Gene expression patterns in CD4+ peripheral blood cells in healthy subjects and stage IV melanoma patients

Abstract

Melanoma patients exhibit changes in immune responsiveness in the local tumor environment, draining lymph nodes, and peripheral blood. Immune-targeting therapies are revolutionizing melanoma patient care increasingly, and studies show that patients derive clinical benefit from these newer agents. Nonetheless, predicting which patients will benefit from these costly therapies remains a challenge. In an effort to capture individual differences in immune responsiveness, we are analyzing patterns of gene expression in human peripheral blood cells using RNAseq. Focusing on CD4+ peripheral blood cells, we describe multiple categories of immune regulating genes, which are expressed in highly ordered patterns shared by cohorts of healthy subjects and stage IV melanoma patients. Despite displaying conservation in overall transcriptome structure, CD4+ peripheral blood cells from melanoma patients differ quantitatively from healthy subjects in the expression of more than 2000 genes. Moreover, 1300 differentially expressed genes are found in transcript response patterns following activation of CD4+ cells ex vivo, suggesting that widespread functional discrepancies differentiate the immune systems of healthy subjects and melanoma patients. While our analysis reveals that the transcriptome architecture characteristic of healthy subjects is maintained in cancer patients, the genes expressed differentially among individuals and across cohorts provide opportunities for understanding variable immune states as well as response potentials, thus establishing a foundation for predicting individual responses to stimuli such as immunotherapeutic agents.

Electronic supplementary material

The online version of this article (doi:10.1007/s00262-015-1745-x) contains supplementary material, which is available to authorized users.

Introduction

Attempts to mobilize the immune response to destroy disseminated cancer cells have met with variable success and are challenged by local [1, 2] and systemic immune suppressive effects [3, 4] of advancing cancers, degradation of the immune system by standard cancer therapies, and variability introduced by genetic and epigenetic determinants of host responsiveness [5–7]. In order to delineate how these variables impact the status of the immune system and immune response potential, we have begun assessing the underlying gene expression profiles of cells comprising the immune system. As these cells circulate throughout the body and survey the cellular and molecular environments of most tissues and draining lymph nodes, we began our studies by focusing on cells from peripheral blood. This material is accessible from most patient populations, and we assume that the reactive “state” of circulating cells is an indication of how those cells are poised to respond to stimuli encountered within an inflamed tissue or tumor.

Cutting-edge technologies reveal the inner workings of immune cells and define molecular determinants of the human immune response [8]. Ability to measure these molecular determinants is fundamental to evaluating individual inflammatory and immune responses to environmental cues, yet traditional approaches focusing on antibody and cytokine production or the expression of select cell-associated molecules are not sufficient to provide informative answers. More comprehensive approaches such as characterizing the transcriptome of peripheral blood mononuclear cells have been stymied by the complexity of the cell types present in these mixed populations [9, 10]. Efforts to investigate factors differentiating closely related cell types are beginning to define pathways governing cellular differentiation [11–13] and define parameters of individual variability [14–16]. Technologies have advanced to the point where the transcriptome of individual cells can be analyzed systematically [17].

We have adopted an intermediate strategy by analyzing subsets of cells that share major lineage markers used to define cellular elements of the immune system. This approach has the advantage of substantially reducing complexity of the cell populations studied, while retaining some elements of complexity that might be important for determining immune response status or defining individual-specific characteristics of immunity; for example, finding that one individual had a different composition of CD4+ cellular subsets may be a determining trait of that individual. By quantitating the transcriptomes of 10⁶ cells, the response patterns observed are averages, which remove the cell-to-cell temporal variation in gene expression observed when analyzing single cells. Whereas recent studies investigating the transcriptomes of immune cells have used semiquantitative gene array profiling [14–16], we used RNAseq [18, 19] to glean a more complete and complex look at the mRNA content of cells at the individual gene level. Recent comparisons of technical platforms for assessing global gene expression indicate that RNAseq is superior to hybridization-based approaches for measuring mRNA levels of genes with transcript levels below the 50th percentile [19]. We reason that many genes of importance may be expressed at these middling-to-lower levels, even in lineage-enriched cell populations, and thus chose to sequence transcriptomes of individuals at the depth of 37 million reads to capture accurately RNA present at levels as low as 1 read per million (RPM).

