Simultaneous Profiling of 194 Distinct Receptor Transcripts in Human Cells

See allHide authors and affiliations

Science Signaling  06 Aug 2013:
Vol. 6, Issue 287, pp. rs13
DOI: 10.1126/scisignal.2003624


Many signal transduction cascades are initiated by transmembrane receptors with the presence or absence and abundance of receptors dictating cellular responsiveness. We provide a validated array of quantitative reverse transcription polymerase chain reaction (qRT-PCR) reagents for high-throughput profiling of the presence and relative abundance of transcripts for 194 transmembrane receptors in the human genome. We found that the qRT-PCR array had greater sensitivity and specificity for the detected receptor transcript profiles compared to conventional oligonucleotide microarrays or exon microarrays. The qRT-PCR array also distinguished functional receptor presence versus absence more accurately than deep sequencing of adenylated RNA species by RNA sequencing (RNA-seq). By applying qRT-PCR–based receptor transcript profiling to 40 human cell lines representing four main tissues (pancreas, skin, breast, and colon), we identified clusters of cell lines with enhanced signaling capabilities and revealed a role for receptor silencing in defining tissue lineage. Ectopic expression of the interleukin-10 (IL-10) receptor–encoding gene IL10RA in melanoma cells engaged an IL-10 autocrine loop not otherwise present in this cell type, which altered signaling, gene expression, and cellular responses to proinflammatory stimuli. Our array provides a rapid, inexpensive, and convenient means for assigning a receptor signature to any human cell or tissue type.


Transmembrane signaling receptors are the genetically encoded sensors of the extracellular environment (1). A cell can display millions of receptor copies on its cell surface (2), yet intracellular responses can be triggered when just a few dozen receptors bind their cognate ligands (3, 4). Complete absence of a signaling receptor renders a cell unresponsive to its ligands (5, 6), meaning that the cell is “blind” to that class of environmental inputs. Transmembrane proteins are enriched in the low-abundance fractions of the transcriptome and proteome (7). These low-abundance transcripts and the surface proteins that they encode are also effective indicators of cell lineage (7, 8). The qualitative presence or absence of signaling receptors thus defines a critical facet of a cell’s identity and its response capabilities.

Large profiles of receptor families can be extracted from transcriptome measurements obtained by oligonucleotide microarrays (9), but the extracted profiles are not definitive. Microarrays have a compressed dynamic range and poorer detection sensitivity relative to single-gene methods (10), and some probe sets on established platforms are still plagued with cross-hybridization artifacts (11). Compared to microarrays, digital transcript counting by RNA sequencing (RNA-seq) is more specific and shows substantially improved dynamic range (12) and sensitivity (13). However, RNA-seq is methodologically inefficient because the technique must repeatedly measure high-abundance transcripts to achieve maximal sensitivity toward the low-abundance targets (14). There is additional evidence that the rarest transcripts identified by RNA-seq are nonfunctional (15), which hinders the ability of RNA-seq to determine whether signaling-competent receptors are truly present or absent in a cell population.

Gene expression measurements from microarrays or RNA-seq are often validated with quantitative reverse transcription polymerase chain reaction (qRT-PCR) (10, 15). Because of its high sensitivity, wide dynamic range, and verifiable specificity, qRT-PCR is routinely viewed as a gold standard for expression studies with individual genes. Inspired by an effort aimed at characterizing the transcriptional profile of a subset of G protein (heterotrimeric guanine nucleotide–binding protein)–coupled receptors (16), here we developed and validated arrayed qRT-PCR reagents for 194 transmembrane signaling receptors in the human genome. By exploiting the array’s sensitivity, we found that the presence of receptor transcripts was far more widespread than typically reported by oligonucleotide microarrays. Surprisingly, the qRT-PCR array was also a more specific predictor of protein presence or absence than RNA-seq. For multiple receptors, we confirmed the accuracy of our profiling approach biochemically and functionally in cells. The throughput of the approach was then leveraged to define the receptor transcript signatures for 40 commonly used human cell lines, representing predominantly cancers of the pancreas, breast, colon, and skin (specifically, melanoma). This pilot study revealed collections of receptors with transcripts that were highly abundant in a lineage-specific manner, as well as several receptors that were selectively silenced. Lentiviral transduction of the IL10RA gene, encoding interleukin-10 (IL-10) receptor subunit α, into melanoma cells lacking this receptor subunit created a constitutive, artificial autocrine circuit involving endogenous IL-10. Autocrine IL-10 perturbed basal signaling, inducible gene expression, and the sensitivity of melanoma cells to apoptotic stimuli, thus showing that receptor absence was a critical mechanism for preventing this autocrine loop and controlling the cellular response. Our approach provides a general tool for surveying the signaling capabilities of human cell populations, and the method of simultaneous transcript profiling can be easily adapted to include additional receptor families.


qRT-PCR receptome profiling is accurate, precise, and more sensitive than oligonucleotide microarrays

We defined a signaling “receptome” (17) that includes all human receptor serine-threonine and tyrosine kinases, all cytokine and chemokine receptors, as well as all receptors of the Toll-like, Frizzled, Notch, and Patched families (Fig. 1A and file S1). These signaling receptors bind a diverse range of macromolecular ligands and show widespread, but selective, tissue expression. A defined panel enabled in-depth validation of gene-specific reagents that together were readily accommodated in a 96-well format for high-throughput profiling.

Fig. 1 Defining a human signaling receptome for profiling by arrayed qRT-PCR.

(A) Distribution of signaling receptor families and subfamilies comprising the receptome-profiling assay. TGFβR, transforming growth factor–β receptor; BMPR, bone morphogenetic protein receptor; ACVR, activin A receptor; EPH, ephrin receptor; EGFR, epidermal growth factor receptor; FGFR, fibroblast growth factor receptor; INSR, insulin receptor; PDGFR, platelet-derived growth factor receptor; TRK, tropomyosin receptor kinase; VEGFR, vascular endothelial growth factor receptor; DDR, discoidin domain receptor; LTK, leukocyte receptor tyrosine kinase; MET, mesenchymal epithelial transition factor; ROR, retinoic acid receptor–related orphan receptor; TIE, tyrosine kinase with immunoglobulin-like and EGF-like domains; ILR, interleukin receptor; TNFR, tumor necrosis factor receptor; CSFR, colony-stimulating factor receptor; IFNR, interferon receptor; CCR, chemokine (C-C motif) receptor; CXCR, chemokine (C-X-C motif) receptor; TLR, Toll-like receptor; FZD, frizzled receptor; Ptch, patched. The complete list of 194 genes is shown in file S1. (B) Reproducibility of receptome profiling across assay replicates. Receptor transcripts detected in at least one assay replicate (red, one replicate; blue, two replicates) were plotted as log2 relative abundance estimated by qRT-PCR cycle threshold assuming 100% amplification efficiency. En dash (–) indicates not detected. The Spearman (ρ) and Pearson (R) correlation coefficients are shown for DM13 cells, and the complete set of pairwise comparisons is shown in fig. S1.

