Single-cell immunoblotting resolves estrogen receptor-α isoforms in breast cancer

An array of isoforms of the nuclear estrogen receptor alpha (ER-α) protein contribute to heterogeneous response in breast cancer (BCa); yet, a single-cell analysis tool that distinguishes the full-length ER-α66 protein from the activation function-1 deficient ER-α46 isoform has not been reported. Specific detection of protein isoforms is a gap in single-cell analysis tools, as the de facto standard immunoassay requires isoform-specific antibody probes. Consequently, to scrutinize hormone response heterogeneity among BCa tumor cells, we develop a precision tool to specifically measure ER-α66, ER- α46, and eight ER-signaling proteins with single-cell resolution in the highly hetero-clonal MCF-7 BCa cell line. With a literature-validated pan-ER immunoprobe, we distinguish ER-α66 from ER-α46 in each individual cell. We identify ER-α46 in 5.5% of hormone-sensitive (MCF-7) and 4.2% of hormone-insensitive (MDA-MB-231) BCa cell lines. To examine whether the single-cell immunoblotting can capture cellular responses to hormones, we treat cells with tamoxifen and identify different sub-populations of ER-α46: (i) ER-α46 induces phospho-AKT at Ser473, (ii) S6-ribosomal protein, an upstream ER target, activates both ER-α66 and ER-α46 in MCF-7 cells, and (iii) ER-α46 partitions MDA-MB-231 subpopulations, which are responsive to tamoxifen. Unlike other single-cell immunoassays, multiplexed single-cell immunoblotting reports–in the same cell–tamoxifen effects on ER signaling proteins and on distinct isoforms of the ER-α protein.


Introduction
The estrogen receptor-α (ER-α66, Uniport P03372) is a steroid receptor expressed or overexpressed in~75% of breast cancers (BCa) [1][2][3][4]. To block ER-α66 overexpression, adjuvant hormone therapies including tamoxifen (TAM) are used. TAM is a nonsteroidaltriphenylethylene selective estrogen receptor modulator (SERM) that was structurally derived from diethylstilbestrol-like estrogens and antiestrogens [5][6][7]. TAM mediates canonical ER signaling action, in which ER-α66 binds to estrogen response element (ERE) sites in DNA, thereby triggering transcription of estrogen-dependent genes [8]. However, BCa is a heterogeneous disease such that classification based on nuclear ER-α66 may be insufficient for hormone therapy selection [9]. Based on the Early Breast Cancer Trialists Group meta-analysis of 46,000 women who were disease-free after the first 5 years of hormone therapy, 21% of stage I patients had recurrence events at 20 years, 14% of which were distant metastasis [10][11][12]. Nearly all late-stage BCa patients develop clinical resistance to hormone therapies via a variety of mechanisms [13,14]. A single-cell tool that discerns full-length and truncated ER-α isoforms may provide an insight for BCa response to hormone therapy. ER-α46 (46 kDa form of the 66 kDa full-length protein) is an alternatively spliced isoform with a missing activation function (AF-1) at the Nterminus. ER-α46 dimerizes with the full-length ER-α66 form to repress transcription [15][16][17]. Further, overexpression of ER-α46 has been observed to partially recover hormone sensitivity in hormone-insensitive BCa cell lines [18,19].
Despite being implicated in hormone response, ER-α46 is difficult to distinguish from ER-α66 at the single-cell level. Widely used for biomarker discovery and cancer prognosis, protein microarrays and immunohistochemistry (IHC) [20,21] identify cell-to-cell variation in oncoprotein expression. Because of homology between ER-α46 and ER-α66, isoform-specific antibodies are unable to distinguish ER-α46 from ER-α66 [16,17]. Imaging mass cytometry offers subcellular resolution and target multiplexing (>30 protein), but like all immunoassays requires isoform-specific antibodies to distinguish ER-α46 from ER-α66 [22,23]. Slab-gel immunoblotting resolves protein targets by differences in molecular mass and immunoprobing with a pan-ER antibody. Given detection sensitivity limitations of slab-gel immunoblotting, pooling of cells is required for detection. Pooling of cells obscures sub-populations with protein expression differences. Consequently, a single-cell tool that offers sub-population resolution and multiplexing of ER signaling is needed [24].
Here, we develop a single-cell immunoblotting that classifies BCa subtypes based on 10 protein targets involved in ER signaling, including the challenging separation of ER-α66 and ER-α46 isoforms, as described above. Seeking to validate single-cell detection of clonal subpopulations, we follow the studies of Leung, et al. [25] and Nugoli, et al. [26] and scrutinize the BCa cell line MCF-7 owing to expected high hetero-clonality and genetic plasticity. We utilize the monoclonal pan-ER (SP-1, C-terminal domain) antibody-tested in BCa cell lines [27,28], mouse models [29], patient tumor ER-α status [30,31]-as an immunoreagent to detect the frequency and expression levels of ER-α isoforms. As a negative control cell line lacking the ER-α isoforms, we follow published studies [32,33] and employ the human embryonic kidney cell line HEK293. To study single-cell ER-α protein changes, we treat cells with either E2 or TAM. Like 4-hydroxytamoxifen, TAM is a nonsteroidal antiestrogen that binds to ER at a low affinity of dissociation constant at 4.8 nM and inhibits cell growth at 10 μM [34,35]. Following the ligand treatment, we investigate BCa subpopulations based on the hormone response. The protein target multiplexing and isoform specificity offered by single-cell immunoblotting is used to gain understanding of the predictive potential of ER-α isoforms in heterogeneous BCa cells.

