating to IgG secretion. (A) SEC-seq data47 were used to link single-cell IgG

442

secretion to transcriptome for a diverse population of B cells. The UMAPs are colored by

443

major sources of single-cell heterogeneity (left to right: Leiden transcriptomic clusters,

444

expression of IGH chain subtype, secreted IgG concentration as quantified by SEC-seq,

445

expression of IGL chain subtype. (B) Dominance analysis quantifies relative importance

446

of secRecon genesets explaining the variation in single-cell IgG secretion for the IgG

447

secreting population. (C) The IgG secreting population is largely composed of 5 distinct

448

subpopulations as annotated by Leiden clustering. (D) Dominance analysis is applied to

449

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

each Leiden cluster to reveal distinct secretory fingerprints associated with IgG secretion

450

heterogeneity. (E) Pairwise secRecon gene expression and (F) gene-level topologies

451

were correlated using Spearman rho and Baker’s Gamma index respectively for

452

pseudobulk populations within each Leiden cluster.

453

454

Expression of secretory pathway subprocesses is strongly associated with

455

plasma cell differentiation

456

Rewiring of secretory machinery is crucial for preparing activated B cells to

457

transition to the stressful antibody-secreting state; although, the associated molecular

458

mechanisms involved remain poorly understood41,42,44. Thus to dissect secretory

459

machinery underlying the differentiation trajectory, we applied Dominance and correlation

460

analysis to predict secRecon genesets and individual genes associated with pseudotime

461

(Figure 7). Specifically, early pseudotime, corresponding to the activated B cell state,

462

was associated with elevated activity in “cytoskeleton remodeling”, “post-Golgi vesicle

463

trafficking”, and “ERAD” processes. In contrast, late pseudotime, corresponding to the

464

mature plasma cell state, was associated with elevated activity of “glycosaminoglycan”

465

post-translational modifications, “ER pre-quality control”, and “Golgi organization”

466

processes (Figure 7C). Previous studies identified UPR and autophagy activation as

467

hallmarks of early antibody-secreting cell (ASC) differentiation, facilitating extensive

468

secretory organelle remodeling such as ER expansion, preventing stress-induced

469

apoptosis, and priming for increased cargo load, whereas disulfide bond formation and

470

protein folding become more prominent in the mature antibody-secreting state40,41,42,44.

471

Interestingly, while transcription of antibody heavy chain genes was moderately

472

correlated with pseudotime (r = 0.28, BH-corrected P < 0.001), the secreted IgG

473

concentration was weakly correlated with pseudotime (r = -0.04 , BH-corrected P <

474

0.001). This suggests that post-transcriptional processes, such as the secretory

475

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

processes identified in our Dominance Analysis, may act independently of well-known

476

differentiation hallmarks to regulate IgG secretion.

477

To further investigate potential markers of differentiation, we performed a

478

secRecon gene-level correlation analysis against pseudotime progression (Figure 7D).

479

Among the top positively correlated genes, we found multiple plasma cell differentiation

480

markers (XBP146,54, FKBP1155, DERL342, PRDX454). Additionally, many of the top

481

correlated genes, such as SSR4, SSR3, SPCS3, KDELR, SPCS2, SAR1B, are annotated

482

in the Human Protein Atlas as plasma cell markers but remain functionally

483

uncharacterized (Figure 7D). Notably, most secRecon genes that significantly correlated

484

with pseudotime are novel or uncharacterized in the context of plasma cell differentiation.

485

These novel targets present opportunities for future CRISPR screens and functional

486

studies to further elucidate the molecular mechanisms underlying plasma cell

487

differentiation.

488

Altogether, we demonstrate that secRecon’s ontology and annotations enable a

489

comprehensive analysis from systems down to the gene level, contextualizing specific

490

secretory processes and machinery underlying single-cell phenotypic heterogeneity.

491

Furthermore, our findings reveal novel markers and pathways that could be of significant

492

interest to the plasma cell research community, potentially unlocking opportunities to

493

further understand plasma cell differentiation and antibody secretion.

494

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

495

Figure 7. Identifying secRecon processes and machinery underlying plasma cell

496

differentiation trajectory: (A) Pseudotime trajectory analysis of SEC-seq data provides

497

a continuous scale of differentiation from activated B cell to antibody secreting plasma

498

cells (PC)47. (B) Dominance analysis identifies predictive secRecon genesets collectively

499

explaining variation along plasma cell differentiation trajectory, as indicated by

500

pseudotime. secRecon (C) subsystems activity and (D) gene expression were correlated

501

to pseudotime, with the highest correlates labeled. Underlined subsystems and genes

502

were identified in existing studies to be associated with plasma cell differentiation. Genes

503

with subscript “H” indicate scRNA plasma cell markers annotated in Human Protein Atlas

504

but not functionally characterized.

505

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Discussion

506

The secretory pathway plays an essential role in shaping the overall function of the

507

cell. Disruptions in this pathway are associated with various pathological conditions11–14.

508

Furthermore, in biotechnology, the secretory pathway is harnessed to manufacture

509

diverse recombinant proteins7. Optimizing protein secretion in host cells, particularly

510

mammalian cells, is crucial for enhancing production yield and quality8,9. Thus, a better

511

understanding of the processes and components within the secretory pathway is critical

512

for advancing both therapeutic protein production and the development of personalized

513

therapies.

514

Reconstructions of the secretory pathway provide comprehensive representations

515

of secretory profiles under specific conditions. Building these models efficiently can

516

support new biomedical and biotechnological applications. However, the accuracy of

517

these reconstructions relies on the depth of annotated data and the quality of the curation

518

of the network. Existing secretory pathway reconstructions1,17,18 provide insightful

519

charters however they vary in depth–such as the level of annotations and hierarchical

520

organization–and coverage, including the number of secretory pathway components.

521

Some integrate protein-protein interaction information, while others provide details on

522

protein complexes. However, none currently provide a comprehensive annotation

523

encompassing all these features in a single framework.

524

Addressing the need for a more detailed, unified, and hierarchically organized

525

model, we introduce secRecon, a comprehensive map of the mammalian secretory

526

pathway, comprising 1127 functionally annotated genes. The reconstruction captures the

527

functional roles of individual genes within specific secretory processes and their spatial

528

localization across cellular compartments, providing a detailed map of where secretory

529

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

machinery operates within the cell. This resource offers mechanistic insights by

530

annotating PPI networks, which can reveal how the pathway machinery collaborates to

531

carry out complex, multi-step processes required for protein synthesis and secretion. We

532

demonstrated here how it can help uncover both direct and indirect relationships between

533

proteins, in both spatially and temporally regulated manner, thus enabling more accurate

534

predictions of how disruptions in these networks may lead to altered secretory

535

phenotypes, disrupted signaling, and disease states. Moreover, both the functional (NDEx

536

UUID: 4e7b9729-722a-11ef-ad6c-005056ae3c32) and PPI topology networks (NDEx

537

UUID: efb15b63-722c-11ef-ad6c-005056ae3c32) are publicly available in the NDEx

538

repository, allowing broader access for researchers to explore and utilize these networks.

539

Applying secRecon to a multi-omics study of diverse antibody-secreting cell lines

540

–including CHO cells, murine, and human plasma cells–we explored the topology of the

541

secretory pathway using RNA-Seq and proteomic data. Despite species-specific

542

differences at the individual gene expression and protein abundance levels, we found that

543

the individual regulation and activity of secretory machinery ultimately converge at the

544

pathway level. By identifying and analyzing upregulated secretory subprocesses in

545

plasma cells within the context of protein-protein interactions, we revealed tightly

546

clustered secretory machinery genes central to the global secretory interaction topology,

547

suggesting their potential roles in regulating plasma cell protein secretion efficiency.