We designed qRT-PCR primers for each gene in the receptome and individually optimized the primers so that they produced a consistent amplicon size under the same rapid-cycling conditions (see Materials and Methods). During the initial primer validation, we diagnosed correct amplicons by melt curve analysis, gel electrophoresis, and (when necessary) sequencing. The validation experiments produced a verified list of gene-specific melting temperatures for direct assessment of receptor transcript presence-absence after each profiling experiment (file S1).

Because qRT-PCR of extremely low abundance targets can be sporadic (18), we profiled the receptome of each sample in separate duplicates. Between duplicate qRT-PCR plates, we observed strong pairwise correlations in cycle threshold (CT) values (median Spearman ρ = 0.84, Pearson R = 0.78) (Fig. 1B and fig. S1). This indicated that plate-to-plate amplification efficiencies were comparable and that average CT values could be used as a semiquantitative log2 measure of relative transcript abundance across independent qRT-PCRs. In addition, receptor transcript status could be qualitatively scored as present or absent on the basis of whether specific amplification was detected in at least one of two replicates. Analysis of blank qRT-PCRs lacking sample indicated that the leading cause for missed detection in a replicate was competition of the desired amplicon with nonspecific primer-dimer products that arose during the late cycles of amplification. We reduced these artifacts by minimizing the primer concentration while maintaining the amplification efficiency of the desired RT-PCR product (file S1). Nonetheless, a few receptor transcripts (EPHA8, ERBB2, NRTK1, IL2RB, IL22RA2, TNFRSF10A, and TNFRSF25) were significantly variable (P < 0.01, Bonferroni-corrected binomial test) because of primer-dimer competition, reinforcing the need for duplicate measurements across the receptome.

To investigate the sensitivity of qRT-PCR receptome profiling, we selected HT-29 colon adenocarcinoma cells treated with interferon-γ (IFN-γ), which previously served as the base condition for a large signaling data set (1921). We assessed transcript presence or absence in HT-29 cells exposed to IFN-γ by receptome profiling and by transcriptional profiling with conventional oligonucleotide microarrays. Compared to the present-absent calls of the commercial microarray analysis software, we found that receptome profiling was more sensitive (Fig. 2A and file S2). Only eight receptor transcripts were called present by microarray and absent by receptome profiling, and literature suggests that several of these receptors are false-positives on the microarray. For example, both ERBB4 and EPOR were called present by microarray, but ERBB4 mRNA (22), ERBB4 protein (23), and EPOR protein and receptor signaling (24) are undetectable in HT-29 cells. By contrast, qRT-PCR receptome profiling detected 54 additional receptor transcripts that were called absent by microarrays. Many of these additional receptors have been detected or functionally validated in HT-29 cells previously (table S1). This suggested that conventional microarray present-absent calls largely reflect differences in detection rather than true presence or absence of a transcript.

Fig. 2 qRT-PCR receptome profiling is significantly more sensitive for detecting receptor transcripts than conventional oligonucleotide microarrays.

(A) Present-absent calls for 177 receptor transcripts monitored on Affymetrix U133A microarrays were compared to receptome-profiling results for HT-29 cells treated with IFN-γ (200 U/ml) for 24 hours. Statistical significance was assessed by Fisher’s exact test. (B) Detection of FAS in HT-29 cells with or without IFN-γ sensitization. (C) Caspase-3 (casp-3) cleavage in IFN-γ–treated HT-29 cells after FAS cross-linking with anti-APO (1 μg/ml) for 24 hours. (D) Replicated densitometry of caspase-3 cleavage in HT-29 cells. Data are shown as the mean relative abundance of cleaved caspase-3 (normalized so that the mean vehicle control equals one) ± SEM of three independent samples. *P < 0.05 by Welch’s one-sided t test. For (B) and (C), cells were immunoblotted for the indicated proteins, with tubulin used as a loading control. All immunoblots are representative of at least three independent experiments.

We further evaluated the specificity of receptome profiling by analyzing receptor presence or absence through a panel of independent measurements. We selected the death receptor–encoding gene FAS as a gene with lineage-specific expression (25). FAS mRNA was predicted to be absent in IFN-γ–treated HT-29 cells by microarray but present by receptome profiling (file S2). We examined FAS abundance by immunoblotting and found that it was present and its abundance was increased by IFN-γ (Fig. 2B), consistent with reports in other cell types (25). Accordingly, stimulation of IFN-γ–treated HT-29 cells with the FAS cross-linking antibody anti-APO resulted in a strong apoptotic response as indicated by caspase-3 cleavage (Fig. 2, C and D). Thus, qRT-PCR receptome profiling uncovered signaling capabilities missed by conventional oligonucleotide microarray methods.

To assess the accuracy of absent calls, we performed reciprocal experiments with two breast epithelial lines, MDA-MB-436 and MCF10A, and tested for the expression of CSF1R, encoding the cytokine receptor for macrophage colony-stimulating factor (M-CSF). qRT-PCR receptome profiling predicted that CSF1R transcripts were absent in MDA-MB-436 cells but present in MCF10A cells, whereas microarray data did not detect CSF1R in either cell line (26). Using an antibody that recognizes CSF1R protein, we immunoblotted MCF10A cell lysates and detected immunoreactive bands at the predicted molecular weight of CSF1R, which were absent in MDA-MB-436 cell lysates (Fig. 3A). We stimulated both cell lines with M-CSF and monitored extracellular signal–regulated kinase 1 and 2 (ERK1/2) phosphorylation as a downstream signaling readout. Phosphorylated ERK1/2 (pERK1/2) immunoreactivity increased significantly at 15 min after M-CSF stimulation in MCF10A cells (Fig. 3, B and C). Conversely, no increases in pERK1/2 were observed in MDA-MB-436 cells at any time after M-CSF treatment (Fig. 3D). The lack of pERK1/2 signaling in MDA-MB-436 cells was not due to a general defect in upstream kinases because we observed robust ERK1/2 phosphorylation upon epidermal growth factor (EGF) stimulation (Fig. 3E). These data indicated that MDA-MB-436 cells lack CSF1R transcripts, validating the accuracy of the absent calls made by qRT-PCR receptome profiling.

Fig. 3 qRT-PCR receptome profiling accurately distinguishes receptor absence.

(A) Detection of CSF1R in MCF10A cells but not in MDA-MB-436 cells. Asterisk marks a nonspecific band. (B) ERK1/2 phosphorylation in MCF10A cells after treatment with M-CSF (100 ng/ml) for 15 min. (C) Replicated densitometry of M-CSF–induced ERK1/2 phosphorylation in MCF10A cells. Data are shown as the mean relative abundance of ERK1/2 phosphorylation (normalized so that the mean vehicle control equals one) ± SEM of three independent samples. *P < 0.05 by Welch’s one-sided t test. (D) ERK1/2 phosphorylation in MDA-MB-436 cells after treatment with M-CSF (100 ng/ml) for 15 min. (E) ERK1/2 phosphorylation in MDA-MB-436 cells after treatment with EGF (100 ng/ml) for 5 min. For all immunoblot panels, cells were immunoblotted for the indicated proteins, with tubulin used as a loading control, and data are representative of at least three independent experiments.

qRT-PCR receptome profiling is more sensitive than exon arrays

Conventional oligonucleotide microarrays are heavily 3′-biased and thus lack the probe density of newer arrays that target all known exons (27). Bioinformatic comparisons between exon-targeted and 3′-biased arrays have suggested that exon arrays are more sensitive and specific for detecting expressed transcripts than 3′ arrays (28). This raised the possibility that exon array data would compare more favorably with qRT-PCR receptome profiling for predicting receptor presence or absence.