Primary tissue dissociation
Primary human tissues, which were slowly frozen in fetal bovine serum (FBS) with 10% dimethyl sulfoxide. Our Institutional Review Board deemed the study to be "not human subjects research", owing to the authors' use of Stanford Tissue Bank tissues that: existed before the research began, were not collected by the authors, and were de-identified prior to receipt by the authors. The authors did not collect potentially identifying genetic information. Tissue information is listed in S1 Table in S1 File. After quickly thawing, tissues were diced and incubated in a solution of collagenase type 3 (3000 unit/mL; 07423; Stemcell Technologies) and DNAse type 1 (D4263-5VL; 100 Kunitz unit/μl; Sigma-Aldrich) at 37˚C for 4 h. After digesting extracellular matrices, cell clumps were dissociated by a 40 μm cell strainer (352340; Corning). The dissociated cells were then resuspended with Hank's Balanced Salt Solution (14025076; Thermo Fisher Scientific) with 2% FBS.

Cell lines and cell culture
MCF-7, MDA-MB-231, HEK293 were obtained from the American Type Culture Collection (ATCC). HEK293 was cultured in Eagle's Minimum Essential Medium (EMEM) ; ATCC) supplemented with 1% penicillin streptomycin (PS) and 10% FBS. MCF-7 and MDA-MB-231 were maintained in RPMI 1640 (11875-093; Thermo Fisher Scientific), supplemented with 1% PS and 10% FBS. All cell lines were incubated in a humidified incubator held at 37˚C under 5% CO 2 . All cell lines were authenticated and free of mycoplasma using short tandem repeat analysis by UC Berkeley Cell Culture Facility. To limit sub-culturing effect, cell lines at low passage numbers (< 20) after thaw were only used for study.

Tamoxifen
Prior to ligand treatment, cells were incubated in phenol free RPMI1640 (11835030; Thermo Fisher Scientific) and charcoal stripped FBS (A3382101; Thermo Fisher Scientific) with 1% PS for 48 h. Like 4-hydroxytamoxifen, tamoxifen (TAM, T5648; Sigma-Aldrich) is a nonsteroidal antiestrogen that binds to ER and inhibits cell growth at a low affinity of dissociation constant at 4.8 nM [34,35]. Thus, similar to previous literature protocols [36,37], cells were treated with TAM with final concentration of 10 μM for 24 h. For negative control, cells were treated with 100% EtOH with equal volume as in the TAM treatment for 24 h. After the treatment, cells were detached from cell culture dish with 10 mM EDTA (AM9260G; Thermo Fisher Scientific) and proceeded with the single-cell immunoblotting.

