Analytical workflow profiling gene expression in murine macrophages

Abstract

Comprehensive and simultaneous analysis of all genes in a biological sample is a capability of RNA-Seq technology. Analysis of the entire transcriptome benefits from summarization of genes at the functional level. As a cellular response of interest not previously explored with RNA-Seq, peritoneal macrophages from mice under two conditions (control and immunologically challenged) were analyzed for gene expression differences. Quantification of individual transcripts modeled RNA-Seq read distribution and uncertainty (using a Beta Negative Binomial distribution), then tested for differential transcript expression (False Discovery Rate-adjusted p-value < 0.05). Enrichment of functional categories utilized the list of differentially expressed genes. A total of 2079 differentially expressed transcripts representing 1884 genes were detected. Enrichment of 92 categories from Gene Ontology Biological Processes and Molecular Functions, and KEGG pathways were grouped into 6 clusters. Clusters included defense and inflammatory response (Enrichment Score = 11.24) and ribosomal activity (Enrichment Score = 17.89). Our work provides a context to the fine detail of individual gene expression differences in murine peritoneal macrophages during immunological challenge with high throughput RNA-Seq.

1. Introduction

Identification and analysis of an individual gene may offer limited insights. While genes serve as one of the smallest units by which biological change can be measured, critical information comes from considering the sum of their individual effects. Expanding the “snapshot view” available for differential expression motivates a drive towards the enlargement of analyses from single gene studies with quantitative real-time PCR to microarrays, and more recently RNA-Seq.¹

The range of tools available for RNA-Seq analysis, as well as the tools themselves, undergoes a rapid pace of modification. These changes demand a thorough understanding of how the tools operate to choose appropriate settings for a particular experiment. Without a singular accepted method or settings to address all applications, transcriptomics relies upon the validation of data quality and controls.² TopHat, Cufflinks, and Cuffdiff comprise a set of tools for analyzing RNA-Seq datasets.³ These tools have gained popularity for the capability to handle intron-spanning reads, and options to address various biological- and technical-biases that are of concern during analysis.⁴

Mapping RNA reads to an annotated genome is one of the popular and well-established methods for differential expression testing between treatments in model organisms.⁵ With the potential to detect thousands of differentially expressed genes, organizing these differences into more interpretable groups becomes the purpose of downstream tools. One possibility that is explored here involves grouping the gene information into groups based upon their functional actions, a form of gene set enrichment.⁶ This study examines the capability of a RNA-Seq-based workflow to evaluate transcriptomic changes. Efficient identification of differentially expressed genes and the functions they impact elucidates their modification of the biological state between treatments. The novelty of this experiment is in the application of RNA-Seq and the associated algorithms to a particular biological model, the analysis of peritoneal macrophages from Bacille Calmette–Guérin (BCG)-challenged mice compared to those receiving a saline control.⁷ This challenge has been associated with substantial changes in sickness and depression-like indicators.⁸ The characterization of the transcriptome during immunological resolution and behavioral transition seven days after initial challenge is of interest. RNA-Seq has yet to be applied to characterize the transcriptome at this time point.^9–11 The application of RNA-Seq and downstream methods to analyze changes in transcriptomics in this model has not been reported, providing a new level of capability in constructing an inflammation-induced immunological response profile. Transcript profiles were further studied and interpreted using functional enrichment analyses to uncover categories that may be over-represented among particular profiles.

2. Materials and Methods

RNA-Seq technology was used to study changes in gene expression in macrophages taken from mice following a previously established immune-challenge model.¹⁰ Male adult (~ 22 weeks of age; n = 6/group) C57BL/6J mice were injected into the peritoneum with TICE strain BCG (Organon USA Inc., USA) or equal volume (10 mg) physiological saline (Control). Utilizing the same inbred strain as used for the Mouse Genome Project minimizes genetic variations that could hinder mapping.¹² RNA was isolated from macrophages collected from the peritoneal cavity seven days post-challenge.^13,14 This timing of collection was selected to capture transcriptome changes during a period of immunological and behavioral transitions.^8,11

The workflow of RNA-Seq data analysis is presented in Fig. 1. Transcriptomic analysis with RNA-Seq involves producing libraries of reads that represent gene transcripts from the samples for quantitative comparison. Individual mouse RNA-Seq libraries were sequenced using Illumina HiSeq2000 (Illumina, San Diego, CA) to produce paired-end 100-bp reads, summarized as “left” and “right” reads. One library of reads per biological sample was examined for sequencing errors prior to mapping to genome and transcriptome features. Quality control of sequence reads used FastQC (Fig. 2).¹⁵ Quality was determined by the reported score at each base position (> 30), a Qphred quality value which is the negative logarithmic transformation of the estimated probability of error (Eq. (1)).¹⁶

Fig. 1.