To make the direct comparison, we prepared total RNA from IFN-γ–treated HT-29 cells, MCF10A cells, and MDA-MB-436 cells and hybridized these samples to Human Exon 1.0 ST arrays. Exon arrays do not provide a discrete present-absent call, so we analyzed the receiver operating characteristics (ROC) of the background-corrected expression index (29) for each transcript with respect to the corresponding present-absent call made by qRT-PCR receptome profiling (Fig. 4A). At a false-positive rate of 10%, we found that exon arrays achieved a true-positive rate of 40 to 55% for signaling receptor transcripts, consistent with earlier transcriptome-wide analyses (47% true-positive rate relative to serial analysis of gene expression) (28). FAS transcripts were readily detected in IFN-γ–treated HT-29 cells below the 10% false-positive rate (Fig. 4A), illustrating that exon arrays are more sensitive than 3′ arrays for certain targets.

Fig. 4 qRT-PCR receptome profiling is more sensitive for detecting receptor transcripts than exon arrays.

(A) ROC curves relating exon array expression index to qRT-PCR present-absent calls for the indicated cells. HT-29 cells were treated with IFN-γ (200 U/ml) for 24 hours. In all three ROC curves, the dashed line indicates expression index = 20, a representative threshold that properly distinguishes CSF1R expression in MCF10A and MDA-MB-436 cells. The area under the ROC curve (AUC, integrated from zero to one false-positive rate) indicates the overall quality of the present-absent classification based on exon array data, with AUC = 1 indicating perfect classification and AUC = 0.5 indicating random guessing. (B) Detection of IL2RG in HT-29, MCF10A, and MDA-MB-436 cells but not in A375 cells. Cells were immunoblotted for IL2RG, with tubulin used as a loading control. Immunoblots are representative of three independent experiments.

For CSF1R, however, we found that a false-positive rate of 40 to 60% must be tolerated to distinguish MCF10A and MDA-MB-436 cells properly (Fig. 4A). At this relaxed expression threshold (expression index = 20), the γ subunit of the IL-2 receptor IL2RG was predicted by exon arrays to be present in HT-29 cells and absent in MCF10A and MDA-MB-436 cells (Fig. 4A). By contrast, qRT-PCR receptome profiling predicted that IL2RG should be present in all three but absent in a fourth cell line, A375 melanoma cells (file S3). Functional testing of IL2RG presence through IL-2 stimulation was not possible because receptome profiling indicated that these cell lines lacked one or more of the requisite subunits for the IL-2 receptor heterotrimer (IL2RA and IL2RB). Nevertheless, we found by immunoblotting that IL2RG was detected in HT-29, MCF10A, and MDA-MB-436 cells but not in A375 cells, and the relative abundance of IL2RG protein was consistent with its relative transcript abundance obtained by qRT-PCR receptome profiling (Fig. 4B and file S3). These data indicated that sensitivity remains a challenge for exon-targeted microarrays when compared to receptome profiling by qRT-PCR.

qRT-PCR receptome profiling is more specific for mature transcripts than RNA-seq

A third alternative for global receptome profiling is RNA-seq (1214), which is more sensitive than oligonucleotide microarrays (13, 15). To compare RNA-seq directly with qRT-PCR receptome profiling, we magnetically purified positive-strand polyadenylated [poly(A)+] RNA from lysates of HT-29 cells treated with IFN-γ, MDA-MB-436 cells, or MCF10A cells and sequenced at two depths: 25 million (M) reads and 50M reads (IFN-γ–treated HT-29, MDA-MB-436) or 25M reads and 100M reads (MCF10A). As expected, the RNA-seq analyses were strongly correlated across duplicates (fig. S2A), and the 50M to 100M analyses detected sequences from substantially more genes than the matched 25M analyses (fig. S2B). The RNA-seq data provided an unbiased, comprehensive, and replicated set of measurements to compare with qRT-PCR receptome profiling.

We normalized the RNA-seq data to yield relative transcript abundances as reads per kilobase per million mapped reads (RPKM) (14) and generated ROC curves with respect to the qRT-PCR present-absent calls made by receptome profiling. For nontumorigenic MCF10A cells, there was a strong concordance between RPKM and qRT-PCR receptome profiling, which improved slightly with the depth of sequencing (Fig. 5A). For instance, using a detection threshold of 0.3 RPKM (8), we found that RNA-seq could correctly distinguish the presence of CSF1R in MCF10A cells (~0.6 RPKM) from its absence in MDA-MB-436 cells (~0.02 RPKM) (Fig. 5, A and B). However, for MDA-MB-436 cells, the RPKM-qRT-PCR agreement was much poorer because false-positives increased proportionally with false-negatives for most RPKM thresholds (Fig. 5B). This pattern was also observed in IFN-γ–treated HT-29 cells, with false-positives increasing abruptly at thresholds as high as 10 to 20 RPKM (Fig. 5C). The discrepancies in the two cancer cell lines were not resolved by deeper sequencing (Fig. 5, B and C, lower graphs), suggesting a fundamental difference between RNA-seq and qRT-PCR receptome profiling.

Fig. 5 qRT-PCR receptome profiling is more specific for detecting functional receptor genes than RNA-seq.

(A to C) ROC curves relating RNA-seq RPKM to qRT-PCR present-absent calls for MCF10A cells analyzed at 25M total reads or 100M total reads (A), MDA-MB-436 cells analyzed at 25M total reads or 50M total reads (B), and IFN-γ–treated HT-29 cells analyzed at 25M total reads or 50M total reads (C). The dashed line indicates a previously reported RPKM threshold for gene detection by RNA-seq (8). Note: The larger the RPKM value, the more abundant the transcript is predicted to be and the more likely the receptor transcript is to be identified as present. The area under the ROC curve (AUC) is shown as in Fig. 4. (D) Detection of FGFR1 by immunoblot in various colon cancer cell lines and MCF10A cells but not in IFN-γ–treated HT-29 cells. (E) Detection of ERBB3 by immunoblot in IFN-γ–treated HT-29 cells and MCF10A cells but not in MDA-MB-436 cells. (F) Coverage of RNA-seq reads across portions of the FGFR1 (upper) and ERBB3 (lower) loci for the indicated cell lines. Introns showing consistent coverage above background are underlined in green. The chromosomal position of each locus is indicated in red. Blue boxes are exons, blue lines are introns, and blue ticks indicate the direction of transcription.