Single-cell immunoblotting procedure
A single-cell immunoblot device is composed of a 30-μm thick polyacrylamide gel (8%T, 2.7% C) patterned with an array of 30-μm diameter microwells on a standard microscope glass slide. Starting with a suspension of cells at 25,000 cells/ml in 1x PBS (10010023; Thermo Fisher Scientific), gravitational sedimentation (10 min) populates microwells with cells, typically at 1 cell/microwell occupancy. After carefully washing the single-cell immunoblot with 1x PBS, more than 94% of microwells containing cells are occupied with single cell as determined by brightfield microscopy (S1 Fig in S1 File). Next, cells were lysed in situ for 30 s by pouring 15 ml of chemical lysis buffer at 37˚C. The chemical lysis buffer is comprised of 8 M Urea (U5378, Sigma Aldrich), 1% sodium dodecyl sulfate (SDS, L3771; Sigma Aldrich), 0.1% Triton X-100 (X100; Sigma Aldrich), 1x Tris-glycine (D6750; Sigma Aldrich). Following cell lysis, an electric field at 40 V/cm was applied across the single-cell immunoblot device, driving for protein polyacrylamide gel electrophoresis (PAGE) for 30 s. Immediately after PAGE, separated proteins were covalently bounded to the gel (via light-activated benzophenone) by applying UV (40 mW cm -2 , 45 s, Lightningcure LC5; Hamamatsu). Then, the single-cell immunoblot was washed with 1x TBS with Tween 20 (TBST, 77500; Affymetrix) for 1 h prior to immunoprobing. For immunoprobing, 0.1 g/l of primary and secondary antibodies were diluted with 1x TBST with 2% BSA and probed the device for 3 h and 2 h, respectively. After each probing step, 1x TBST was used for washing for 1 h. Lastly, the device was dried and scanned with a fluorescence microarray scanner (GenePix 4300A; Molecular Devices).

Single-cell immunoblotting data and statistical analyses
Images were processed by applying a median filter with a 2-pixel radius and a threshold value of 50 (ImageJ). Protein peaks from the single-cell immunoblot were quantitated with in-house MATLAB scripts [38]. The peaks were fitted by Gaussian functions in MATLAB (R2016b) and processed by extracting Gaussian parameters for peak width, location, and area-under-curve for protein expression. The protein peaks with Gaussian fitting R 2 � 0.65 and signal-to-noise ratio (SNR) > 3 were analyzed [38].
For statistical comparison of single-cell expression level, Mann-Whitney test was used. Kruskal-Wallis test with Dunn's multiple comparison test was used for > 2 mean comparison of the single-cell expression level. Unpaired t-test with Welch's correction was used to compare the cell subpopulation frequencies. The level of significance (p) is 0.05. For correlation studies, we used Spearman's correlation coefficients (ρ) with Dunn and Sidák correction and accounted correlations with the p value � 0.05.
Principal component analysis in MATLAB (2016b) is used for the multivariate analysis of protein expression levels from the single-cell immunoblotting. MATLAB's zscore function is applied to standardize the protein expression levels with a mean of 0 and a standard deviation of 1. MATLAB's pca function is used to compute the principal component coefficients, scores, and variances. The 95% confidence ellipses are calculated by eigenvalue decomposition with two standard deviations.