RNA-Seq workflow, showing the analysis of each sample individually by Tophat and Cufflinks (inset) before the collective analysis of all samples in Cuffdiff to test for differential expression (?) between conditions (BCG and Control or Ctrl groups).

Fig. 2.

Quality box-and-whisker graphs via FastQC illustrating quality scores across the read length in the left and right reads from BCG and Control (Ctrl) samples.

$$Q_{\text{PHRED}}=-10\times {log}_{10}(P_e).$$ (1)

Reads were mapped to the mouse genome (GRCm38) and assembled using TopHat2 (TopHat v2.0.9) and Cufflinks and analyzed using Cuffmerge, and Cuffdiff 2 (v2.1.1, Fig. 1).³ TopHat2 maps reads via the use of Bowtie2, the core read-alignment program, while TopHat2 deals with splicing concerns from mapping intron-spanning RNA reads to a DNA genome.³ Due to the computational scale of mapping millions of reads to large genomes, Bowtie2 implements Burrows–Wheeler transformation to efficiently scan the genome during mapping.¹⁷ TopHat2 was chosen for its two-step method to deal with spliced alignments and preferential alignment of reads onto real genes from an annotation.¹⁸

Reads were assembled based upon mapping information into gene transcripts, with transcripts quantified by condition for differential comparison as elaborated in Ref. 3. The Cufflinks program (http://cufflinks.cbcb.umd.edu/) takes the mapping information from TopHat2, and assembles the reads back into the biologically relevant transcripts that would have produced them. Cufflinks offers optional assembly methods that correct for biological and technical biases, including biases in Illumina’s read-creation process.¹⁹ Options to correct for fragment bias during transcription priming with random hexamers and estimation of appropriate counting for those reads that can map to multiple sites were used.^20,21 Upper Quartile normalization was enabled for its superior performance compared to the default Total Count method available in Cufflinks.²²

Cuffdiff 2 (referred to from here on simply as “Cuffdiff”) performs differential expression testing between conditions by checking if each gene follows a beta negative binomial distribution. The beta negative binomial distribution can account for potential overdispersion between groups or uncertainty in read counts that may otherwise be ignored by simpler models.⁴ Before any testing for significance, all loci in the genome first needed a minimum number of fragment alignments (10 fragments; Test Status “OK”). Genes within a locus could be analyzed for significance after this minimum alignment (MA) within Cuffdiff was satisfied. Of those genes in locations with > 10 fragment alignments, a list of genes exhibiting significant differential expression between conditions (False Discovery Rate or FDR-adjusted p-value < 0.05) was obtained. The genes were named based upon annotation available from the UCSC database (see Ref. 23, www.genome.ucsc.edu).

Two complementary approaches were used to identify functional categories among transcript profiles. Enrichment based on the hypergeometric test applied to a list of differentially abundant transcript isoforms and gene set enrichment analysis (GSEA) of all transcript isoforms based on the Kolmogorov–Smirnov statistics were evaluated.²⁴ Gene Ontological (GO, see Ref. 25, www.geneontology.org) terms related to Biological Process (BP) and Molecular Function (MF) were tested, along with the Kyoto Encyclopedia of Genes and Genomes (KEGG)-Pathway database (see Ref. 26, http://www.genome.jp/kegg). For the hypergeometric test, functional category enrichment and functional annotation clustering were performed in the Database for Annotation, Visualization, and Integrated Discovery (DAVID).⁶ Specifically the GO FAT categories within DAVID were tested, a filter of GO categories to minimize repetition of general categories and to focus on more specific term identification. Individual categories in DAVID are deemed enriched by using a one-tailed jackknifed Fisher exact test, the EASE score.²⁷ The downstream functional annotation clustering of these categories used Enrichment Score (ES), calculated as the −log scale geometric mean of the EASE scores of member categories.²⁸