In this study, we delineate structural features of the human CD4+ transcriptome. CD4+ peripheral blood lymphocytes (PBL) of T cell lineages (e.g., Th1, Th2, Th17, Treg [20, 21]) and other less frequent cell types such as CD4+ macrophages [22] should be captured in this assay. We chose to focus initially on CD4+ cells (primarily T cells) because these cells not only provide helper and regulatory functions for CD8 T cells, but they also interact with and modulate the functions of many other immune cells [23]. Our thesis is that individual variation in CD4+ subpopulations is a trait detectable by the relative abundance of specific RNAs recovered from the lineage positive PBL, a view consistent with recent studies suggesting gene expression, as opposed to levels of regulation is the major factor determining phenotype [24]. At any given time, the variation in relative gene expression by lineage positive populations distinguishing individuals is expected to be the product of differences in genetic hard wiring and epigenetic changes stemming from biological rhythms and environmental stimuli. Here, we demonstrate RNAseq as a quantitative tool for human studies and document the remarkable homogeneity in mRNA expression patterns among both healthy individuals as well as patients with stage IV melanoma for genes specifying many shared functions of CD4+ T cells. Our data show that other populations of mRNA are more variably expressed and suggest that these more complex expression patterns can be used as a source for elucidating mechanisms of systemic immune dysfunction in melanoma and how immune therapies can be appropriately harnessed to improve patient outcomes.

Materials and methods

Study subjects

Participant recruitment and blood sample collection were conducted using protocols approval by the Mayo Clinic Institutional Review Board (IRB#12-002580 and IRB#13-002293). Normal healthy subjects (seven females and three males; median age 34 years) were from a southeastern Minnesota community cohort of adults self-identifying as having no autoimmunity, allergy, immunodeficiency, immunosuppressive treatments, or cancer. The second cohort (five females and five males; median age 60) was comprised of patients with a diagnosis of stage IV melanoma, but having different degrees of disease burden and a variety of treatment histories, planning to receive immune therapy (see Supplementary Table 1).

CD4+ T cell isolation and stimulation

Blood was drawn into sodium heparin tubes (APP Pharmaceuticals, NDC 63323-540-11); CD4+ cells were isolated from whole blood using anti-CD4 microbeads (Miltenyi Biotec) and MS column purification as per the manufacturer’s instructions. A small portion of the cell isolate was set aside for flow cytometry using anti CD3 APC-H7 and anti CD4 PE-Cy7 (BD PharMingen) to assess purity. Isolated cells were re-suspended at 1 million/ml in RPMI containing 10 % FBS and either immediately lysed in 0.7 ml QIAzol lysis reagent (Qiagen) or incubated for 4 h at 37 °C with 25 μl/ml anti-CD3/anti-CD28 Human T-Activator Dynabeads (Invitrogen) prior to lysis. Lysates were stored at −80 °C until processing for RNAseq.

RNA and cDNA library preparation

Samples with RNA integrity values of ≥7.0 by Agilent Bioanalyzer (Santa Clara, CA) were processed by the Mayo Clinic Medical Genome Facility Gene Expression Core into TruSeq libraries according to the manufacturer’s instructions (RNA Prep Kit v2, Illumina, San Diego, CA) and published elsewhere [25]. Paired-end DNA adaptors (Illumina) with a single “T” base overhang at the 3′ end were immediately ligated to the “A-tailed” cDNA population. Unique indexes, included in the standard TruSeq Kits (12-Set A and 12-Set B), were incorporated at the adaptor ligation step for multiplex sample loading on the flow cells. Libraries (8–10 pM) were loaded onto paired-end flow cells to generate cluster densities of 700,000/mm² following Illumina’s standard protocol for cBot and cBot paired-end cluster kit version 3. The flow cells were sequenced as 51 X 2 paired-end reads on an Illumina HiSeq 2000 using TruSeq SBS sequencing kit version 3 and HCS v2.0.12 data collection software. Base-calling was performed using RTA version 1.17.21.3.