Results
Exclusive reliance on nuclear overexpression of full-length ER-α66 as an indicator for hormone therapy may be insufficient [39][40][41][42][43]. The roles of truncated ER-α isoforms and noncanonical ER-α mechanisms are also important. Consequently, we investigated 10 distinct ER signaling proteins, related to canonical and non-canonical ER signaling pathways, at singlecell resolution (Fig 1A). We develop a single-cell immunoblot to scrutinize ER signaling and isoforms in hormone-sensitive (MCF-7) BCa, hormone-insensitive (MDA-MB-231) BCa, and patient-derived dissociated ER-α 3+ BCa tumors (Fig 1B and 1C). As a model to detect clonal subpopulations with the single-cell immunoblot, MCF-7 was chosen as a cell line with high heteroclonality and genetic plasticity [25,26]. Of note, HEK293 was used as a control cell line that lacks ER-α isoforms (S2 Fig in S1 File) [32,33].
The single-cell immunoblot utilizes an open microfluidic device design (i.e., no enclosed microchannels or pneumatic control) to prepend single-cell polyacrylamide gel electrophoresis (PAGE) for size-based protein separation to an in-gel immunoassay (Fig 1B, S3 Fig in S1 File). As illustrated in Fig 1C, same-cell protein target multiplexing (up to 10 targets here) is achieved by immobilizing the separated proteins by UV, detecting with cocktails of compatible antibody probes, and thorough chemical stripping and re-probing of antibody probes for different protein targets [44].
To discern ER-α isoforms, we developed the single-cell immunoblotting by testing cell lysis conditions (SDS, urea) and several pan-ER-α antibodies in MCF-7, MDA-MB-231, and HEK293 cells (S4 Fig in S1 File). As corroborated by previous literature [27] and conventional assays (S2 Fig in S1 File), the SP-1 antibody identified ER-α isoforms without non-specific background signals in the single-cell immunoblot. After confirming molecular sizing with housekeeping proteins in in slab-gel and single-cell immunoblots (S2, S4 Figs in S1 File), we chose the monoclonal SP-1 antibody-widely used in cell lines [27,28], mouse models [29], patient ER-α status [30,31]-to investigate the frequency and expression levels of ER-α isoforms in BCa cell lines.

MCF-7 and MDA-MB-231 as BCa models for ER signaling
Before we use the single-cell immunobloting to measure drug response in cell lines, we sought to understand how well hormone-sensitive (MCF-7) and hormone-insensitive (MDA-MB-

Fig 1. Single-cell immunoblotting resolves ER signaling proteins and ER-α isoforms each BCa cell. (A)
Schematic of ER signaling, which includes crosstalk between canonical and non-canonical pathways. Network nodes indicate ER signaling proteins scrutinized. Canonical nodes in blue; non-canonical nodes in red. Edges represent protein-protein associations, arrows indicate activation, bars indicate inhibition, and circles represent an unspecified interaction. (B) Image and workflow of a single-cell immunoblotting device, which is a glass microscope slide layered with polyacrylamide gel (30-μm thick), stippled with an array of microwells (30-μm diameter). Schematic of assay workflow: after cells are settled into the microwells, each cell is chemically lysed (30 s) and the resultant single-cell lysate is separated by PAGE (40 V/cm, 30 s). Next, protein targets are photo-immobilized in the gel and interrogated with a sequence of antibody probes. (C) False-color fluorescence micrographs of single-cell immunoblots report both canonical and non-canonical ER signaling pathway targets in two relevant cell lines and cells dissociated from ER-α 3+ BCa tumor (4318-1). Log-linear plots report molecular mass versus protein peak location and confirm protein target identity. Error bars represent the variance in protein target peak location, from a set of 10 single-cell immunoblots. Coefficient of determination for log-linear regression is R 2 MCF-7, MDA-MB-231,BCa_Tissue = 0.9. https://doi.org/10.1371/journal.pone.0254783.g001
In another case, the ER-α 3+ BCa tumor (4318-1) had the highest mean ER-α46 expression   ; Fig 2A). Subsequently, we sought to compare variance in ER signaling between the ER-α + BCa tumors and the two cell lines. For this investigation, we performed dimensional reduction on the multivariate analysis of the ER signaling protein levels (ER-α66, ER-α46, CD44, Cyclin A, p38 MAPK, pAKT, pS6). We tested whether the ER-α + BCa tumors and the cell lines differ in the ER-signaling target expression level by carrying out principal component analysis (PCA). The first and second principal components (PC) explain the major variance (48.6%) of the ERsignaling target expression level (Fig 2B). Using the first and second principal components (PC), we investigated mean scores with confidence ellipses (Fig 2B). The convergence of the 95% confidence ellipses between the ER-α + BCa tumors and MCF-7 in the PC1 and PC2 score plot explains similarity between the ER-α + breast tumors and MCF-in the variance of ER-signaling target expression level (Fig 2B). Conversely, we observe the 95% confidence ellipse of MDA-MB-231 diverging from the confidence ellipses of MCF-7 and ER-α + breast tumors at PC1 > 2. As the PC1 score increases, positive correlations are found with ER-α46 (ρ ER-α46 = 0.30), p38 MAPK (ρ p38_MAPK = 0.55), pAKT (ρ pAKT = 0.45), and pS6 (ρ pS6 = 0.48) markers (Fig 2C). For the PC2, ER-α66 (ρ ER-α66 = 0.69) and pAKT (ρ pAKT = 0.51) are two dominant correlation coefficients. Consequently, the confidence ellipses in PCA of ER-signaling target expression level indicate similar variance in the ER-signaling pathway between the ER-α + BCa tumors and the cell lines.