For the purposes of clustering, DAVID by default considers categories individually by their EASE score (EASE ≤ 0.1) without concern for experiment-wide false-detection.²⁷ To avoid errors related to multiple tests, categories were only considered enriched if they were significant at FDR-adjusted EASE score based p-value < 0.1.²⁹ Cluster ES were re-calculated to reflect the remaining member categories (ES > 4). The GSEA methodology was implemented using the software package GSEA-P and enrichment was tested against the functional categories present in the Molecular Signature Database (MSigDB).²⁴ The recommended GSEA FDR-adjusted p-value < 0.25 threshold was used in agreement with the statistical testing implemented.²⁴ Categories consistent between the hypergeometric and GSEA approaches are reported and discussed. These results are robust to differences in assumptions and methodologies between the approaches.

3. Results

The quality control was evaluated for every sample. No evidence of low quality reads was observed within the samples, with quality scores greater than 30 across the entire length of the reads. Quality scores were similarly high across both Control and BCG groups (Fig. 2). Scores ranged between 30 and 40, indicating accuracies between 99.9% and 99.99% for the bases at those positions. Based upon the observed quality of the sample data as well as the read filtering internal to TopHat2, trimming was not needed.

The RNA-Seq reads produced 54 ± 8.5 million reads and 64 ± 6 million paired-end reads of 100 bp in length per sample for Control and BCG, respectively. On average 91% of total reads was mapped to the genome for both Control and BCG. The percentage of reads per sample that successfully mapped to the genome ranged from 74% to 95%, the percentage of reads that produced aligning pairs was also in the same range (Table 1).

Table 1.

Read count of the paired-end (left/right) sample data.

Group	Sample	Readsa	Direction	Mapped (%)	Aligned pairs (%)
Ctrl	1	48,137,202	Left	78	77
		48,137,202	Right	81
	2	54,465,049	Left	94	92
		54,465,049	Right	93
	3	50,384,929	Left	95	94
		50,384,929	Right	95
	4	58,747,069	Left	95	94
		58,747,069	Right	95
	5	68,739,003	Left	95	93
		68,739,003	Right	94
	6	45,314,712	Left	92	88
		45,314,712	Right	89
BCG	1	64,092,232	Left	75	74
		64,092,232	Right	77
	2	72,364,837	Left	94	92
		72,364,837	Right	94
	3	67,524,327	Left	95	94
		67,524,327	Right	95
	4	61,335,634	Left	95	94
		61,335,634	Right	95
	5	54,744,239	Left	94	93
		54,744,239	Right	94
	6	62,302,008	Left	92	89
		62,302,008	Right	90

^aNumber of reads for the sample, for each of the paired runs (direction).

Following evaluation of read quality, their assembly into transcripts produced over 60,000 transcripts among all samples. Prior to differential testing between the groups, these transcripts were filtered based upon sufficient alignment coverage and experiment-wide significance cut-offs. The number of differentially expressed transcripts between Control and BCG groups was 2079 (Table 2; 1373: FDR p-value < 0.01; 706: 0.01 < FDR p-value < 0.05), representing 1884 genes.

Table 2.

Transcript and gene counts within Cuffdiff.

	Total tested	MA > 10a	Significant (FDR p-value < 0.05)	Named genesb
Transcript	62,490	29,844	2258	2079
Gene	23,274	12,009	1885	1884

^aMA: Minimum alignment; a locus (i.e. transcript) needs at least this many fragments aligned before significance testing will be performed.

^bNamed genes were determined using the UCSC database (http://genome.ucsc.edu).

Among the differentially expressed genes, 802 were under-expressed in BCG vs. Control, indicating similar quantity of up- and down-regulated genes post-challenge. However, there was a predominance of genes overexpressed in BCG relative to Control among the most significant profiles. The most significantly differentially expressed genes (FDR p-value < 0.01) are listed separately for those over- (Table 3) and underexpressed (Table 4) in BCG vs. Control, together with supporting references when previously associated with macrophage populations and their immunological response profile.

Table 3.

Twenty five genes showing the greatest differential overexpression in BCG relative to Control (FDR p-value < 0.01).