548

Additionally, the enriched subprocesses in plasma cells allowed us to narrow down a list

549

of 23 genes consistent in fold-change magnitudes at both the transcript and protein levels.

550

Further investigation of these specific processes and genes could guide efforts to improve

551

the scalability and efficiency of producing complex biologics, to meet the growing

552

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

demands of the biotherapeutics industry30,56. Moreover, the framework could also support

553

future studies into how the secretory pathway dynamics impact diseases related to protein

554

secretion and cell-cell communication.

555

We then applied secRecon to probe antibody-secreting plasma cells at single-cell

556

resolution. By linking single-cell IgG secretion with transcriptomic profiles for human

557

plasma cells, we leveraged secRecon to shed light on secretory processes and

558

machinery associated with IgG secretion heterogeneity and plasma cell differentiation.

559

This analysis identified specific subprocesses and genes within UPR, ERAD, vesicle

560

trafficking, and protein folding associated with the heterogeneity in IgG secretion, while

561

further resolving cluster-specific secretory phenotypes within the bulk population.

562

Additionally, we uncovered novel insights into the secretory machinery involved in plasma

563

cell differentiation. Although existing studies have identified broad secretory hallmarks

564

associated with early and mature plasma cell differentiation40–42,57,58, system to gene-level

565

mechanistic understanding of the complex secretory process remains largely unresolved.

566

Notably, a substantial portion of variation along the single-cell differentiation trajectory

567

could be explained by secRecon subprocess activity, including cytoskeletal remodeling,

568

ER preemptive quality control, post-Golgi vesicle trafficking, and glycosaminoglycan post-

569

translational modifications. These processes are aligned with findings that the ER and

570

Golgi systems are subjected to extensive remodeling to prime B cells for a highly stressful,

571

antibody-secreting state32,41,42,57,59 while the use of secRecon ontology and Dominance

572

Analysis is able to quantify the distribution of specific, granular processes underlying this

573

trajectory of dynamic cellular remodeling. secRecon gene-level correlations further

574

revealed targets previously identified by the community alongside a substantial number

575

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

of novel candidates strongly associated with plasma cell differentiation. As we were able

576

to identify groups of coordinated secretory machinery signatures among a highly

577

heterogeneous population of plasma cells, it would be valuable to further investigate what

578

and how cell-cell interactions might drive this continuum of states and phenotypes, and

579

to examine whether these correlated secretory machinery patterns are regulated by such

580

signaling. Altogether, we illustrate secRecon's ability to contextualize multi-omics data at

581

multiple resolutions, from secretory systems to individual genes. These analyses

582

underscore the value of secRecon in revealing novel insights into the distinct secretory

583

wiring of different mammalian cell types and potential to unravel the complexity of

584

secretory pathways across diverse biological contexts.

585

While this work established a robust framework to map the complex secretory

586

network, there remain numerous opportunities for further exploration and enhancement

587

of secRecon. One notable opportunity for future exploration is the integration of

588

transcription factors (TFs), which function as regulators of processes within the secretory

589

pathway. To lay the groundwork for future incorporation, we conducted a transcription

590

factor enrichment analysis using our list of curated secRecon machinery (Supplementary

591

Table S4-S6). Moreover, secRecon is positioned to evolve into a comprehensive

592

resource for reconstructing secretory pathway reactions. By expanding on the Genome-

593

Scale Model (GEM) framework proposed previously1 and integrating directional topology

594

from established pathway databases such as KEGG, secRecon can facilitate the

595

development of more sophisticated mechanistic models. Our comparative analysis of

596

secretory topologies in plasma and CHO cells revealed that, although secretory

597

subprocess activities are largely conserved, the regulation and wiring of individual genes

598

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

and proteins may vary across species and down to the level of single-cell resolved

599

clusters. Such consideration can be important when pathway databases (e.g., KEGG,

600

IPA) are used to contextualize omics data and output actionable hits, as resources are

601

often based on model organisms and may miss species- or cell line-specific nuances in

602

secretory regulation. This highlights a potential need for hybrid models that overlay multi-

603

omics data onto protein-protein interaction (PPI) networks to gain deeper, species-

604

specific insights into secretory regulation60–62.

605

Additionally, we recognize potential inherent biases in ontologies and

606

knowledgebases that rely on manual curation and public literature. Indeed, genes with

607

higher global expression levels or localized to specific organelles are more frequently

608

studied and better characterized. This can confound community-wide functional

609

understanding and experimental design63. This systematic bias is a concern across many

610

public network constructions and should be considered when using and interpreting

611

ontologies like secRecon that rely on published studies. While we explored this and did

612

not find any strong associations of secRecon curated metrics with the transcriptomic

613

profiles from the Genotype-Tissue expression project64,65 (GTEx) (Supplementary

614

Figure 7), this does not rule out the presence of study and community bias. AI tools and

615

LLMs could help mitigate these biases by systematically analyzing large amounts of data

616

to identify under-studied genes of phenotypic importances; this can help provide more

617

balanced functional annotations. Future work to enhance the quality of such

618

knowledgebases should further investigate and address such biases, e.g., by

619

experimentally validating groups of secretory machinery functions or interactions using

620

well-defined experiments that span diverse cell types.

621

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Materials and Methods

622

Manual curation of mammalian secretory pathway genes

623

A preliminary list of secretory genes was drafted based on previous

624

reconstructions from Feizi et al.18, Lund et al.17, and Gutierrez et al1. Genes involved in

625

glycosylation were added to the list based on their annotation in the GlycoGene DataBase

626

(GGDB)20. Each gene was manually removed or linked to one or more processes in our

627

secRecon ontology based on literature surveys. Literature-based relevance scores

628

ranging from poor to strong association to secRecon terms were assigned as follows: (1)

629

loose association with secretory pathway, (2) localized to secretory pathway with

630

probable evidence but missing functional information, (3) evidence supporting

631

association/interaction with core secretory machinery, (4) concrete evidence of functional

632

role in the secretory pathway well supported and characterized in literature. Based on the

633

identified literature evidence and curated secRecon ontology term annotations, a

634

comprehensive description of the secretory functional role was manually generated for

635

each gene.

636

The secRecon ontology was manually defined and organized based on extensive

637

literature research of the secretory pathway, resulting in 77 distinct terms

638

(Supplementary Table S2, Supplementary Material S7). Each was categorized under

639

five primary secretory systems: translocation, protein conformation, post-translational

640

modifications, proteostasis, and vesicle trafficking. Each system is further subdivided into

641

subsystems, processes, and subprocesses, reflecting the nested and interconnected

642

organization of the secretory pathway.

643

644

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Ortholog mapping and integration of database annotations

645

Various databases were used to further annotate the secretory pathway genes. A

646

custom Python script (See Supplementary Material; Feature_Extraction.ipynb) was

647

used to map the default human gene symbol (HGNC) to: i) aliases, ii) Ensembl IDs, iii)

648

Entrez IDs, iv) Gene names, and v) UniProt IDs. CHO and Mouse orthologs were mapped

649

using the Entrez API66 and the Human Entrez IDs as input. Subcellular localization was

650

mapped using a two-step process: localization was primarily assigned using organelle

651

immunoprecipitation data generated in Hein et al.24 genes absent from this dataset were

652

then mapped according to subcellular localizations annotated in UniProt using the

653

UniProt IDs as input. Protein complex information was acquired from the CORUM21

654

database and filtered for complexes that contain other secRecon partners or contain a

655

secretory relevant functional description. Interaction partners for each gene were

656

retrieved from the STRING database (See Supplementary Material;

657

PPI_Ontology_Network.ipynb).