To determine which data type corresponded more closely to signaling competency, we selected two receptors with large RPKM values that were predicted to be absent by qRT-PCR receptome profiling. The fibroblast growth factor receptor 1–encoding gene FGFR1 is overexpressed in some colon cancers (30) and was detected at ~20 RPKM in IFN-γ–treated HT-29 cells, but FGFR1 transcripts were predicted to be absent by receptome profiling. Using a C-terminal antibody recognizing multiple splice variants of FGFR1, we detected FGFR1 in multiple colon cancer cell lines but not in IFN-γ–treated HT-29 cells (Fig. 5D). Another discrepancy was found with the epidermal growth factor receptor family member ERBB3, which was sequenced at ~3 RPKM in MCF10A cells and was present by qRT-PCR, ~90 RPKM in MDA-MB-436 cells and was absent by qRT-PCR, and ~0.4 RPKM in HT-29 cells and was present by qRT-PCR. We immunoblotted for the cytoplasmic domain of ERBB3 and detected strong immunoreactivity in MCF10A cells, which was weaker in HT-29 cells and absent in MDA-MB-436 cells (Fig. 5E), consistent with the relative abundances predicted by qRT-PCR receptome profiling (file S3). Therefore, abundant RNA-seq alignments did not necessarily correspond to functional receptors in cancer cells.

We examined FGFR1 and ERBB3 further by inspecting the coverage of aligned sequences across each locus. In both instances where the corresponding protein was absent despite high RPKM, we identified a subset of introns that were detected, suggesting incomplete splicing or aberrant intron retention (Fig. 5F, green). The observed introns were not likely caused by assembly or alignment errors because we obtained multiple paired-end reads spanning the intron-exon junctions of each retention event. MDA-MB-436 and HT-29 cells showed the same overall coverage of intronic and other noncoding RNA sequences compared to MCF10A cells (fig. S3A). However, when focusing on the putative false-positives detected by RNA-seq in the cancer cell lines, we found that these genes were significantly enriched for intronic sequences relative to receptor transcripts that were also called present by qRT-PCR (fig. S3B). These data suggest that incompletely spliced RNA sequences can be discriminated more effectively by qRT-PCR–based profiling than by current implementations of RNA-seq.

Receptome profiling defines signaling signatures enriched in specific tissue lineages

To demonstrate an application of receptome profiling, we surveyed the signaling receptomes of 40 human cell lines (file S3). The collection was weighted toward pancreatic, melanocytic, breast, and colonic lineages to evaluate the link between receptome signatures and tissue origin. As expected (31), we found that receptome signatures clustered significantly according to lineage (Fig. 6A and fig. S4). Lineage enrichment was associated with the high abundance of certain signaling receptor transcripts (Fig. 6B, yellow, and fig. S5). For example, transcripts for the receptor tyrosine kinase ERBB3 were increased among breast epithelia, which may explain why some breast cancers have amplification of ERBB2, which encodes a dimerization partner of ERBB3 (32). Many tissue-enhanced patterns were supported by previous studies, although roughly half of the patterns uncovered by receptome profiling had not been described to our knowledge (Table 1).

Fig. 6 Receptome profiling of 40 cell lines reveals that lineage is not defined by the qualitative presence of signaling receptors.

(A and B) Lineage characterization by the relative abundance of specific receptor transcripts. One-way (A) and two-way (B) hierarchical clustering of receptor-specific relative abundance after normalization to GAPDH as a loading control. (C and D) Lineage characterization by the absence of specific receptors. One-way (C) and two-way (D) hierarchical clustering of high-sensitivity present-absent calls from receptome profiling. Yellow boxes in (B) and (D) highlight local clusters of receptor patterns that are lineage-specific. Clustering was done using a Euclidean distance metric with Ward’s linkage. Dendrogram branches significantly enriched for specific lineages (P < 0.05) are matched to the color associated with the lineage. Enrichment analysis for cell lineages was performed by the hypergeometric test. n.d., not detected. See figs. S4 to S7 for expanded views of (A) to (D) that include individual receptor gene names. See file S3 for the data.

Table 1 Signaling receptors with abundant transcripts indicating lineage-specific gene expression.
View this table:

To exploit the qualitative sensitivity of receptome profiling, we removed all quantitative information and reclustered the 40 cell lines on the basis of the presence or absence of receptor transcripts. The binary present-absent signature was sufficient to categorize much of the cell line panel according to lineage (Fig. 6C and fig. S6). For the tissue types in the panel, lineage enrichment was not associated with tissue-selective presence of receptor subsets but rather with the absence of transcripts (Fig. 6D, yellow, and fig. S7). Using strict criteria for lineage specificity (see Materials and Methods), we identified seven receptors with tissue-specific absence but only one receptor with tissue-specific presence (Table 2). For example, the chemokine receptor XCR1 was absent in 8 of 10 melanoma lines (P < 0.005, hypergeometric test), consistent with the reported loss of XCR1 in culture compared to primary melanoma tumors (33). A few absent signatures could be inferred from literature reports, but most had not been reported previously (Table 2).

Table 2 Lineage-specific presence or absence of signaling receptors.
View this table:

Ectopic expression of IL10RA in melanoma cells engages an artificial autocrine circuit

We examined the impact of receptor absence on cell function by selecting the IL-10 receptor α subunit IL10RA for follow-up studies. IL10RA was called absent in 9 of 10 melanoma lines by qRT-PCR receptome profiling (Table 2 and file S3), which was notable because melanoma cells are a source of anti-inflammatory IL-10 (34, 35). To determine whether the absence of IL10RA influenced cell behavior, we used A375 melanoma cells, which lack IL10RA but constitutively secrete IL-10 (36). We transduced the cells with either a control luciferase-expressing lentivirus or a lentivirus encoding IL10RA. As expected, IL10RA was not detectable in control luciferase-expressing A375 cells but was present in cells transduced with IL10RA (Fig. 7A). The IL10RA-expressing cells also showed phosphorylation of signal transducer and activator of transcription 3 (STAT3) upon stimulation with recombinant IL-10, whereas the control A375 cells were unresponsive (Fig. 7, B and C). Therefore, the A375 melanoma cell line has all the intracellular machinery for transducing an IL-10 signal except for IL10RA, which acts as a gatekeeper for conferring cellular responsiveness.

Fig. 7 Forced expression of IL10RA in melanoma cells creates an autocrine signaling loop that alters signaling, gene expression, and cell responses.