Bioinformatics analysis

The Illumina standard pipeline version 1.4 was used to process raw images to make base calls and to generate sequence reads. The generated FASTQ sequence reads (51 bp long) were aligned to human genome assembly hg19 using TopHat aligner [26] with the default parameters. The raw read counts of each annotated gene in GENCODE v12 [27] were calculated by the HTSeq program (http://www-huber.embl.de/users/anders/HTSeq/doc/count.html) for further downstream analysis. The initial reference genome selected is applied consistently throughout the analysis, even though updated references periodically become available. However, inferences about differences in gene expression for weakly annotated or previously unidentified loci from this presentation should be tempered until a deeper analysis of the distribution of the reads mapped to the genomic reference to verify that the correct match between the putative locus and mapped transcripts has been identified. For well-defined coding sequences, this is not an issue. For most analyses, the raw read counts of each gene were normalized by the total reads of each individual sample and then standardized to reads per million (RPM, gene counts/total counts of each sample × 1 million). Genes without any reads across all samples were eliminated from downstream analysis. As there were genes with zero count in some of samples, 1 was added to all gene counts in all samples after converting to RPM. Log2-transformed RPM values were used for additional analyses. Non-transformed data were used to calculate fold differences from arithmetic means.

Statistical analyses

The average (mean) and variation in mRNA expression across 10 individual subjects were determined and characterized using the coefficient of variation (CV = STDEV/mean) using Microsoft Excel software. Additional data analyses were conducted using Prism (GraphPad). The distributions of the CV measurements were assessed using the F test. The variances of the distributions of CV were found in some comparisons to be highly significantly different. Consequently, all subsequent comparisons were conducted using nonparametric rank analyses (Mann–Whitney or Kruskal–Wallis) as indicated where appropriate. A probability threshold of less than 0.05 was used to determine statistical significance. A gene-by-gene comparison of means and CV for the described gene sets was conducted by regression analysis in Prism. Two-way ANOVA to derive differentially expressed gene sets, stable between duplicate blood draws, were conducted using Partek [28].

Results

Peripheral blood CD4+ cells were isolated directly from whole blood of healthy donors (HD, n = 10; Supplementary Table 1) and patients with stage IV melanoma (Mel, n = 10) using magnetic beads (Supplementary Figure 1). Typical enrichment efficiency (Supplementary Figure  2) for CD4+ cells was approximately 1000-fold. In order to identify similarities and differences in gene expression as a reflection of both immune state and response potential, RNA was extracted from freshly isolated (i.e., unstimulated, T0) cells, as well as cells placed into short-term culture (4 h in the presence of anti-CD3/anti-CD28 beads, T4) and analyzed by RNAseq. The number of individual mRNAs is reported as log2 of the mapped reads per million reads plus 1 to account for zero reads in some samples (log2 [RPM+1]). Approximately 12,000 different mRNAs were present at greater than a value of 2 which is ~1 copy per cell, assuming a typical resting lymphocyte contains approximately 3.5 × 10⁵ mRNA molecules [17, 29, 30]. Although sequences in this range could be present in only a fraction of the cells, they may represent subset-specific transcripts as CD4 T cells are known to differentiate into several functionally distinct lineages (e.g., Th1, Th2, Th17, Treg). We have also seen that previously unappreciated cell types, present at low frequencies (<1 %), can have profound functional significance [31]. Consequently, the ability to measure mRNA species at low frequency is important for our approach.

We first sought to determine the frequency of mRNAs encoding gene products key to CD4+ T cell functions: (1) 58 genes encoding elements of the T cell receptor (CD3 + CD28) signaling pathway, (2) 41 genes encoding transcription factors regulating immune functions and polarization of CD4 T cells, (3) 17 genes encoding cytokines up-regulated more than tenfold following CD3/CD28 ligation, (4) genes expressed at the highest levels (>400 RPM, non-log2 transformed) in freshly isolated cells (~140 genes), and (5) genes expressed at the highest levels and up-regulated more than tenfold following stimulation (~60 genes).