658

659

Graph-based representation of secRecon

660

To construct network representations of both the functional topology and protein-

661

protein interaction (PPI) networks of secRecon, the NetworkX67 Python library was used.

662

For the functional topology, each gene is represented as a node in a graph "G". The

663

module iterates over each pair of genes, calculating the number of shared secRecon

664

ontology terms, which includes respective parent systems, subsystems and processes,

665

and whether they share any protein complexes. If a gene lacks associated complexes,

666

the shared complex count is set to zero. Similarly, if no shared terms exist, the shared

667

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

process count is set to zero. An edge is added between each pair of genes, with the edge

668

weight determined by the sum of shared processes and complexes. For visualization,

669

only edges with a count greater than 1 are displayed in the plot. The spatial arrangement

670

of nodes is determined using the Fruchterman-Reingold force-directed algorithm25 or

671

'spring layout', which positions genes with similar processes and complexes closer

672

together.

673

The PPI network was obtained from STRING database23. Gene pairs were filtered

674

based on interaction confidence scores, retaining only high-confidence interactions in the

675

final network. A graph "G" was again generated using NetworkX, where each node

676

represents a gene, and edges correspond to known PPIs between the proteins encoded

677

by these genes. STRING interaction scores were used as weights on the edges. For

678

visualization, the spatial arrangement of nodes was computed using the 'spring layout'

679

algorithm to cluster proteins based on interaction strength. Nodes were colored according

680

to either their involvement in secretory subsystems (e.g., vesicle trafficking, post-

681

translational modifications) or their subcellular localization (e.g., ER, Golgi apparatus).

682

In both networks, nodes are colored consistently according to either their

683

involvement in secretory subsystems (e.g., vesicle trafficking, post-translational

684

modifications) or their subcellular localization (e.g., ER, Golgi apparatus). Each node is

685

displayed as a pie chart, with segments color-coded to represent the gene’s subcellular

686

localization or secRecon systems.

687

688

Community detection analysis to investigate secRecon network topology

689

690

To investigate the organizational principles of secRecon, Louvain community

691

detection68 was performed on the two different network topologies aforementioned. For

692

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

each network, Louvain community detection was carried out across a range of resolution

693

parameters, ranging from 0.5 to 2.5 in increments of approximately 0.1 to identify optimal

694

community partitions. Three metrics were used to assess the quality and biological

695

relevance of the communities detected: (1) Modularity: A measure of the quality of the

696

partitioning, indicating how well-separated the communities are within the network. (2)

697

Normalized Mutual Information (NMI) for subcellular localization: Quantifying the

698

agreement between the detected communities and the known subcellular localization

699

annotations. (3) NMI for system categories: Quantifying the alignment of the detected

700

communities with system-level biological functions, as annotated in secRecon.

701

To determine the optimal resolution that balances high modularity with biologically

702

meaningful community structure, a trade-off analysis was conducted using the following

703

approach: (1) Normalization of Metrics: The modularity scores and NMI scores for both

704

subcellular localization and system categories were normalized to a common scale using

705

min-max normalization.

706

𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑+𝑆𝑐𝑜𝑟𝑒!+ = +𝑆𝑐𝑜𝑟𝑒!− +𝑚𝑖𝑛(𝑆𝑐𝑜𝑟𝑒𝑠)

𝑚𝑎𝑥(𝑆𝑐𝑜𝑟𝑒𝑠) +− +𝑚𝑖𝑛(𝑆𝑐𝑜𝑟𝑒𝑠)

707

(2) Combined Score Calculation: For each resolution parameter, a combined score

708

was calculated by summing the normalized modularity and NMI scores.

709

!"#$%&'()*+",' -.",#/0%1'()2"(30/,%45 6.",#/0%1'().27)89"+/0%1/4%"&: 6 .",#/0%1'().27)8*5;4'#;:

710

(3) Identification of Optimal Resolution: The resolution parameter corresponding

711

to the maximum combined score was identified as the optimal resolution. This resolution

712

represents the best trade-off point where the network exhibits high modularity while

713

maintaining strong alignment with biological annotations.

714

715

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Visualization of Secretory Pathway Ontology

716

A sunburst plot was generated using plotly to visualize the hierarchical structure of

717

the secretory pathway's ontology described above. The dataset containing the ontology

718

of the secretory pathway was organized by System, Subsystem, Process, and

719

Subprocess. Each level in the ontology was used as a path to structure the sunburst plot,

720

with colors representing the different systems. To visualize the overlap of genes involved

721

in multiple secretory pathway Systems, an UpSet plot69 was generated using the

722

upsetplot Python library. A dictionary containing gene annotations across the secretory

723

systems (e.g., "Protein conformation", "Translocation", "Post-translational modifications",

724

"Vesicle trafficking", "Proteostasis") was used to create a binary matrix indicating the

725

presence of genes in each system. The occurrence of different combinations of systems

726

was counted and visualized as an UpSet plot.

727

728

Comparison of secRecon coverage and annotations against currently available

729

reconstructions

730

To compare secRecon with other available reconstructions, secretory pathway

731

genes from existing reconstructions, including Feizi et al.18, Lund et al.17, and Gutierrez

732

et al.1, were mapped to the genes in secRecon. Using the secRecon gene list, mappings

733

between gene identifiers (Human Entrez IDs) and other gene identifiers used in these

734

reconstructions (such as ENSEMBL IDs and CHO Entrez IDs) were created to ensure

735

accurate comparisons. A venn diagram was used to visualize the overlap between the

736

gene sets from these reconstructions, allowing quantification of the shared and unique

737

genes across the different reconstructions.

738

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

739

Multi-omic analysis of CHO and plasma cells using secRecon

740

Normalized transcriptomic and proteomic data33 were subset for secRecon genes

741

and proteins. Spearman correlation was quantified between all samples for secRecon

742

genes and proteins. Gene Set Variation Analysis35 (GSVA) quantifying secRecon

743

subsystem activity was conducted using the complete transcriptomic and proteomic data.

744

Cosine similarity of gene expression, protein abundance, and respective GSVA scores

745

was calculated for all 6 cell lines per species (CHO, murine, human for transcriptomic

746

data, CHO and murine for proteomic data). Hierarchical clustering using average linkage

747

was performed for the cosine similarity matrices to create respective species

748

dendrograms. Baker’s gamma index34 to compare subtree clustering topology was

749

calculated using the R dendextend package. Correlation plots were generated in R using

750

corrplot.

751

For network analysis, both the functional and PPI topology networks were filtered

752

to include only genes present in the transcriptomic and proteomic datasets. The networks

753

were further annotated by overlaying gene expression levels onto the network nodes.

754

Node sizes were scaled according to log2 fold-change in gene expression, enabling a

755

visual representation of expression differences between CHO and plasma cells.