Gene namea	Gene ID	Log2 (BCG/Control)	Referenceb
S100a9	20202	10.08	30
Ly6i	57248	9.50	31
Asprv1	67855	9.44	32, 33
Il1f9	215257	8.19	33, 34
Spon1	233744	8.02	35
Nos2	18126	8.01	36, 37
S100a8	20201	7.69	30, 36
Ccl8	20307	7.58	30, 38
Cxcr2	12765	7.50	30
Fcgr1	14129	6.03	30
Xcl1	16963	5.49	33, 39
AW112010	107350	4.98	40
Lst1	16988	4.80	41
Oas3	246727	4.77	67
Rsad2	58185	4.72	35
Smpdl3b	100340	4.52	42
Gbp2	14469	4.38	68
AA467197	433470	4.26	43
Ccl5	20304	3.99	33, 38
Cxcl10	15945	3.81	30
Chi3l3	12655	3.81	30
Acsl1	14081	3.59	35
Isg15	100038882	3.38	35
Ifi27l2a	76933	2.65	44
Arg1	11846	3.35	30, 33

^aAA467197: expressed sequence AA467197; Acsl1: Acyl-CoA synthetase long-chain family member 1; Arg1: Arginase 1; Asprv1: Aspartic peptidase, retroviral-like 1; AW112010: Expressed sequence AW112010; Ccl5: Chemo-kine (C-C motif) ligand 5; Ccl8: Chemokine (C-C motif) ligand 8; Chi3l3: Chitinase-like 3; Cxcr2: Chemokine (C-X-C motif) receptor 2; Cxcl10: Che-mokine (C-X-C motif) ligand 10; Fcgr1: Fc receptor, IgG, high affinity I; Gbp2: Guanylate binding protein 2; Ifi27l2a: Interferon, alpha-inducible protein 27 like 2A; Il1f9: Interleukin 1 family, member 9; Isg15: ISG15 ubi-quitin-like modifier; Lst1: Leukocyte specific transcript 1; Ly6i: Lymphocyte antigen 6 complex, locus I; Nos2: Nitric oxide synthase 2, inducible; Oas3: 2′-5′ oligoadenylate synthetase 3; Rsad2: Radical S-adenosyl methionine domain containing 2; Smpdl3b: Sphingomyelin phosphodiesterase, acid-like 3B; Spon1: Spondin 1; S100a8: S100 calcium binding protein 8; S100a9: S100 calcium binding protein 9; Xcl1: Chemokine (C motif) ligand 1.

^bLiterature associating the listed gene with a macrophage population.

Table 4.

Twenty five genes showing the greatest differential underexpression in the BCG relative to the control group (FDR p-value < 0.01).

Gene namea	Gene ID	Log2 (BCG/Control)	Referenceb
Retnla	57262	−4.90	30, 45
Cxcl13	55985	−4.67	30, 33, 35
Cd209a	170786	−4.42	33
Pf4	56744	−4.07	30, 35
Fcrls	80891	−3.90	30
Adm	11535	−3.51	30, 35
Lyve1	114332	−3.41	46
Vsig4	278180	−3.19	35, 47
Bank1	242248	−3.07	48
Il10	16153	−3.01	30, 33, 35, 49
Cd83	12522	−2.87	35
Faim3	69169	−2.79	50
Blk	12143	−2.79	35
Pou2af1	18985	−2.78	35
Mmd	67468	−2.71	35, 51
Cd79b	15985	−2.66	30
Bcar3	29815	−2.62	42
Cd2	12481	−2.55	35
Fabp4	11770	−2.43	30, 35
Gimap6	231931	−2.36	35
F13a1	74145	−2.35	30
Ptprcap	19265	−2.34	—
Phgdh	236539	−1.94	43
Mt1	17748	−1.70	52
Wfdc17	100034251	−1.22	53

^aAdm: Adrenomedullin; Bank1: B cell scaffold protein with ankyrin repeats 1; Bcar3: Breast cancer anti-estrogen resistance 3; Blk: B lymphoid kinase; Cd2: Cd2 antigen; Cd209a: Cd209a antigen; Cd79b: Cd79b antigen; Cd83: CD83 antigen; Cxcl13: Chemokine (C-X-C motif) ligand 13; Fabp4: Fatty acid binding protein 4, adipocyte; Faim3: Fas apoptotic inhibitory molecule 3; Fcrls: Fc receptor-like S, scavenger receptor; F13a1: Coagulation factor XIII, A1 subunit; Gimap6: GTPase, IMAP family member 6; Il10: Interleukin 10; Lyve1: Lymphatic vessel endothelial hyaluronan receptor 1; Mmd: Monocyte to macrophage differentiation-associated; Mt1: Metallothionein 1; Pf4: Platelet factor 4; Phgdh: 3-phosphoglycerate dehydrogenase; Pou2af1: POU domain, class 2, associating factor 1; Ptprcap: Protein tyrosine phosphatase, receptor type, C polypeptide-associated protein; Retnla: Resistin like alpha; Wfdc17: WAP four-disulfide core domain 17; Vsig4: V-set and immunoglobulin domain containing 4.