The T cell receptor and the CD28 co-stimulatory receptor signaling pathways comprise central mechanisms of intracellular communication used by CD4 T cells [20, 21]. Transcripts encoding these elements, common housekeeping genes, and approximately 5000 other annotated loci were present in CD4+ peripheral blood resting cells at levels ranging from 5.5 to 9.5 (Supplementary Table 2 and data not shown). A comparison of the average expression of each member of this gene set from cohorts of healthy donors (HD) and melanoma patients (Mel) revealed shared, highly ordered patterns of gene expression both before and after stimulation (Fig. 1a, d). Parallel analyses focusing on transcription factors (Supplementary Table 3) and the most highly expressed transcripts (present at greater than 400 reads per million (log 2–8.6; ~150 transcripts, see Supplementary Table 4) revealed similar patterns (Fig. 1b, c, e). Among these most highly expressed mRNA are those encoding cell surface markers known to be expressed by CD4+ cells, including CD4, CD3ε, and IL-10Rα; 12 of the 58 molecules associated with the TcR/CD28 signaling pathways were also in this category. Other transcripts in this list encode functions associated with the mechanics of translation, the cytoskeleton, or energy metabolism, and some have no known function.

Fig. 1.

Regression analysis of mean expression values (log2[RPM + 1]) for a genes encoding elements of the T cell receptor (TcR) signaling pathway, b genes encoding transcription factors typical of immune cells, and c highly expressed genes in freshly isolated CD4+ PBL of healthy donors (HD) and patients with stage IV melanoma (Mel). Data are available in Supplementary Tables 2–4. d, e Similar analysis of TcR pathway and transcription factor transcripts from stimulated CD4+ PBL. Data were calculated using Excel (Microsoft) and analyzed using Prism (GraphPad)

To assess immune responsiveness, CD4+ PBL were stimulated for four hours ex vivo in medium containing beads coated with antibodies specific for CD3 epsilon and the co-stimulatory receptor CD28. The levels of cytokine-encoding mRNAs in stimulated cells were increased greater than tenfold over resting levels (Fig. 2a, Supplementary Table 5), yet the stimulated levels of these cytokines was highly similar in cells from melanoma patients and healthy donors (Fig. 2b). Genes expressed at high levels after ex vivo stimulation and whose expression levels were increased at least tenfold were also highly comparable between these cohorts (Fig. 2c; Supplementary Table 6).

Fig. 2.

a Response to ex vivo stimulation assessed by quantitation of cytokine gene expression using RNA from unstimulated (T0) and stimulated (T4) CD4+ PBL from healthy donors (HD). Data are available in Supplementary Table 5. b Regression analysis of mean expression levels for cytokine genes and c other genes highly expressed and up-regulated tenfold in stimulated CD4+ PBL from HD and Mel. Data are available in Supplementary Tables 5 and 6. Data were plotted and analyzed using Kruskal–Wallis and Dunn’s multiple comparison tests using Prism

We next considered the intracohort variability in gene expression, calculating coefficients of variation (CV) for the gene sets described above. In unstimulated CD4+ PBL, variations in the genes of the TcR pathway from healthy donors and melanoma patients were relatively small, with median CVs of 0.019 and 0.015, respectively (Fig. 3a, b). The median CV of expression for transcription factors in freshly isolated cells from healthy subjects (T0, HD) was significantly higher than the median CV for the signaling molecules (Fig. 3a, P = 0.0202, Mann–Whitney U). Much of this difference can be traced to variation in a subset of the transcription factors with CV values over 0.07, as the top quartile of the transcription factors has a median CV of 0.161 in comparison with the equivalent quartile of the TcR/CD28 signaling pathway with a median CV of 0.039 (Supplementary Tables 3 and 2, respectively). Included in this group are genes TBX21 and RORC encoding master regulators of the Th1 and Th17 response patterns, EOMES, a factor associated with terminally differentiated cells [32, 33], as well as heterodimeric regulators such as CREB5 and FOSB [34]. This disparity in CVs for genes associated with TcR signaling pathway (shared by all CD4+ T cells) and transcription factors in the CD4+ PBL also was observed in cells from the cohort of 10 melanoma patients (Fig. 3b, P = 0.0048). These data suggest that despite disease course and treatment regimens endured by these individual cancer patients, the underlying architecture of CD4+ PBL gene expression pattern is preserved.

Fig. 3.