756

757

Single-cell SEC-seq analysis of human ASCs

758

Single-cell gene expression and IgG secretion data were preprocessed,

759

normalized, and clustered using scanpy as previously published47. Geneset lists were

760

defined according to secRecon ontology or GO-BP53 genesets for additional biological

761

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

processes. secRecon subgraph of protein-protein interaction (PPI) network PCNet70 was

762

created by filtering for edges in the total PCNet graph (NDEx UUID: 4de852d9-9908-

763

11e9-bcaf-0ac135e8bacf) that connect 2 gene nodes within secRecon. Single-cell

764

secRecon geneset average expression was calculated using scanpy FeatureExpression

765

and PPI activity was quantified using ORIGINS249 per geneset edgelist. Dominance

766

Analysis50,51 was performed using dominance-analysis python package with default

767

multiple linear regression parameters. Average expression scores or PPI activity scores

768

per secRecon geneset were used as input features and normalized secreted IgG counts

769

or pseudotime were used as response variables in Dominance Analysis. secRecon

770

subsystems and genes correlating with secreted IgG or pseudotime were identified using

771

Spearman’s rho correlation (|r| > 0.1) and Benjamini-Hochberg multiple test correction

772

(BH p-value < 0.05).

773

To perform topology-based analysis of distinct Leiden clusters, the normalized

774

expression data matrix was subset for differentially expressed secRecon genes between

775

the IgG secreting population clusters. Pseudobulk groups per cluster were then generated

776

by splitting each cluster into random groups of cells and averaging expression of each

777

gene per pseudobulk group. Similar to the previous topological analyses for the bulk

778

transcriptome and proteome CHO and plasma cells dataset, pairwise correlation of gene-

779

expression and Baker’s gamma index of gene expression dendrograms were performed.

780

781

Transcription factor enrichment for secRecon genes

782

Transcription factors regulating secRecon genes were enriched using Ingenuity Pathway

783

Analysis71 (IPA), ChEA372, and the Lund et al. secretory reconstruction17. Human gene

784

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

symbols for all 1127 secRecon genes were input to IPA and ChEA3 using default

785

parameters. Enriched transcription factors from IPA Upstream Regulator analysis were

786

identified by filtering for “transcriptional regulator” molecule type and adjusted p-value <

787

0.05. Enriched transcription factors from ChEA3 analysis were identified by selecting the

788

top 100 ranked factors. As Lund et al. included DNA and protein interaction neighbors in

789

their secretory reconstruction, we intersected the GO:000370053 DNA-binding

790

transcription factor activity gene set list to filter for transcription factors associated with

791

secretory machinery. The three lists are provided as tables with annotations from the

792

respective databases (Supplementary Tables S4-S6).

793

794

Quantifying potential biases in secRecon scoring

795

RNA-seq TPM expression data for 36 different tissues originating from the GTEx

796

Consortium64,65 was obtained from the Human Protein Atlas (HPA) database. Pairwise

797

Spearman correlations were performed for individual secRecon gene expression

798

averaged for each tissue type against various metrics from the secRecon curation

799

(Supplementary Table S1), including the number of annotated secRecon ontology terms

800

(counting all associated parent terms), max annotated relevance score, mean relevance

801

score.

802

Acknowledgements

803

This work was supported by generous funding from NIGMS (R35 GM119850),

804

NSF (CBET-2030039), Novo Nordisk Foundation (NNF20SA0066621), and Sartorius

805

Stedim.

806

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Author Contributions

807

H.O.M., J.T., P.D.G., and N.E.L. led the network reconstruction, analyzed data,

808

and wrote the manuscript. J.T. and P.D.G. created the figures and visualizations for the

809

manuscript. I.S., H.B., M.S., C.M.R., C.C.K., N.K., S.S., A.G., Z.L., and A.M. contributed

810

to the network ontology and curation. A.A. and A.R. participated in the network

811

annotation. A.R. and N.E.L. reviewed and edited the manuscript. N.E.L. supervised and

812

secured funding for the project. All authors read and approved the manuscript.

813

Declaration of generative AI and AI-assisted technologies in the writing

814

process

815

During the preparation of this work the authors used ChatGPT-4 in a limited

816

manner to improve the readability and language of the manuscript. After using this tool,

817

the authors reviewed and edited the content. The authors take full responsibility for the

818

content of the published article.

819

Declaration of Interests

820

H.M. is an employee of Eli Lilly and Company. J.T. is an employee of Amgen, Inc.

821

A.R and A.A are employees of Sartorius. N.E.L is a co-founder of Augment Biologics, Inc.

822

and NeuImmune, Inc. and a board member for CHO Plus, Inc. The remaining authors

823

declare no competing interests.

824

825

Data and Code Availability

826

All data and code supporting the findings of this study are publicly available. The

827

secRecon knowledgebase, annotations and scripts necessary to reproduce the analyses

828

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

and figures presented in this manuscript are available in the GitHub repository at

829

https://github.com/LewisLabUCSD/secRecon-Secretory-Pathway-Reconstruction.

830

Additionally, the supplementary data accompanying the code are hosted on Synapse.org

831

and can be accessed via the DOI: https://doi.org/10.7303/syn64026567.

832

833

834

References

835

1. Gutierrez, J. M. et al. Genome-scale reconstructions of the mammalian secretory pathway

836

predict metabolic costs and limitations of protein secretion. Nat. Commun. 11, 68 (2020).

837

2. Uhlén, M. et al. Proteomics. Tissue-based map of the human proteome. Science 347,

838

1260419 (2015).

839

3. Kondylis, V., Pizette, S. & Rabouille, C. The early secretory pathway in development: a tale

840

of proteins and mRNAs. Semin. Cell Dev. Biol. 20, 817–827 (2009).

841

4. Vázquez-Martínez, R. et al. Revisiting the regulated secretory pathway: from frogs to

842

human. Gen. Comp. Endocrinol. 175, 1–9 (2012).

843

5. Stefan, C. J. et al. Membrane dynamics and organelle biogenesis-lipid pipelines and

844

vesicular carriers. BMC Biol. 15, 102 (2017).

845

6. Barlowe, C. K. & Miller, E. A. Secretory protein biogenesis and traffic in the early secretory

846

pathway. Genetics 193, 383–410 (2013).

847

7. Hou, J., Tyo, K., Liu, Z., Petranovic, D. & Nielsen, J. Engineering of vesicle trafficking

848

improves heterologous protein secretion in Saccharomyces cerevisiae. Metab. Eng. 14,

849

120–127 (2012).

850

8. Schwanhäusser, B. et al. Global quantification of mammalian gene expression control.

851

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Nature 473, 337–342 (2011).

852

9. Nielsen, J. Production of biopharmaceutical proteins by yeast: advances through metabolic

853

engineering. Bioengineered 4, 207–211 (2013).

854

10. Keller, M., Rüegg, A., Werner, S. & Beer, H.-D. Active caspase-1 is a regulator of

855

unconventional protein secretion. Cell 132, 818–831 (2008).

856

11. Hansen, U. et al. A secreted variant of cartilage oligomeric matrix protein carrying a

857

chondrodysplasia-causing mutation (p.H587R) disrupts collagen fibrillogenesis. Arthritis

858

Rheum. 63, 159–167 (2011).

859

12. Huber, R. J. Altered protein secretion in Batten disease. Dis. Model. Mech. 14, (2021).

860

13. Lu, P. J. et al. Mutations alter secretion of fukutin-related protein. Biochim. Biophys. Acta

861

1802, 253–258 (2010).