(A) IL10RA abundance after lentiviral transduction of A375 cells with IL10RA or a luciferase control. IL10RA was detected by immunoblotting. (B) STAT3 phosphorylation in luciferase- or IL10RA-expressing A375 cells after treatment with IL-10 (20 ng/ml) for 20 min. (C) Replicated densitometry of IL-10–induced STAT3 phosphorylation in A375 cells, normalized so that the mean relative abundance of STAT3 phosphorylation in unstimulated luciferase-expressing cells equals one. (D) ELISA quantification of IL-10 in the conditioned medium of luciferase- or IL10RA-expressing A375 cells. (E) qRT-PCR quantification of IL10 mRNA abundance in luciferase- or IL10RA-expressing A375 cells, normalized so that the geometric mean of relative abundance of IL10 mRNA in luciferase-expressing cells equals one. (F) Decrease in baseline STAT3 phosphorylation for A375 cells ectopically expressing IL10RA. Replicated densitometry is normalized so that the mean relative abundance of STAT3 phosphorylation in luciferase-expressing cells equals one. (G) Caspase-3 cleavage (ClvC3) in luciferase- or IL10RA-expressing A375 cells after FAS cross-linking with anti-APO (1 μg/ml) for 24 hours. Replicated densitometry is normalized so that the mean relative abundance of cleaved caspase-3 in anti-APO–treated luciferase-expressing cells equals one. (H) qRT-PCR quantification of NF-κB target genes in luciferase- or IL10RA-expressing A375 cells after treatment with TNF (100 ng/ml) for the indicated time points. NF-κB target genes are normalized so that the geometric mean of relative abundance of the indicated transcript in unstimulated luciferase-expressing cells equals one. For (A), (B), (F), and (G), cells were immunoblotted for the indicated proteins with tubulin or procaspase-3 (ProC3) used as a loading control. For (C), (D), (F), and (G), quantitative data are shown as the means ± SEM of three independent samples. For (E) and (H), data are shown as the geometric mean ± log-transformed SEM of four independent samples. *P < 0.05 by Welch’s one-sided t test (G) or by log-transformed two-way analysis of variance (ANOVA) with Sidák posttest correction (H). All immunoblots are representative of at least three independent experiments.

To determine whether IL10RA had engaged an artificial autocrine circuit in A375 cells, we analyzed the concentration of IL-10 in conditioned medium by enzyme-linked immunosorbent assay (ELISA). IL-10 was readily detected in the medium conditioned by control cells but not in medium conditioned by IL10RA-expressing cells (Fig. 7D). By contrast, IL10 mRNA abundance was the same in the control and IL10RA-expressing cells (Fig. 7E), suggesting that the absence of IL-10 in the medium of IL10RA-expressing cells could be the result of autocrine trapping. We also noted a ~60% reduction in basal STAT3 phosphorylation (Fig. 7F), which may be due to chronic IL-10 signaling causing feedback desensitization of other STAT3-activating pathways in IL10RA-expressing cells (37).

To test whether the IL10RA-triggered autocrine circuit was sufficient to affect cellular responses, we stimulated receptors of the tumor necrosis factor (TNF) family that were detected in A375 cells by qRT-PCR receptome profiling (file S3). IL10RA slightly increased the resistance of A375 cells to apoptosis induced by FAS cross-linking with anti-APO (Fig. 7G). The transcriptional signature of nuclear factor κB (NF-κB) target genes was also altered when IL10RA-expressing A375 cells were stimulated with TNF (Fig. 7H). Whereas some NF-κB targets, such as NFKBIA and CCL2, were unaffected by ectopic IL10RA expression, others were significantly inhibited (IL1A and IL6; Fig. 7H). We conclude that receptor absence is important to insulate cells from secreted proteins that are meant to act as paracrine ligands. Aberrant receptor expression creates autocrine circuitry that traps paracrine factors locally and disrupts signaling, gene expression, and cellular responses.

Environmental stimuli perturb cellular receptome signatures

We explored the plasticity of cellular receptomes by profiling receptor transcript abundance in 293T embryonic kidney cells and MCF7 breast carcinoma cells after exposure to various stimuli (Fig. 8A and file S4). We used EGF as a growth factor stimulus, IFN-γ and TNF as proinflammatory stimuli, and ionizing radiation (IR) as an environmental stress. We found that most stimulus-induced changes in the abundance of receptor transcripts were relatively minor (±2-fold). This was particularly true for IR-treated samples, which gave rise to abundance changes that were highly variable across independently irradiated cultures (file S4). For the proinflammatory stimuli, however, there were several notable transcriptional responses that warranted additional analysis. In both cell lines, the TNF superfamily receptor TNFRSF9 was strongly induced upon TNF stimulation (Fig. 8B), consistent with a previous report (38). We also observed many changes that were specific to cell type, indicating context-specific transcriptional programs. For example, the abundance of the insulin receptor transcript INSR mildly increased in TNF-stimulated MCF7 cells (Fig. 8B). MCF7 cells also showed selective increases in TLR3 and IL15RA transcripts upon stimulation with IFN-γ, which were not observed in 293T cells (Fig. 8B). This difference cannot be attributed to a general lack of IFN-γ responsiveness because 293T cells abundantly express the transcripts of the required receptors (file S3) and IFN-γ triggers changes in the abundance of other transcripts (Fig. 8A) (39). The induction of IL15RA upon IFN-γ treatment of MCF7 cells agrees with a previous study (40), and we further found that IL15RA was also induced in MCF7 cells by TNF (Fig. 8B). TNF-stimulated transcription of IL15RA has not been previously reported, illustrating that receptome profiling can be used as a discovery tool to link environmental changes to transcriptional signatures of other environmental sensors.

Fig. 8 Stimulus-dependent changes in receptome profiles are dependent on cell type.

(A) Receptor abundances in 293T embryonic kidney cells and MCF7 breast carcinoma cells after stimulation with EGF (100 ng/ml) for 4 hours, IFN-γ (200 U/ml) for 4 hours, 5-Gy IR for 2 hours, or TNF (20 ng/ml) for 4 hours. One-way hierarchical clustering was done using a Euclidean distance metric with Ward’s linkage after normalization to GAPDH as a loading control. Data were centered on the cell type–matched untreated (No tx) condition or the median observed abundance across both cell types if the receptor was absent for one of the untreated conditions. (B) Plate-matched qRT-PCR quantification of the indicated receptor transcripts in 293T or MCF7 cells treated with TNF or IFN-γ, normalized so that the geometric mean of relative abundance of the indicated transcript in unstimulated cells equals one. For (A), data are shown as the median cycle threshold (approximate log2 relative abundance) of three independent biological samples. For (B), data are shown as the geometric mean ± log-transformed SEM of three independent biological samples. *P < 0.05 by log-transformed, unpaired one-sided t test. n.d., not detected. See fig. S8 for an expanded view of (A) that includes all individual receptor gene names. See file S4 for the data.

qRT-PCR receptome profiling is compatible with primary human tissues

To illustrate that receptome profiling can be applied to primary tissue samples, we profiled primary specimens of brain and skeletal muscle (Fig. 9 and file S5). Compared to the cell lines, we detected transcripts from significantly fewer types of receptors in the primary tissues (P < 10−6, binomial test assuming 78% of receptors are present based on Fig. 6C). Although some transcripts may have been lost during sample isolation, we attributed the restricted overall expression pattern to the highly specialized tissues examined. Many specific receptors detected in one or both tissues were consistent with the known biology, including the presence of GHR (encoding growth hormone receptor) in muscle (41), SMO and PTCH2 (encoding the Hedgehog receptor Smoothened and its co-receptor target Patched) and FZD family (encoding the Wnt receptors of the Frizzled family) transcripts in brain (42, 43), and INSR in both brain and muscle (44, 45). Conversely, some receptor transcripts that were ubiquitous in cultured epithelial cells, such as EPHA2 and EPHB4, were absent in the brain or muscle isolates, corroborating their reported tissue distribution (46, 47). We conclude that qRT-PCR receptome profiling is a versatile approach for systematic interrogation of canonical receptors involved in cell signaling.