Coefficients of variation (CV) are shown for RNAseq quantitation of gene expression for known elements of the T cell receptor (TcR) signaling pathways and downstream transcription factors (TFs). RNA was from unstimulated (T0) CD4+ cells from a healthy donors (HD; n = 10) and b patients with stage IV melanoma (Mel; n = 10; see Supplementary Tables 2 and 3). Data were plotted and analyzed using Mann–Whitney test in Prism (GraphPad). Each line indicates the median CV for each group. c, d Similar analyses for genes encoding TcR pathway and cytokines at T4 as well as highly expressed genes at T4 induced tenfold by simulation ex vivo. Data are available in Supplementary Tables 2, 5, and 6

When data from stimulated cells were analyzed, cytokine gene expression showed a substantial variation among the healthy individuals in comparison with TcR pathway genes (Fig. 3c, median CV = 0.101 vs. 0.024, P < 0.0001, Mann–Whitney U). A parallel analysis of the transcripts in melanoma patient CD4+ PBL revealed a similar disparity between the CVs of TcR pathway elements and cytokine genes in the stimulated cells (Fig. 3d, P < 0.0001). These findings indicate that some key modulating factor(s) distinguishing individual responses impacts the variability in cytokine gene expression downstream of signaling molecules which are expressed less variably. Relatively low variability was observed for a set of 58 genes expressed at greater than 400 copies per million and up-regulated greater than tenfold following activation (Fig. 3c, Supplementary Table 6; comparing median CVs of 0.101 vs. 0.0250, P < 0.05, Kruskal–Wallis and Dunn’s multiple comparison tests). A very similar pattern was observed in our analysis of CD4+ PBL from melanoma patients (Fig. 3d, P < 0.05). These findings suggest that some genes that are strongly up-regulated post-activation are more variably up-regulated across individuals within cohorts than are others and that this variation of gene expression is not necessarily higher among highly expressed genes.

Next, we examined the transcriptomes from our cohort of healthy subjects for additional RNA transcripts expressed by CD4+ cells with higher variability, a subset of genes which might provide a basis for discriminating individual differences in immune function. Using a high threshold of CV > 0.1, 1818 transcripts were expressed at greater than 2 copies per cell (RPM > 3.3) in fresh CD4+ PBL. Of those genes, ~40 have known roles in immune responses, yet were present in similar quantities and with similar CVs in HD and melanoma patient CD4+ cells (Supplementary Table 7). Similarly, ~400 transcripts present in HD CD4+ PBL following ex vivo stimulation were identified, 37 with known roles in immune responses (Supplementary Table 8). Again, overall levels and variability of expression in melanoma patients paralleled those in healthy donors.

To further investigate whether the patterns of conservation observed in healthy subjects and melanoma patients were interrelated and ordered, we performed a regression analysis comparing the CV for each gene in the sets encoding the CD3/CD28 signaling molecules, the transcription factors subset, and the induced cytokines before and after stimulation. By ordering the measures from healthy subjects (X-axis) and cancer patients (Y-axis) according to average CV (Supplementary Figure 3), a strongly correlated pattern for each comparison was revealed (P < 0.0001). The strongest correlations were for the TCR/CD28 signaling pathways and for transcription factor genes expressed after stimulation. We conclude that the transcriptomes of both healthy subjects and melanoma patients are highly ordered with a strong pattern specifying relative rankings in gene expression levels and patterns of variation.

Having established shared underlying architecture of the transcriptomes of the healthy and melanoma patient cohorts, we evaluated the data further for differences in gene expression patterns in a head-to-head comparison. To control for the variability in expression of some genes irrespective of subject cohort, we analyzed RNAseq data for which we had duplicate, independent blood samples (10 HD and 7 Mel) using a two-way ANOVA. In fresh CD4+ PBL transcriptomes, approximately 6500 loci were found present at consistent levels for these 17 individuals (using a threshold of P ≥ 0.5). In stimulated CD4+ PBL transcriptomes, that number was closer to 6000 loci. There were 2153 genes that differed between the cohorts at P < 0.01 (Table 1); 1413 genes (66 %) were expressed at higher levels (P < 0.01) in cells from melanoma patients. For cells stimulated 4 h, this relationship reversed with 66 % of the 1328 stable genes differentiating healthy and patient cohorts expressed at higher levels in cells from the healthy cohort (Table 1). Regression analyses of the mean expression values confirmed the previous conclusions that the transcriptomes of both fresh and stimulated CD4+ PBL from healthy subjects and metastatic melanoma patients were highly correlative (Fig. 4a, b). Nonetheless, more than 100 genes within the sets differentiating healthy subjects and melanoma patients at P < 0.01 exceeded a threshold of twofold difference (Fig. 4c, d). These patterns of differential gene expression were also observed using data from two additional healthy subjects and two additional melanoma patients where only single blood draw data were available (not shown).