862

14. Braakman, I. & Bulleid, N. J. Protein folding and modification in the mammalian

863

endoplasmic reticulum. Annu. Rev. Biochem. 80, 71–99 (2011).

864

15. Guerriero, C. J. & Brodsky, J. L. The delicate balance between secreted protein folding and

865

endoplasmic reticulum-associated degradation in human physiology. Physiol. Rev. 92,

866

537–576 (2012).

867

16. Li, F. et al. Improving recombinant protein production by yeast through genome-scale

868

modeling using proteome constraints. Nat. Commun. 13, 2969 (2022).

869

17. Lund, A. M. et al. Network reconstruction of the mouse secretory pathway applied on CHO

870

cell transcriptome data. BMC Syst. Biol. 11, 37 (2017).

871

18. Feizi, A., Gatto, F., Uhlen, M. & Nielsen, J. Human protein secretory pathway genes are

872

expressed in a tissue-specific pattern to match processing demands of the secretome. NPJ

873

Syst Biol Appl 3, 22 (2017).

874

19. Feizi, A., Österlund, T., Petranovic, D., Bordel, S. & Nielsen, J. Genome-scale modeling of

875

the protein secretory machinery in yeast. PLoS One 8, e63284 (2013).

876

20. Narimatsu, H. Construction of a human glycogene library and comprehensive functional

877

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

analysis. Glycoconj. J. 21, 17–24 (2004).

878

21. Giurgiu, M. et al. CORUM: the comprehensive resource of mammalian protein complexes—

879

2019. Nucleic Acids Res 47, D559–D563 (2018).

880

22. The UniProt Consortium et al. UniProt: the Universal Protein Knowledgebase in 2023.

881

Nucleic Acids Res 51, D523–D531 (2022).

882

23. Szklarczyk, D. et al. The STRING database in 2023: protein-protein association networks

883

and functional enrichment analyses for any sequenced genome of interest. Nucleic Acids

884

Res 51, D638–D646 (2023).

885

24. Hein, M. Y. et al. Global organelle profiling reveals subcellular localization and remodeling

886

at proteome scale. bioRxiv 2023.12.18.572249 (2023) doi:10.1101/2023.12.18.572249.

887

25. Fruchterman, T. M. J. & Reingold, E. M. Graph drawing by force-directed placement. Softw.

888

Pract. Exp. 21, 1129–1164 (1991).

889

26. Morre, D. J. & Mollenhauer, H. H. The Golgi Apparatus: The First 100 Years. (Springer,

890

New York, NY, 2009). doi:10.1007/978-0-387-74347-9.

891

27. Finley, D. Recognition and processing of ubiquitin-protein conjugates by the proteasome.

892

Annu. Rev. Biochem. 78, 477–513 (2009).

893

28. Berggård, T., Linse, S. & James, P. Methods for the detection and analysis of protein-

894

protein interactions. Proteomics 7, 2833–2842 (2007).

895

29. Bensimon, A., Heck, A. J. R. & Aebersold, R. Mass spectrometry-based proteomics and

896

network biology. Annu. Rev. Biochem. 81, 379–405 (2012).

897

30. Walsh, G. Biopharmaceutical benchmarks 2018. Nat. Biotechnol. 36, 1136–1145 (2018).

898

31. Park, S.-Y. et al. Driving towards digital biomanufacturing by CHO genome-scale models.

899

Trends Biotechnol. 0, (2024).

900

32. Cenci, S. & Sitia, R. Managing and exploiting stress in the antibody factory. FEBS Lett. 581,

901

3652–3657 (2007).

902

33. Raab, N. et al. Nature as blueprint: Global phenotype engineering of CHO production cells

903

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

based on a multi-omics comparison with plasma cells. Metab. Eng. 83, 110–122 (2024).

904

34. Baker, F. B. Stability of two hierarchical grouping techniques case I: Sensitivity to data

905

errors. J. Am. Stat. Assoc. 69, 440–445 (1974).

906

35. Hänzelmann, S., Castelo, R. & Guinney, J. GSVA: gene set variation analysis for

907

microarray and RNA-Seq data. BMC Bioinformatics 14, 1–15 (2013).

908

36. Moremen, K. W., Tiemeyer, M. & Nairn, A. V. Vertebrate protein glycosylation: diversity,

909

synthesis and function. Nat. Rev. Mol. Cell Biol. 13, 448–462 (2012).

910

37. Breton, C., Šnajdrová, L., Jeanneau, C., Koča, J. & Imberty, A. Structures and mechanisms

911

of glycosyltransferases. Glycobiology 16, 29R–37R (2005).

912

38. Hetz, C., Zhang, K. & Kaufman, R. J. Mechanisms, regulation and functions of the unfolded

913

protein response. Nat. Rev. Mol. Cell Biol. 21, 421–438 (2020).

914

39. Rapoport, T. A., Li, L. & Park, E. Structural and Mechanistic Insights into Protein

915

Translocation. Annu. Rev. Cell Dev. Biol. 33, (2017).

916

40. Shapiro-Shelef, M. & Calame, K. Regulation of plasma-cell development. Nat. Rev.

917

Immunol. 5, 230–242 (2005).

918

41. Shi, W. et al. Transcriptional profiling of mouse B cell terminal differentiation defines a

919

signature for antibody-secreting plasma cells. Nat. Immunol. 16, 663–673 (2015).

920

42. Gaudette, B. T., Jones, D. D., Bortnick, A., Argon, Y. & Allman, D. mTORC1 coordinates an

921

immediate unfolded protein response-related transcriptome in activated B cells preceding

922

antibody secretion. Nat. Commun. 11, 723 (2020).

923

43. King, H. W. et al. Single-cell analysis of human B cell maturation predicts how antibody

924

class switching shapes selection dynamics. Sci Immunol 6, (2021).

925

44. Duan, M. et al. Understanding heterogeneity of human bone marrow plasma cell maturation

926

and survival pathways by single-cell analyses. Cell Rep. 42, 112682 (2023).

927

45. Trezise, S. et al. An arrayed CRISPR screen of primary B cells reveals the essential

928

elements of the antibody secretion pathway. Front. Immunol. 14, 1089243 (2023).

929

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

46. Xiong, E. et al. A CRISPR/Cas9-mediated screen identifies determinants of early plasma

930

cell differentiation. Front. Immunol. 13, 1083119 (2022).

931

47. Cheng, R. Y.-H. et al. SEC-seq: association of molecular signatures with antibody secretion

932

in thousands of single human plasma cells. Nat. Commun. 14, 3567 (2023).

933

48. Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587.e29

934

(2021).

935

49. Senra, D., Guisoni, N. & Diambra, L. Cell annotation using scRNA-seq data: A protein-

936

protein interaction network approach. MethodsX 10, 102179 (2023).

937

50. Luo, W. & Azen, R. Determining Predictor Importance in Hierarchical Linear Models Using

938

Dominance Analysis. J. Educ. Behav. Stat. 38, 3–31 (2013).

939

51. Budescu, D. V. Dominance analysis: A new approach to the problem of relative importance

940

of predictors in multiple regression. Psychol. Bull. 113, 542–551 (1993).

941

52. Ashburner, M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology

942

Consortium. Nat. Genet. 25, 25–29 (2000).

943

53. Gene Ontology Consortium et al. The Gene Ontology knowledgebase in 2023. Genetics

944

224, (2023).

945

54. Bertolotti, M. et al. B- to plasma-cell terminal differentiation entails oxidative stress and

946

profound reshaping of the antioxidant responses. Antioxid. Redox Signal. 13, 1133–1144

947

(2010).