Fig. 9 Receptome profiling is compatible with primary tissue samples.

(A and B) Relative receptor abundances (A) and present-absent calls (B) obtained by receptome profiling in a primary human brain sample and a primary human muscle sample. n.d., not detected. Receptor transcripts mentioned in the text are highlighted. See fig. S9 for an expanded view that includes all individual receptor gene names. See file S5 for the data.


The presence or absence of signaling receptors determines a cell’s ability to respond to its environment. By defining a receptome panel and validating each qRT-PCR reagent in the array individually, we provide a convenient tool for establishing the boundaries of cellular responsiveness to the ligands that activate these receptors. Detection of the mRNA of a receptor does not always imply that this receptor will be properly translated and localized to bind ligands and transmit signals. However, we showed that lack of mature receptor transcripts was consistent with cellular unresponsiveness (Figs. 3, A and D, and 7, A and B), a finding that required the sensitivity and specificity of the profiling approach described here.

We verified that qRT-PCR receptome profiling is substantially more sensitive for discerning receptor presence or absence than microarrays, irrespective of the microarray probe coverage along the transcript. This result was expected considering the stringency of microarray hybridization that is required to gauge specificity reliably using perfect match and mismatch probes. More surprising was the superior specificity of qRT-PCR–based profiling compared to RNA-seq when the receptomes of cancer cells were profiled. The difference here may be related to the methods used for mRNA isolation during the two measurement techniques. Our first-strand synthesis for qRT-PCR is primed with oligo(dT)24, and high-stringency reverse transcription is performed at 50°C, ensuring that most, if not all, complementary DNAs (cDNAs) contain at least poly(A)24 (48). For RNA-seq, however, poly(A)+ transcripts are isolated by magnetic separation after room temperature annealing to oligo(dT)25 beads, which may copurify mRNAs with much shorter oligo(A) tails. The distinction is important because shorter oligo(A) tails remain on transcripts undergoing nonsense-mediated decay, which is triggered when premature stop codons are encountered after aberrant splicing events, such as intron retention (49, 50). Nonsense-mediated decay may be specifically enhanced in cancer cells to suppress antitumor immune responses (51, 52), which could explain why we observed most RNA-seq discrepancies in transformed cells. The discrepancies can, in theory, be avoided by sequencing the transcriptome as oligo(dT)24-primed cDNA, but this decreases the uniformity of coverage along transcripts (14), which is a major advantage of RNA-seq.

Our collection of qRT-PCR receptome profiles across 40 human cell lines complements other work, showing that an exceedingly small fraction of proteins is detected in a purely cell- or tissue-specific manner (53). Furthermore, the binary present-absent signatures indicate that receptor silencing might be just as important in defining a lineage as the receptors that are highly abundant. Receptor silencing could be an important mechanism for enabling effective paracrine communication without the complications of autocrine cross talk. For example, forced expression of IL10RA in melanoma cells would not only sequester IL-10 away from neighboring immune cells but would also severely dampen the induction of anti-inflammatory signals, such as IL6. We confirmed these predictions by showing that the IL10RA-expressing cells had less IL-10 in the medium (Fig. 7D) and produced fewer IL6 transcripts in response to TNF (Fig. 7H). Thus, IL10RA-harboring melanoma cells would be predicted to be more immunogenic overall than their naturally occurring counterparts. Besides lineage-specific silencing, it may also be worth examining receptors that are lost in individual cancer lines to get a sense of how transformed cells evolve resistance to ligands that inhibit tumor growth.

A major challenge for deciphering the microenvironment is the complex cocktail of ligands that cells encounter physiologically (54). We can gain a clearer understanding of the microenvironment by distinguishing the ligands that activate intracellular signaling from those that are ignored. The qRT-PCR array described here provides a straightforward and scalable way to make this discrimination. With better sensitivity than microarrays and better specificity than RNA-seq at less than one-tenth of the cost, qRT-PCR receptome profiling could be readily incorporated into large-scale characterizations of cell lines, primary tissues, and tumors.

Materials and Methods

Cell culture

293T, CCRF-CEM, DLD-1, HCT-8, HCT-15, HT-29, AU-565, HCC1500, MCF7, MDA-MB-231, MDA-MB-361, MDA-MB-468, AsPC-1, BxPC-3, Capan-1, Capan-2, CFPAC-1, HPAF-II, L3.6 pl, Mia PaCa-2, Panc-1, PaTu 8902, PL45, SU.86.86.86, SW1990, and Yap-C cells were cultured according to American Type Culture Collection recommendations. MDA-MB-231, MDA-MB-361, and MDA-MB-468 cells were cultured without CO2. The 5E clone of MCF10A cells was cultured as described (55, 56). MDA-MB-436 cells were cultured in L-15 medium with 10% fetal bovine serum (FBS) without CO2. Human pancreatic duct epithelial cells were cultured as described (57). A375, HT144, SK-MEL-2, and SLM2 cells were cultured in RPMI medium with 10% FBS and 5% CO2. HeLa cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) with 10% FBS and 5% CO2. DM13, DM122, DM331, SK-MEL-24, VMM18, and VMM39 cells were cultured in RPMI medium with 5% FBS and 5% CO2.

Primary tissues

Brain and muscle samples were obtained as anonymized, snap-frozen cadaver tissue through the Biorepository and Tissue Research Facility at the University of Virginia.

Primer design

The receptors were selected from the Human Plasma Membrane Receptome (17), and receptor RefSeq mRNA sequences were obtained from the National Center for Biotechnology Information (NCBI). Primers were designed using Primer3 (58), with each search constrained to a product size of 150 to 200 bp, primer sequence length of 18 to 22 bp, and GC (guanine-cytosine) content 40 to 60%. The specificity of primer targets was verified using NCBI’s Basic Local Alignment Search Tool (BLAST). The generality of primer sets was confirmed with the NCBI single-nucleotide polymorphism database to ensure that no reported polymorphism was located at the 3′ end of any primer.

Quantitative RT-PCR

RNA from cultured cells was isolated with the RNeasy Plus Mini kit (Qiagen) according to the manufacturer’s protocol. RNA from primary tissues was isolated with RNA STAT-60 (Tel-Test) after homogenization on a TissueLyser LT (Qiagen). First-strand cDNA synthesis and qRT-PCR were performed as described (48). For qRT-PCR experiments other than those used to generate receptome profiles, data were normalized to the geometric mean of three housekeeping genes, and stability of the normalization was qualitatively assessed with a fourth housekeeping gene among the following candidates: GAPDH, HINT1, PPIA, PRDX6, B2M, and GUSB.

Primer validation

Each primer set was tested on cDNA together with a no reverse transcription sample to control for genomic contamination and a blank sample to control for primer dimers. The size of any amplicon above a melting temperature of 77°C was verified by gel electrophoresis to confirm the expected amplicon size. Ambiguous amplicons were gel-purified and analyzed by conventional DNA sequencing. Primer concentration was initially set at 10 pmol per 15 μl of reaction and was empirically adjusted to optimize the specificity of amplification (file S1).