Heterogeneous ER-α66 and ER-α46 response to TAM in hormone-sensitive (MCF-7) BCa
We next sought to understand how TAM affects ER-α66 and ER-α46 frequencies and mean expression levels. At the single-cell level, we hypothesize that TAM would reduce the ER-α66 + subpopulation. Since the TAM effect on ER-α46 is not known, we examined the ER-α46 expression level by using the single-cell immunoblots with TAM-treated cells.

Rare subpopulation of hormone-insensitive BCa cells expresses ER-α46
After characterizing ER-α isoforms in hormone-sensitive MCF-7 cells, we sought to understand the heterogeneity of ER-α46 expression level in a triple-negative BCa cell line, MDA-MB-231. The MDA-MB-231 cells lack the full-length ER-α66 protein and exhibit highly invasive phenotypes [55]. As expected, we did not detect ER-α66 in individual MDA-MB-231 cells using single-cell immunoblotting (Fig 4A).
Two cell subpopulations were identified: MDA Cell Type 1 with ER-α46 + and MDA Cell Type 2 with ER-α46 - (Fig 4A). The MDA Cell Type 1 subpopulation accounts for 4.2%  Fig 4C). The variance of the ER-α46 expression level was lower in the MDA-MB-231 cells (CV = 77.7%) than in the MCF-7 cells (CV = 105.3%), indicating less cellto-cell variation in the ER-α46 expression level within the population of MDA-MB-231 cells (Fig 4C).

PLOS ONE
Single-cell immunoblotting resolves estrogen receptor-α isoforms in breast cancer

ER signaling proteins are highly correlated in TAM-treated hormonesensitive and hormone-insensitive BCa cells
To evaluate whether TAM affects both canonical and non-canonical ER actions [57], we assessed associations between ER-α isoforms and ER signaling proteins in hormone-sensitive MCF-7 cells and hormone-insensitive MDA-MB-231 cells. We sought to investigate the canonical ER signaling response by measuring EGFR, p38 MAPK, and phospho-AKT (pAKT, phosphorylation at Ser473) protein targets, which are translated from genes enriched with ERE [45,46], while CD44, pS6, and Cyclin A for non-canonical ER signaling.
At a basal level, the full-length ER-α66 protein is associated with the ER-α46 protein (ρ = 0.96) and the ER-β protein (ρ = 0.52, Fig 5A). With TAM treatment, while we did not observe significant changes in protein expression of ER-signaling targets with the exception of pAKT (S2 Table in Fig 5A). In contrast, ER-α46 is less correlated (< 0.4) with any ER signaling targets in the TAM-treated group (Fig 5A). Taken together, in hormone-sensitive MCF-7, TAM reduces ER-α66 isoform expression but activates both canonical and non-canonical pathway (Fig 5A).