948

55. Preisendörfer, S. et al. FK506-Binding Protein 11 Is a Novel Plasma Cell-Specific Antibody

949

Folding Catalyst with Increased Expression in Idiopathic Pulmonary Fibrosis. Cells 11,

950

(2022).

951

56. Kunert, R. & Reinhart, D. Advances in recombinant antibody manufacturing. Appl.

952

Microbiol. Biotechnol. 100, 3451–3461 (2016).

953

57. Reimold, A. M. et al. Plasma cell differentiation requires the transcription factor XBP-1.

954

Nature 412, 300–307 (2001).

955

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

58. Nutt, S. L., Hodgkin, P. D., Tarlinton, D. M. & Corcoran, L. M. The generation of antibody-

956

secreting plasma cells. Nature Reviews Immunology 15, 160–171 (2015).

957

59. van Anken, E. & Braakman, I. Endoplasmic reticulum stress and the making of a

958

professional secretory cell. Crit. Rev. Biochem. Mol. Biol. 40, 269–283 (2005).

959

60. Schinn, S.-M., Morrison, C., Wei, W., Zhang, L. & Lewis, N. E. A genome-scale metabolic

960

network model and machine learning predict amino acid concentrations in Chinese

961

Hamster Ovary cell cultures. Biotechnol. Bioeng. 118, 2118–2123 (2021).

962

61. Systematic evaluation of parameters for genome-scale metabolic models of cultured

963

mammalian cells. Metab. Eng. 66, 21–30 (2021).

964

62. Gopalakrishnan, S. et al. COSMIC-dFBA: A novel multi-scale hybrid framework for

965

bioprocess modeling. Metab. Eng. 82, 183–192 (2024).

966

63. Luck, K. et al. A reference map of the human binary protein interactome. Nature 580, 402–

967

408 (2020).

968

64. GTEx Consortium. The GTEx Consortium atlas of genetic regulatory effects across human

969

tissues. Science 369, 1318–1330 (2020).

970

65. GTEx Consortium et al. Genetic effects on gene expression across human tissues. Nature

971

550, 204–213 (2017).

972

66. Sayers, E. A General Introduction to the E-utilities. in Entrez Programming Utilities Help

973

[Internet] (National Center for Biotechnology Information (US), 2022).

974

67. Exploring Network Structure, Dynamics, and Function Using Networkx. (2008).

975

68. Traag, V. A., Waltman, L. & van Eck, N. J. From Louvain to Leiden: guaranteeing well-

976

connected communities. Sci Rep 9, 5233 (2019).

977

69. Lex, A., Gehlenborg, N., Strobelt, H., Vuillemot, R. & Pfister, H. UpSet: Visualization of

978

intersecting sets. IEEE Trans. Vis. Comput. Graph. 20, 1983–1992 (2014).

979

70. Huang, J. K. et al. Systematic Evaluation of Molecular Networks for Discovery of Disease

980

Genes. Cell Syst 6, 484–495.e5 (2018).

981

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

71. Krämer, A., Green, J., Pollard, J., Jr & Tugendreich, S. Causal analysis approaches in

982

Ingenuity Pathway Analysis. Bioinformatics 30, 523–530 (2014).

983

72. Keenan, A. B. et al. ChEA3: transcription factor enrichment analysis by orthogonal omics

984

integration. Nucleic Acids Res. 47, W212–W224 (2019).

985

986

987

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Supplementary Figures

988

989

Supplementary Figure 1. secRecon is a manually curated knowledgebase of the

990

mammalian secretory pathway. (A) Previous secretory pathway reconstructions

991

(Feizi18 N=575, Lund17 N=793 and Gutierrez1 N=271) were integrated and more

992

extensively annotated to build secRecon. In addition, genes involved in glycosylation

993

were added from GGDB20. (B) The draft was subsequently refined and expanded via

994

manual curation. Each individual gene was assigned to one or more specific processes

995

within our secretory pathway ontology along with its relevance score and functional

996

description backed by literature survey. (C) Additional annotations were integrated from

997

various databases, including mapping gene symbols to aliases, Ensembl IDs, Entrez IDs,

998

gene names, and UniProt IDs, and identifying orthologs, subcellular localizations, protein

999

complexes, and interaction partners (See Materials and Methods).

1000

1001

1002

1003

1004

1005

1006

1007

1008

1009

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Supplementary Figure 2. Elbow Plot Analysis of Community Detection in

1010

Functional and PPI Network Topologies: Elbow plots illustrating the results of Louvain

1011

community detection across a range of resolution parameters for (A) the functional

1012

topology network and (B) the protein-protein interaction (PPI) topology network. For each

1013

network, three metrics are plotted against the resolution parameter: modularity (blue line),

1014

normalized mutual information (NMI) with subcellular localization categories (red line),

1015

and NMI with secRecon system categories (green line). Modularity measures the strength

1016

of the community structure in the network, with higher values indicating more defined

1017

communities. The NMI scores quantify how well the detected community structures align

1018

with known biological attributes—either subcellular localization or functional system

1019

categories.

1020

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

1021

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Supplementary Figure 3. Network-based Proteomics Analysis of Plasma Cells vs

1022

CHO Cells: Log2 fold changes in protein levels between plasma and CHO cells were

1023

calculated and proteins from the proteomics dataset were overlaid with secRecon genes.

1024

Differentially expressed proteins were visualized in separate network plots for

1025

upregulated (A and C) and downregulated (B and D) proteins. Panels A and B display

1026

protein levels in a network layout based on the secRecon system ontology, where nodes

1027

are colored according to the major secretory pathway systems (left), or subcellular

1028

localization (right) they belong to. Panels C and D show the same sets of proteins plotted

1029

in a network representation based on protein-protein interactions, with node colors

1030

according to the major secretory pathway systems (left), or subcellular localization (right)

1031

they belong to. The size of each node corresponds to the magnitude of fold change in

1032

expression.

1033

1034

1035

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

1036

Supplementary Figure 4. Differentially enriched secRecon processes between

1037

plasma and CHO cells: Log fold-change of secretory machinery annotated under select

1038

processes significantly upregulated in both the transcript or protein enrichments were

1039

overlaid on PPI networks for (C) transcripts or (D) protein. The size of each node

1040

corresponds to the magnitude of fold change in expression in plasma vs CHO cells (bigger

1041

nodes indicate higher expression in plasma cells), while the color of each node indicates

1042

the process to which it belongs. Nodes associated with other secRecon processes are

1043

colored in gray.

1044

1045

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

1046

Supplementary Figure 5. Identifying differentially expressed secRecon genes in

1047

plasma cells for future engineering and characterization in CHO cells: Differentially

1048

expressed genes and differentially abundant proteins33 annotated within secRecon

1049

(N=112) were plotted according to respective log2-fold change in plasma cells relative to

1050

CHO cells. Genes are colored by differentially enriched secRecon processes identified in

1051

Figure 6 or otherwise colored gray. Boxed text labels indicate genes within these enriched

1052

processes that are similarly differentially expressed at both transcript and protein levels

1053

(e.g. positive fold-change above 1.5 or negative fold-change below -1.5 in both datasets)

1054

in plasma cells.