Receptor profiling assay

Primers were lyophilized in 96-well low-profile PCR plates (Bio-Rad) for 24 hours at 0.110 mbar (Labconco). Reverse-transcribed cDNA (10 μl) was diluted in 740 μl of H2O and mixed on ice with 750 μl of 2× master mix: 2× PCR Buffer II (Applied Biosystems), 8 mM MgCl2, 400 μM dNTP (deoxynucleotide triphosphate), bovine serum albumin (300 μg/ml), 10% glycerol, 0.5× SYBR Green (Invitrogen), and Taq polymerase (0.05 U/ml) (Roche) (48). Master mix–cDNA mixture (15 μl) was loaded into each well of the lyophilized plate, and qRT-PCR was performed on a CFX96 real-time PCR instrument (Bio-Rad) with the following amplification protocol: denaturation at 95°C for 90 s; amplification cycles of 95°C for 10 s, 60°C for 10 s, and 72°C for 12 s repeated 40 times; a fusion step of 65° to 95°C increased at a rate of 0.1°C/s and measured at 0.5°C increments. For each plate, GAPDH was used as the loading control, and a blank well with no primer served as a negative control.

Raw receptor-profiling data were extracted using CFX Manager 2.0 (Bio-Rad). The baseline for cycle threshold values (CT) was set at 25 RFU (relative fluorescence units), and the baseline for the melting temperature (Tm) estimate was set at 15 −d(RFU)/dT. These data were exported and compared against a database of Tm ranges for each receptor amplicon (files S1 and S6) to make present-absent calls on the basis of specificity of the amplification. Absent calls were made only if both duplicate runs were called absent.

For relative quantification, CT values were normalized to GAPDH from each plate, and the CT values from duplicate runs of each receptor were averaged. If only one run was called present, we used the CT value from that run. For clustering, absent genes were nominally assigned a CT value that was three cycles higher than the highest CT value observed for that gene. MATLAB software for present-absent calls and relative quantification is available as file S6 along with an example data set to illustrate plate layouts and formatting.

Oligonucleotide microarrays

HT-29 cells were plated at 50,000 cells/cm2 for 24 hours and stimulated with IFN-γ (200 U/ml) (Roche) for 24 hours. RNA was isolated with the RNeasy Mini Kit (Qiagen), and overall RNA integrity was confirmed with a Bioanalyzer (Agilent). Biotin-labeled complementary RNA (cRNA) was prepared using the T7-based BioArray HighYield RNA Transcript Labeling Kit (Enzo), and samples were hybridized to GeneChip Human Genome U133A Arrays (Affymetrix) and scanned according to the manufacturer’s recommendations.

Processing and analysis of microarray data

The scanned images of HT-29 microarrays were analyzed using Expression Console 1.1 (Affymetrix). The Microarray Suite (MAS) 5.0 algorithm was used to determine present-absent calls of the HT-29 data and the raw breast cancer cell line microarray data (ArrayExpress #E-TABM-157) (26). The HT-29 data were then compared to the receptome-profiling data as follows. For genes that have multiple probes, a present call was made if at least one probe with an “_at” or “_a_at” suffix designation was called present for at least one biological sample. An absent call was made if all “_at” and “_a_at” probes for a gene for all three biological samples were called absent. If there were no “_at” or “_a_at” probes for a gene, the present-absent calls were made using probes with an “_s_at” suffix designation, and if there were no “_s_at” probes, the calls were made using probes with an “_x_at” suffix. This approach used the best available probes for each receptor transcript.

Exon array

RNA was isolated with the RNeasy Plus Mini Kit (Qiagen), and overall RNA integrity was confirmed on a Bioanalyzer (Agilent). Sense-strand cDNA synthesis was performed with the Ambion WT Expression Kit (Applied Biosystems). Briefly, double-stranded cDNA was synthesized with engineered primers containing a T7 promoter sequence. The cDNA was used as a template for antisense cRNA synthesis by in vitro transcription using T7 RNA polymerase. The cRNA was reverse-transcribed with random primers to synthesize single-stranded, sense-strand cDNA. The cDNA was then fragmented, labeled, and hybridized to a GeneChip Human Exon 1.0 ST Array (Affymetrix) according the manufacturer’s recommendations. The chips were scanned on a GeneChip Scanner 3000 7G (Affymetrix).

Processing and analysis of exon array data

Exon array data were analyzed with the R package JETTA (29). The gene expression index for each transcript cluster was calculated for the core probe set after median GC background correction and normalization by median scaling. Background correction was performed on the exon array data relative to an earlier data set of 178 human cell lines (GSE29682).

RNA sequencing

All RNA-seq data were generated by the Genomics Services Lab at the HudsonAlpha Institute for Biotechnology (Huntsville, AL). RNA was isolated with the RNeasy Plus Mini Kit (Qiagen), and 1 μg of total RNA was enriched for poly(A)+ transcripts with oligo(dT)25 Dynabeads (Invitrogen). Each cDNA library was prepared with the NEBNext First Strand Synthesis, Second-Strand Synthesis, End Repair, dA Tailing, and Quick Ligation modules (New England Biolabs). Libraries were indexed with standard Illumina-type adapters and sequenced on an Illumina HiSeq 2000 using version 3 reagents that generate 180M to 200M reads per lane. Samples were 50-bp paired-end sequenced in duplicate at 25M and 50M or 100M reads per sample.

Processing and analysis of RNA-seq data

RNA-seq reads were filtered for signal to noise (chastity filtered), assessed for overall quality with FastQC (, and then mapped with STAR 2.2.0 (59) against the human genome build hg19. Reads that map to each gene were counted with HTSeq ( under the union set, whereby reads that do not completely overlap a gene are still counted. The percentage of reads mapping to exons, introns, and untranslated regions was calculated by intersecting the data with features from the hg19 assembly by using BEDTools (60). RPKM calculations were performed by normalizing to the median transcript length for each gene and the total library size of each sample. Receptor genes with partial intronic coverage were counted manually using the Integrative Genomics Viewer (61).

ROC analysis

ROC curves were generated in R using the ROCR package (62).

Plasmids and viral transduction

V5-tagged luciferase and IL10RA vectors were prepared by recombination of donor plasmids into the lentiviral destination vector pLX302 (63) with LR Clonase (Invitrogen). Donor plasmids were verified by sequencing, and recombined plasmids were verified by restriction digest. Lentiviruses were packaged as previously described (64). Stably transduced A375 cells were selected with puromycin (1 μg/ml) until control plates had cleared.


A375 cells were plated at 50,000 cells/cm2 for 24 hours, and conditioned medium was collected. After centrifuging to remove dead cells, supernatants were analyzed for IL-10 by ELISA (R&D Systems) according to the manufacturer’s instructions.