^bLiterature associating the listed gene with a macrophage population.

Functional categorization of the gene list resulted in 92 significantly enriched terms (BP: 69 terms; MF: 20 terms; KEGG: 3 terms; FDR p-value < 0.1; not listed). Clustering the enriched terms further reduced the list to 6 highly enriched clusters (ES > 4), listed by score in Table 5. Clusters were dominated by GO BP terms as they were the majority of the significantly enriched term list, with terms in the clusters underscoring the activation and regulation of the immune system following challenge. These clusters accounted for 24 of the significantly enriched terms.

Table 5.

List of member terms for each functional clusters (ES > 4).

Cluster identifier	ES (Genes)a	Identifier (Genes)b	Term name
1	17.89 (98)	GO:0006412 (69)	Translation
		GO:0003735 (59)	Structural constituent of ribosome
		GO:0005198 (85)	Structural molecule activity
		Mmu03010 (60)	Ribosome
2	11.24 (118)	GO:0009611 (80)	Response to wounding
		GO:0006952 (92)	Defense response
		GO:0006954 (54)	Inflammatory response
3	7.15 (66)	GO:0042330 (35)	Taxis
		GO:0006935 (35)	Chemotaxis
		GO:0007626 (46)	Locomotory behavior
		GO:0008009 (17)	Chemokine activity
		GO:0042379 (17)	Chemokine receptor binding
		GO:0005125 (38)	Cytokine activity
4	5.89 (67)	GO:0030246 (67)	Carbohydrate binding
		GO:0030247 (30)	Polysaccharide binding
		GO:0001871 (30)	Pattern binding
		GO:0005539 (27)	Glycosaminoglycan binding
5	5.83 (83)	GO:0006915 (79)	Apoptosis
		GO:0012501 (79)	Programmed cell death
		GO:0008219 (83)	Cell death
		GO:0016265 (83)	Death
6	5.45 (88)	GO:0042981 (88)	Regulation of apoptosis
		GO:0043067 (88)	Regulation of programmed cell death
		GO:0010941 (88)	Regulation of cell death

^aES: Enrichment Score; Listed in parenthesis, the ES is the number of genes enriching the cluster.

^bMember terms of the cluster, with the number of genes enriching that term in parentheses.

4. Discussion

Quality control of the input reads is an important step to successful downstream mapping. Once the reads were determined to be of high quality, the filtering controls implemented by TopHat2 prior to mapping made additional trimming of the reads unnecessary (Table 1).⁵⁴ Percentages of mapped reads were similar to those reported in previous high-stringency methods, and approached the percentages seen when previously tested on simulated error-free data.^18,55 The mapping capability of aligners like TopHat2 is dependent upon the genome and annotation, meaning unmapped reads may include those associated with transcripts not yet represented in the annotation. Findings from these RNA-Seq confirmed several results from previous studies that used similar models and quantitative real-time PCR or microarray technologies and uncovered additional profiles and enriched categories. This study centered on one type of peripheral macrophage, collected at one time point and using a specific collection method on macrophage activation status. A longitudinal study of additional macrophage populations using alternative collection methods is necessary for extrapolation of our findings to wider conditions.