1055

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

1056

Supplementary Figure 6. Identifying secretory pathway signatures underlying

1057

single-cell IgG secretion in plasma cells: (A) SEC-seq links single-cell IgG secretion

1058

to transcriptome for a diverse population of B cells47 (B) Dominance Analysis identifies

1059

secRecon genesets and biological processes explaining the variation in single-cell IgG

1060

secretion were identified for the IgG secreting population. (top) secRecon genesets

1061

explain 10.9% variation in secreted IgG concentration while (bottom) additional cellular

1062

processes such as IgG gene transcription, cytoplasmic translation (GO-BP), and oxidative

1063

phosphorylation (GO-BP) increase the amount of variation explained for secreted IgG.

1064

1065

secRecon geneset activity in predicted IgG concentration. (D) secRecon subsystem

1066

activity and (E) gene expression were correlated with secreted IgG, with the highest

1067

correlates labeled.

1068

1069

1070

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

1071

Supplementary Figure 7. Identifying potential biases in secRecon curation and

1072

process scoring with physiological gene expression: secRecon gene expression was

1073

averaged for each tissue type in the Genotype-Tissue Expression (GTEx) Project and

1074

correlated against secRecon features: max secRecon process confidence score, mean

1075

secRecon process confidence score, and number of annotated secRecon processes.

1076

Spearman correlation coefficients of all pairwise tests are compiled in a heatmap.

1077

The copyright holder for this preprintthis version posted November 15, 2024. ; https://doi.org/10.1101/2024.11.14.623668doi: bioRxiv preprint

Fantastic genes and where to find them expressed in CHO

Article

Apr 2025

The transcriptome of Chinese hamster ovary (CHO) cells plays a crucial role in determining cellular characteristics that are essential for biopharmaceutical applications. RNA-sequencing has been extensively used to profile gene expression patterns, aiming to gain a better understanding of intracellular behavior and mechanisms. Individual datasets, however, do not provide a comprehensive overview and characterization of the CHO cell's transcriptome, such that the fundamental structure of the transcriptome remains unknown. Using 15 RNA-sequencing datasets, encompassing almost 300 samples of various experimental setups, conditions and cell lines, we explore and classify the protein-coding transcriptome of CHO cells. Differences in cell line lineages are found to be the primary source of variation in transcribed genes. By employing a novel approach, we identified the core transcriptome that is ubiquitously expressed in all cell lines and culture conditions, as well as genes that remain entirely non-expressed. Additionally, we identified a set of genes that may be active or inactive depending on different conditions, which are linked to biological processes including translation as well as immune and stress response. Lastly, by integrating chromatin states derived from histone modifications, we provided additional context on the epigenetic level that supports our protein-coding gene classification. Our study offers a comprehensive insight into the CHO cell transcriptome and lays the foundation for future research into cellular adaptation to changing conditions and the development of phenotypes.

The Gene Ontology knowledgebase in 2023

Article

Full-text available

May 2023
GENETICS

The Gene Ontology (GO) knowledgebase (http://geneontology.org) is a comprehensive resource concerning the functions of genes and gene products (proteins and noncoding RNAs). GO annotations cover genes from organisms across the tree of life as well as viruses, though most gene function knowledge currently derives from experiments carried out in a relatively small number of model organisms. Here, we provide an updated overview of the GO knowledgebase, as well as the efforts of the broad, international consortium of scientists that develops, maintains, and updates the GO knowledgebase. The GO knowledgebase consists of three components: (1) the GO—a computational knowledge structure describing the functional characteristics of genes; (2) GO annotations—evidence-supported statements asserting that a specific gene product has a particular functional characteristic; and (3) GO Causal Activity Models (GO-CAMs)—mechanistic models of molecular “pathways” (GO biological processes) created by linking multiple GO annotations using defined relations. Each of these components is continually expanded, revised, and updated in response to newly published discoveries and receives extensive QA checks, reviews, and user feedback. For each of these components, we provide a description of the current contents, recent developments to keep the knowledgebase up to date with new discoveries, and guidance on how users can best make use of the data that we provide. We conclude with future directions for the project.

SEC-seq: association of molecular signatures with antibody secretion in thousands of single human plasma cells

Article

Full-text available

Jun 2023

The secreted products of cells drive many functions in vivo; however, methods to link this functional information to surface markers and transcriptomes have been lacking. By accumulating secretions close to secreting cells held within cavity-containing hydrogel nanovials, we demonstrate workflows to analyze the amount of IgG secreted from single human B cells and link this information to surface markers and transcriptomes from the same cells. Measurements using flow cytometry and imaging flow cytometry corroborate the association between IgG secretion and CD38/CD138. By using oligonucleotide-labeled antibodies we find that upregulation of pathways for protein localization to the endoplasmic reticulum and mitochondrial oxidative phosphorylation are most associated with high IgG secretion, and uncover surrogate plasma cell surface markers (e.g., CD59) defined by the ability to secrete IgG. Altogether, this method links quantity of secretion with single-cell sequencing (SEC-seq) and enables researchers to fully explore the links between genome and function, laying the foundation for discoveries in immunology, stem cell biology, and beyond.

Cell annotation using scRNA-seq data: A protein-protein interaction network approach

Article

Full-text available

Apr 2023

Pathway analysis is an important step in the interpretation of single cell transcriptomic data, as it provides powerful information to detect which cellular processes are active in each individual cell. We have recently developed a protein-protein interaction network-based framework to quantify pluripotency associated pathways from scRNA-seq data. On this occasion, we extend this approach to quantify the activity of a pathway associated with any biological process, or even any list of genes. A systems-level characterization of pathway activities across multiple cell types provides a broadly applicable tool for the analysis of pathways in both healthy and disease conditions. Dysregulated cellular functions are a hallmark of a wide spectrum of human disorders, including cancer and autoimmune diseases. Here, we illustrate our method by analyzing various biological processes in healthy and cancer breast samples. Using this approach we found that tumor breast cells, even when they form a single group in the UMAP space, keep diverse biological programs active in a differentiated manner within the cluster.•We implement a protein-protein interaction network-based approach to quantify the activity of different biological processes. •The methodology can be used for cell annotation in scRNA-seq studies and is freely available as R package.

A CRISPR/Cas9-mediated screen identifies determinants of early plasma cell differentiation

Article

Full-text available

Jan 2023

Introduction The differentiation of B cells into antibody-secreting plasma cells depends on cell division-coupled, epigenetic and other cellular processes that are incompletely understood. Methods We have developed a CRISPR/Cas9-based screen that models an early stage of T cell-dependent plasma cell differentiation and measures B cell survival or proliferation versus the formation of CD138+ plasmablasts. Here, we refined and extended this screen to more than 500 candidate genes that are highly expressed in plasma cells. Results Among known genes whose deletion preferentially or mostly affected plasmablast formation were the transcription factors Prdm1 (BLIMP1), Irf4 and Pou2af1 (OBF-1), and the Ern1 gene encoding IRE1a, while deletion of XBP1, the transcriptional master regulator that specifies the expansion of the secretory program in plasma cells, had no effect. Defective plasmablast formation caused by Ern1 deletion could not be rescued by the active, spliced form of XBP1 whose processing is dependent on and downstream of IRE1a, suggesting that in early plasma cell differentiation IRE1a acts independently of XBP1. Moreover, we newly identified several genes involved in NF-kB signaling (Nfkbia), vesicle trafficking (Arf4, Preb) and epigenetic regulators that form part of the NuRD complex (Hdac1, Mta2, Mbd2) to be required for plasmablast formation. Deletion of ARF4, a small GTPase required for COPI vesicle formation, impaired plasmablast formation and blocked antibody secretion. After Hdac1 deletion plasmablast differentiation was consistently reduced by about 50%, while deletion of the closely related Hdac2 gene had no effect. Hdac1 knock-out led to strongly perturbed protein expression of antagonistic transcription factors that govern plasma cell versus B cell identity (by decreasing IRF4 and BLIMP1 and increasing BACH2 and PAX5). Discussion Taken together, our results highlight specific and non-redundant roles for Ern1, Arf4 and Hdac1 in the early steps of plasma cell differentiation.