Cell stimulation

HT-29 cells were plated at 50,000 cells/cm2 for 24 hours, pretreated with IFN-γ (200 U/ml) (Roche) for 24 hours, and stimulated with FAS cross-linking antibody (1 μg/ml) (APO-1-3, Axxora) for 24 hours. MDA-MB-436 cells were plated at 50,000 cells/cm2 for 24 hours and treated with M-CSF (100 ng/ml) (PeproTech) or EGF (100 ng/ml) (PeproTech) for the indicated times before lysis. MCF10A cells were plated at 25,000 cells/cm2 for 24 hours and treated with M-CSF (100 ng/ml) (PeproTech) for 15 min. A375 cells were plated at 50,000 cells/cm2 for 24 hours and treated with IL-10 (50 ng/ml) (PeproTech) for 20 min, FAS cross-linking antibody (1 μg/ml) (APO-1-3, Axxora) for 24 hours, or TNF (100 ng/ml) (PeproTech) for the indicated times. 293T cells were plated at 50,000 cells/cm2 and MCF7 cells at 25,000 cells/cm2 for 24 hours before stimulation with EGF (100 ng/ml) (PeproTech) for 4 hours, IFN-γ (200 U/ml) (Roche) for 4 hours, 5-Gy IR (60Co) for 2 hours, or TNF (20 ng/ml) (PeproTech) for 4 hours.

Western blot analysis

Cells were lysed in radioimmunoprecipitation assay buffer [50 mM tris (pH 7.5), 150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 0.1% SDS, 0.5% sodium deoxycholate). Equal amounts of clarified lysates (20 μg) were separated by SDS–polyacrylamide gel electrophoresis and transferred onto polyvinylidene difluoride membranes (Millipore). Membranes were blocked for 1 hour in 0.5× blocking solution (LI-COR) diluted with phosphate-buffered saline. Membranes were incubated overnight with primary antibodies recognizing the following proteins or epitopes: FAS (Cell Signaling Technology; 1:1000), CSF1R (Santa Cruz Biotechnology; 1:1000), phosphorylated ERK1/2 (T202/Y204, Cell Signaling Technology; 1:1000), ERK1/2 (Millipore; 1:1000), caspase-3 (Cell Signaling Technology; 1:1000), phosphorylated STAT3 (Y705, Cell Signaling Technology; 1:1000), STAT3 (Cell Signaling Technology; 1:1000), ERBB3 (Cell Signaling Technology; 1:1000), FGFR1 (Cell Signaling Technology; 1:1000), IL10RA (Millipore; 1:1000), IL-2Rγ (Santa Cruz Biotechnology; 1:1000), or α-tubulin (Abcam, 1:20,000, or Cell Signaling Technology, 1:1000). Subsequently, membranes were incubated with secondary IRDye-conjugated antibodies (LI-COR; 1:20,000) or with horseradish peroxidase–conjugated secondary antibodies (Jackson ImmunoResearch; 1:10,000). Protein bands were detected by an Odyssey infrared scanner (LI-COR) or by enhanced chemiluminescence (Pierce) on a ChemiDoc MP camera-based detection system (Bio-Rad). Densitometry of bands was performed in ImageJ.

Hierarchical clustering

Hierarchical clustering was performed with the clustergram function in MATLAB by the unweighted pair group method with a Euclidean distance metric and Ward’s linkage.

Statistical analysis

Comparison of the receptor present-absent calls for receptome profiling and microarrays was performed by Fisher’s exact test. qRT-PCR time courses were compared by two-way ANOVA after log transformation to allow for parametric analysis (65), using the Sidák correction to account for multiple-hypothesis testing. Individual cell stimulations were compared by unpaired one-sided t test after log transformation. Lineage enrichment within the clustered receptome profiles was determined by the hypergeometric test. Receptors with lineage-specific absence must be highly enriched for absence (P < 0.01, hypergeometric test), absent in ≥75% of the cell lines comprising that lineage, and present in ≥50% of all cell lines tested. Receptors with lineage-specific presence must be highly enriched for presence (P < 0.01, hypergeometric test), present in ≥75% of the cell lines representing that lineage, and absent in ≥50% of the cell lines tested.

Supplementary Materials

Fig. S1. Receptome profiling is consistently reproducible across assay duplicates.

Fig. S2. RNA-seq is consistently reproducible across assay duplicates with sensitivity that depends on sequencing depth.

Fig. S3. Receptor discrepancies between qRT-PCR and RNA-seq in cancer cells are enriched for retained intronic sequences.

Fig. S4. One-way hierarchical clustering of relative receptor abundances in 40 cell lines obtained by receptome profiling.

Fig. S5. Two-way hierarchical clustering of relative receptor abundances in 40 cell lines obtained by receptome profiling.

Fig. S6. One-way hierarchical clustering of high-sensitivity present-absent calls in 40 cell lines obtained by receptome profiling.

Fig. S7. Two-way hierarchical clustering of high-sensitivity present-absent calls in 40 cell lines obtained by receptome profiling.

Fig. S8. Changes in receptor abundance in 293T and MCF7 cells upon stimulation with EGF, IFN-γ, IR, and TNF.

Fig. S9. Relative receptor abundance and present-absent calls in human brain and muscle tissue.

Table S1. Literature support for present calls in HT-29 cells.

File S1. Sequences of validated qRT-PCR primers for receptome profiling.

File S2. Comparison of present-absent calls obtained by receptome profiling and oligonucleotide microarrays for HT-29 cells exposed to IFN-γ.

File S3. Mean-centered cycle thresholds and present-absent calls of the signaling receptome in 40 human cell lines.

File S4. Response of 293T and MCF7 signaling receptomes to EGF, IFN-γ, IR, and TNF.

File S5. Signaling receptomes of primary human brain and muscle tissue.

File S6. MATLAB software for present-absent calls and relative quantification.

References (101112)

References and Notes

Acknowledgments: We thank C.-C. Wang for designing qRT-PCR primers, J. Beech and D. Roller for providing cDNA samples, Y. Bao and A. Prorock of the University of Virginia (UVA) DNA Sciences Core for help with the exon arrays, and S. Turner and A. Koeppel of the UVA Bioinformatics Core for help in processing the RNA-seq data. Funding: This work was supported by the NIH Director’s New Innovator Award Program (1-DP2-OD006464), the Pew Scholars Program in the Biomedical Sciences, and the David and Lucile Packard Foundation to K.A.J. K.J.J. is partly supported by predoctoral awards from the Joanna M. Nicolay Melanoma Foundation and the Achievement Rewards for College Scientists Foundation. Author contributions: B.H.K. and J.A.H. validated the individual qRT-PCR reagents and performed all receptome-profiling experiments. B.H.K. also performed the bioinformatic analyses of the receptome-profiling and microarray data. K.J.J. performed all follow-up biochemical experiments, including the IL10RA studies with A375 cells. K.A.J. conceived the study, designed the experiments, and wrote the manuscript with assistance from B.H.K., K.J.J., and J.A.H. Competing interests: The authors declare that they have no competing interests. Data and materials availability: Microarray, exon array, and RNA-seq data are available through the NCBI Gene Expression Omnibus (GSE38516, GSE45195, and GSE45258).
View Abstract

Navigate This Article