Table 1.

Numbers of genes distinguishing the transcriptomes of CD4+ PBL from healthy donors and stage IV melanoma patients

CD4+ PBL	Total stably expressed genes distinguishing melanoma from healthy subjects over probability threshold (P < 0.01)	Genes expressed at higher levels in melanoma patients (P < 0.01)	Subset of genes expressed > twofold higher in melanoma patients (P < 0.01)	Subset of genes expressed > twofold higher in healthy subjects (P < 0.01)
Fresh (T0)	2153	1413 (66 %)	93 (4.3 %)	28 (1.3 %)
Stimulated (T4)	1328	449 (34 %)	49 (3.7 %)	64 (4.8 %)

Fig. 4.

Regression analysis of stably and differentially expressed genes in Mel compared with HD. a Comparison (R ² = 0.943) of mean RPM values for ~2150 genes from fresh CD4+ PBL with stable expression values between duplicate blood draws (P ≥ 0.5) and distinguishing HD and metastatic melanoma patients at P < 0.01. b Similar to a but with ~1300 genes from stimulated PBL (R ² = 0.938). c 121 gene subset from a that differed between cohorts by at least twofold (R ² = 0.664). d 113 gene subset from b that differed between cohorts by at least twofold (R ² = 0.601)

The top 50 genes whose expression differed by at least twofold between HD and Mel were ranked by P value and used to construct a heat map for both T0 and T4 data (Fig. 5a, b). Of note, the genes comprising this list derived from T0 data were over-expressed in the Mel CD4+ PBL compared with HD cells, whereas the genes derived from T4 data were under-expressed in the Mel cells compared with HD samples. As the ages of the subjects in our cohorts were not the same, additional analyses of an age-matched sample subset were performed, showing that the contribution of age to the gene expression patterns was not substantial (Fig. 5c; Supplementary Tables 9 and 10, and data not shown).

Fig. 5.

a Heat map displaying 50 stably expressed genes whose expression levels best differentiate (based on P value) melanoma patients (n = 7) from healthy donors (n = 10). Expression levels of the genes are shown for each individual melanoma patient relative to healthy donor mean expression levels of those genes. Blue gene expressed at lower RPM in Mel. Red gene expressed at higher RPM in Mel. b Similar to a but with transcriptomes of stimulated (T4) CD4+ Mel (n = 6) relative to HD PBL (n = 10). c Similar to a and b but showing results of analysis using data only from age-matched HD and Mel

Finally, the cumulative data analyzed in this report were evaluated to determine whether the cohort of melanoma patients displayed greater variability in gene expression relative to the healthy subjects. The distributions of the evaluated CVs for the expression of the 236 genes in freshly isolated CD4+ PBL (Supplementary Tables 2, 3, and 4) and the 75 genes in stimulated PBL (Supplementary Tables 5 and 6) were found significantly more variable in the melanoma patients (P = 0.015 and P = 0.002, respectively, by Mann–Whitney U analysis), perhaps reflecting differences in disease burden and treatment history.

Discussion

Stage IV melanoma patients have been a targeted population for immunotherapy clinical trials. To investigate the status of the immune system of these advanced cancer patients, the structure and patterns of variability within the transcriptome of CD4+ PBL were compared with the transcriptomes of CD4+ PBL from healthy subjects. We used RNAseq to develop global assessments of immune cells and found this approach measures gene expression with precision, as indicated by the low coefficients of variations (<0.02) for 50 % of the genes evaluated. This initial assessment of CD4+ PBL from very advanced cancer patients reveals that, despite having had different treatment histories and tumor burdens, the foundation of the CD4+ cell transcriptome from these patients is well conserved in terms of both transcript abundance and variability between members of the cohort. Moreover, the transcripts dictating CD4 T cell functions, including those encoding the T cell receptor and associated signaling molecules, are expressed very similarly in healthy individuals and patients with stage IV melanoma. Nonetheless, there are approximately 2000 genes whose expression levels are statistically different levels in the melanoma patients and healthy subject cohorts, even after controlling for variability between repeated blood draws. Comparison of age-matched subsets from these cohorts revealed similar findings. The variability in the expression levels of specific genes in cells from melanoma patients is greater in response to ex vivo stimulation compared with the variability of the same genes in cells from healthy subjects. We conclude that systemic changes in the immune status of melanoma patients are clearly evident and that these changes are reflected in altered immune responsiveness to strong stimuli.