Improving recombinant protein production by yeast through genome-scale modeling using proteome constraints

Article

Full-text available

May 2022

Eukaryotic cells are used as cell factories to produce and secrete multitudes of recombinant pharmaceutical proteins, including several of the current top-selling drugs. Due to the essential role and complexity of the secretory pathway, improvement for recombinant protein production through metabolic engineering has traditionally been relatively ad-hoc; and a more systematic approach is required to generate novel design principles. Here, we present the proteome-constrained genome-scale protein secretory model of yeast Saccharomyces cerevisiae (pcSecYeast), which enables us to simulate and explain phenotypes caused by limited secretory capacity. We further apply the pcSecYeast model to predict overexpression targets for the production of several recombinant proteins. We experimentally validate many of the predicted targets for α-amylase production to demonstrate pcSecYeast application as a computational tool in guiding yeast engineering and improving recombinant protein production. Due to the complexity of the protein secretory pathway, strategy suitable for the production of a certain recombination protein cannot be generalized. Here, the authors construct a proteome-constrained genome-scale protein secretory model for yeast and show its application in the production of different misfolded or recombinant proteins.

FK506-Binding Protein 11 Is a Novel Plasma Cell-Specific Antibody Folding Catalyst with Increased Expression in Idiopathic Pulmonary Fibrosis

Article

Full-text available

Apr 2022

Antibodies are central effectors of the adaptive immune response, widespread used therapeutics, but also potentially disease-causing biomolecules. Antibody folding catalysts in the plasma cell are incompletely defined. Idiopathic pulmonary fibrosis (IPF) is a fatal chronic lung disease with increasingly recognized autoimmune features. We found elevated expression of FK506-binding protein 11 (FKBP11) in IPF lungs where FKBP11 specifically localized to antibody-producing plasma cells. Suggesting a general role in plasma cells, plasma cell-specific FKBP11 expression was equally observed in lymphatic tissues, and in vitro B cell to plasma cell differentiation was accompanied by induction of FKBP11 expression. Recombinant human FKBP11 was able to refold IgG antibody in vitro and inhibited by FK506, strongly supporting a function as antibody peptidyl-prolyl cis-trans isomerase. Induction of ER stress in cell lines demonstrated induction of FKBP11 in the context of the unfolded protein response in an X-box-binding protein 1 (XBP1)-dependent manner. While deficiency of FKBP11 increased susceptibility to ER stress-mediated cell death in an alveolar epithelial cell line, FKBP11 knockdown in an antibody-producing hybridoma cell line neither induced cell death nor decreased expression or secretion of IgG antibody. Similarly, antibody secretion by the same hybridoma cell line was not affected by knockdown of the established antibody peptidyl-prolyl isomerase cyclophilin B. The results are consistent with FKBP11 as a novel XBP1-regulated antibody peptidyl-prolyl cis-trans isomerase and indicate significant redundancy in the ER-resident folding machinery of antibody-producing hybridoma cells.

Altered protein secretion in Batten disease

Article

Full-text available

Dec 2021

Robert J. Huber

The neuronal ceroid lipofuscinoses (NCLs), collectively known as Batten disease, are a group of neurological diseases that affect all ages and ethnicities worldwide. There are 13 different subtypes of NCL, each caused by a mutation in a distinct gene. The NCLs are characterized by the accumulation of undigestible lipids and proteins in various cell types. This leads to progressive neurodegeneration and clinical symptoms including vision loss, progressive motor and cognitive decline, seizures, and premature death. These diseases have commonly been characterized by lysosomal defects leading to the accumulation of undigestible material but further research on the NCLs suggests that altered protein secretion may also play an important role. This has been strengthened by recent work in biomedical model organisms, including Dictyostelium discoideum, mice, and sheep. Research in D. discoideum has reported the extracellular localization of some NCL-related proteins and the effects of NCL-related gene loss on protein secretion during unicellular growth and multicellular development. Aberrant protein secretion has also been observed in mammalian models of NCL, which has allowed examination of patient-derived cerebrospinal fluid and urine for potential diagnostic and prognostic biomarkers. Accumulated evidence links seven of the 13 known NCL-related genes to protein secretion, suggesting that altered secretion is a common hallmark of multiple NCL subtypes. This Review highlights the impact of altered protein secretion in the NCLs, identifies potential biomarkers of interest and suggests that future work in this area can provide new therapeutic insight.

Integrated analysis of multimodal single-cell data

Article

Full-text available

May 2021
CELL

The simultaneous measurement of multiple modalities represents an exciting frontier for single-cell genomics and necessitates computational methods that can define cellular states based on multimodal data. Here, we introduce “weighted-nearest neighbor” analysis, an unsupervised framework to learn the relative utility of each data type in each cell, enabling an integrative analysis of multiple modalities. We apply our procedure to a CITE-seq dataset of 211,000 human peripheral blood mononuclear cells (PBMCs) with panels extending to 228 antibodies to construct a multimodal reference atlas of the circulating immune system. Multimodal analysis substantially improves our ability to resolve cell states, allowing us to identify and validate previously unreported lymphoid subpopulations. Moreover, we demonstrate how to leverage this reference to rapidly map new datasets and to interpret immune responses to vaccination and coronavirus disease 2019 (COVID-19). Our approach represents a broadly applicable strategy to analyze single-cell multimodal datasets and to look beyond the transcriptome toward a unified and multimodal definition of cellular identity.

Genetic effects on gene expression across human tissues

Article

Full-text available

Sep 2016

A. Roger Little

Characterization of the molecular function of the human genome and its variation across individuals is essential for identifying the cellular mechanisms that underlie human genetic traits and diseases. The Genotype-Tissue Expression (GTEx) project aims to characterize variation in gene expression levels across individuals and diverse tissues of the human body, many of which are not easily accessible. Here we describe genetic effects on gene expression levels across 44 human tissues. We find that local genetic variation affects gene expression levels for the majority of genes, and we further identify inter-chromosomal genetic effects for 93 genes and 112 loci. On the basis of the identified genetic effects, we characterize patterns of tissue specificity, compare local and distal effects, and evaluate the functional properties of the genetic effects. We also demonstrate that multi-tissue, multi-individual data can be used to identify genes and pathways affected by human disease-associated variation, enabling a mechanistic interpretation of gene regulation and the genetic basis of disease.

COSMIC-dFBA: A novel multi-scale hybrid framework for bioprocess modeling

Article

Feb 2024
METAB ENG

A reconstruction of the mammalian secretory pathway identifies mechanisms regulating antibody production

Abstract and Figures

Recommended publications

A Reconstruction of the Mammalian Secretory Pathway Identifies Mechanisms Regulating Antibody Produc...

From omics to Cellular mechanisms in mammalian cell factory development

Dysregulation of the secretory pathway connects Alzheimer’s disease genetics to aggregate formation

In situ detection of protein interactions for recombinant therapeutic enzymes