The workflow described here effectively identifies genes that are differentially expressed during an immunological challenge and clusters these results based upon functionality. Significantly differentially expressed genes illustrated the extended expression response after BCG-challenge. Among the overexpressed genes, the most overexpressed gene S100a9 works as a heterodimer with S100a8, also found to be in the overexpressed list (Table 3). As both are associated with inflammatory events and are inducible in mature macrophages, their presence after BCG-challenge is expected.⁵⁶ The overexpression of CCL5 and CXCL10 (Table 3) was also unsurprising, considering the inhibitory action of IL-10 upon both, and that reduced IL-10 levels were associated with increased resistance to intracellular pathogens⁴⁹ such as BCG (Il10, Table 4). The overexpression of Arg1 in the BCG group is consistent with previous work studying the effect of this challenge in macrophages.⁵⁷ Along with the most underexpressed gene in BCG compared to Control, Retnla, these indicate underexpression of Th2-associated genes due to the classic Th1-response to BCG.^45,58 It is interesting to find Mt1 and Il10 together in the underexpressed category. Although studies were previously performed in T cells, Mt1/Mt2-deficient mice were found to produce increased levels of Il10 following an immune challenge with anti-CD3/CD28.⁵⁹ Still, the role of metallothionein genes during immune challenges and inflammation are not fully elucidated, and low expression of Mt1 supports the proinflammatory nature of the response at the measured time-point.⁶⁰ Direct association in the literature between macrophages and the underexpressed gene Ptprcap was less clear, although it has been found in the monocyte precursors to macrophages.⁶¹ However, Ptprcap is known as a CD45-associate, regulating the interaction of CD45 with other proteins. As CD45 regulates apoptosis, this may explain the relationship to immune-challenge.⁶² Although the number of differentially expressed genes were similarly split between overexpressed and underexpressed in the BCG relative to the control group, a more stringent significance cutoff found a predominance of genes overexpressed in the BCG group. These results are consistent with other reports of overexpression in the microglia of genes associated with inflammation response in response to an inflammatory challenge.^63,64

Table 5 summarizes the enriched functional categories consistently detected by the hypergeometric test and GSEA approaches. Enrichment analysis highlighted the biological response of macrophages to an immunological challenge (inflammation-based defense responses; clusters 2 and 3 in Table 5). Clustering was effective at identifying cytokine and chemokine activity in immune cells that are typically associated with activation of macrophages.⁶⁵ Categories previously associated with similar immune challenges⁶⁶ were clustered to better clarify the transcriptomic differences between experimental groups. Several genes that were overexpressed in BCG relative to Control (Table 3) are affiliated to regulation of locomotion (Table 5) including Xcl1, Cxcl13, Cxcr2, Cxcl10, and Ccl5. These associations could be related to the typical amelioration of sickness behaviors and higher activity observed in mice seven days post-challenge.^8,10 A ribosomal cluster (cluster 1) dominating the list is expected, as protein regulation is at the core of immunological response.⁶⁵ The gene lists and resulting clusters from RNA-Seq technology allows for analysis based upon the shared and unique genes. In future studies, this response profile of immunologically challenged peritoneal macrophages can be compared to similar constructed profiles of other cell populations or challenges to identify profile characteristics unique to each combination.

Biographies

Scott E. Nixon is currently a Ph.D. Candidate in Bioinformatics in the laboratory of Dr. Rodriguez-Zas at the University of Illinois at Urbana-Champaign.

Dianelys González-Peña is currently a Ph.D. Candidate in Animal Genetics in the laboratory of Dr. Rodriguez-Zas at the University of Illinois at Urbana-Champaign.

Marcus A. Lawson is a Postdoctoral Researcher in the Division of Nutritional Sciences at the University of Illinois at Urbana-Champaign.

Robert H. McCusker is a Research Assistant Professor in the Department of Animal Sciences at the University of Illinois at Urbana-Champaign. His key areas of research include Neuroimmunology and the associated impacts on behavior.

Alvaro G. Hernandez is the Director of the High-Throughput Sequencing and Genotyping Unit, which is a part of the Roy J. Carver Biotechnology Center at the University of Illinois at Urbana-Champaign.

Jason C. O’Connor is an Assistant Professor in the Department of Pharmacology at the University of Texas Health Science Center in San Antonio. His key areas of research include Neuroinflammation and Inflammation-Associated Depression.

Robert Dantzer is a Professor in the Department of Symptom Research at the University of Texas MD Anderson Cancer Center in Houston. His key areas of research include Psychoneuroimmunology as well as Brain Expression and Action of Cytokines.

Keith W. Kelley is a Professor Emeritus in the Department of Animal Sciences at the University of Illinois at Urbana-Champaign. His key areas of research involve Immunophysiology and Behavior.

Sandra L. Rodriguez-Zas is a Professor in the Department of Animal Sciences, an Adjunct Professor in the Department of Statistics, an active member of the Neuroscience Program, and an affiliate with the Institute of Genomic Biology, all at the University of Illinois at Urbana-Champaign. Her key areas of research involve Statistical Genomics and Bioinformatics.