Transcriptome analysis allows for simultaneous evaluation of cell type-specific molecules for their roles in dictating cellular functions and also as a way to seek out mechanisms contributing to individual variations in cellular (and here, immune) responses. Our data show that the most abundantly expressed mRNA molecules (more than 400 copies per 10⁶ transcripts) identified from freshly isolated human CD4+ cells include CD4 itself, elements of the T cell receptor (TcR constant region and CD3 components), other cell surface leukocyte markers, kinases, adaptor molecules, and transcription factors and that these basic immune response elements are largely intact in advanced melanoma patients. Moreover, the individual subject variability in the frequency of differentiated CD4 T cell subtypes (e.g., Th1, Th2, Th17, or Treg cells), based on mRNA levels and their variability for transcription factors that differentiate these cell types (e.g., T-bet (TBX21), GATA3, RORγt (RORC), FOXP3; [35]) were similar between PBL of healthy subjects and melanoma patients. However, transcripts encoding the AP1 components Fos, Jun, and JunB were repressed in freshly isolated CD4 PBL from melanoma patients relative to our healthy donors. Functional annotation clustering using DAVID Bioinformatics Resources [36, 37] revealed that many genes differentially expressed by cells from melanoma patients were enriched for inflammatory response genes and encoded membrane proteins involved in cell–cell communication, cell adhesion and migration, and molecule transport. How specific changes in gene expression relate to altered states of immune responsiveness will require additional study.

Our data also underscore how little is understood about the connections between genes expressed in specific types of cells; for example, stimulating CD4 T cells ex vivo for 4 h resulted in ~2000 transcripts increased or decreased by at least a factor of 4. How the signals are elaborated from the engaged cell surface receptors to such a large network of genes in just a few hours is not understood. Furthermore, our data show that most of the central players identified in CD4 cells are maintained in late-stage cancer patients and that those cells can respond to stimuli. How this information can be used to deduce critical points of dysregulation central to restoring homeostasis and focusing immunity against the cancer remains a challenge. Our RNAseq analysis reveals that many of the genes differentiating CD4 cells from healthy subjects and melanoma patients are expressed at relatively low levels which may be best addressed using sensitive quantitative assays such as RNAseq. We hypothesize that a gene expression signature from immune cells that has clinical consequence will be one that takes into account a summation of the contributions of several hundred genes, each expressed at measurably different levels by subjects in healthy and cancer-burdened cohorts. In addition, the contribution of the transcriptomes from other PBL subtypes will likely be important.

During the course of this work, a study by Benoist and colleagues identified sources of individual variation of gene expression linking RNA expression to GWAS analyses and assessed their relationships to cellular activation and autoimmunity [16]. Our approach is different, focused on a summation of variation across many genes to create patterns that may identify sources of individuality in immune responsiveness. It will be of interest to determine to what extent the differentially expressed or the variably expressed gene subsets identified in our study correlate with other measures of differences in immune responses among individuals in various cohorts. Future studies will aim to determine which of these gene expression patterns can be resolved into predictive measures of how cancer patients respond to therapeutic interventions.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

CV

Coefficient of variation

HD

Healthy donors

HGNC

HUGO Gene Nomenclature Committee

Mel

Melanoma patients

PBL

Peripheral blood leukocyte

RPM

Reads per million

TcR

T cell receptor

Compliance with ethical standards

Written informed consent was obtained from all individual participants included in this study. All procedures involving human participants were in accordance with the ethical standards and approval of the Institutional Review Board (protocols #12-002580 and #13-002293) and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Conflict of interest

No author has a financial and/or commercial conflict of interest.