pAKT is a key regulator of TAM sensitivity in the ER signaling pathway
We further investigated the relationship between ER-α isoform and ER signaling proteins by analyzing expression levels in each subpopulation. Since pAKT interacts both upstream and downstream in ER signaling pathways [58], we hypothesized that pAKT and ER-α isoforms would influence each other in TAM-treated hormone-sensitive BCa (MCF-7). In the ER-α66 +
Next, we sought to understand the relationship between ER-α46 and pAKT. We clustered the ER-α46 + \ pAKT + MCF-7 cell subpopulation and measured responses to TAM treatment. Unlike the pAKT expression level at the population level (S6 Fig, S2 Table in S1 File), we did not observe repression in the mean pAKT expression level within the ER-α46 + \ pAKT + subpopulation upon TAM treatment (Fig 6D, S3 Table in S1 File). We observed that the pAKT expression level is higher in the MCF Cell Type 1 (ER-α66 + \ ER-α46 + ) than in the MCF Cell Type 2 (ER-α66 + \ ER-α46 -) subpopulation ( Fig 6F); however, the difference in pAKT expression is attributable to a greater decrease in ER-α66 + in the MCF Cell Type 2 (vs. MCF Cell Type 1) subpopulation and may not be associated with ER-α46 expression. Taken together, we suspect that the non-canonical ER signaling action of TAM is linked with the ER-α66 isoform (and not the ER-α46 isoform) via the pAKT signaling pathway in MCF-7 hormone sensitive cells (Fig 6G).

p38 MAPK is associated with ER-α66 upon TAM treatment
In addition to the PI3K/AKT/mTOR pathway, we sought to scrutinize the interaction between ER isoforms and p38 MAPK in hormone-sensitive BCa cells (MCF-7). Given the significant changes in correlation between p38 MAPK and ER-α66 with and without TAM treatment (ρ = -0.27 in Neg to ρ = 0.83 in TAM, Fig 5A), we hypothesized that TAM affects the p38 MAPK expression level in an ER-α66 dependent manner. We did not observe significant changes in p38 MAPK expression at the population level (S2 Table in S1 File), but we did observe that TAM increased the mean p38 MAPK expression level by 17% in the ER-α66 + \ p38 MAPK + subpopulation (Fig 7A, S3 Table in S1 File). The upregulation of p38 MAPK is associated with the presence of the ER-α66 protein: mean p38 MAPK expression is 31% higher in the ER-α66 + \ p38 MAPK + subpopulation, as compared to the ER-α66 -\ p38 MAPK + subpopulation (Fig 7B, S4 Table in S1 File). TAM significantly decreased the mean ER-α66 expression level (77% decrease, Fig 7A, S3 Table in S1 File). The subpopulation analysis suggests a unidirectional relationship between ER-α66 and p38 MAPK, in which TAM affects ER-α66 to alter p38 MAPK pathway (Fig 7G).

pS6 upregulates ER-α isoforms via the non-canonical ER signaling pathway
Next, we sought to scrutinize the interaction of pS6 (an indicator of activity in the PI3K/ pAKT/mTOR signaling pathway) with ER-α isoforms in the hormone-sensitive MCF-7 cells [59]. Unlike pAKT, we did not observe perturbation of pS6 expression in either the whole population (S6 Fig, S2 Table in S1 File), the ER-α66 + subpopulation (Fig 7C), or the ER-α46 + subpopulations (Fig 7D, S3 Table in S1 File) upon TAM treatment. Instead, TAM affected ER-α isoforms in the pS6 + subpopulations. TAM significantly altered the mean ER-α66 expression level in the ER-α66 + \ pS6 + subpopulation and the mean ER-α46 expression level in the ER-α46 + \ pS6 + subpopulation (Fig 7C and 7D, S3 Table in S1 File). In order to understand if ER-α isoform responses are linked to the presence of pS6, we compared the pS6 + and the pS6subpopulations (Fig 7E and 7F). Interestingly, we discovered that pS6 appears to mitigate TAM repression of ER-α66: the mean ER-α66 expression in the pS6 + subpopulation is 50% greater than the ER-α66 expression in the pS6subpopulation (Fig 7E, S4 Table in S1  File). Similarly, we observed that the mean ER-α46 expression is greater in the pS6 + subpopulation (Fig 7F, S4 Table in S1 File). Taken together, the single-cell protein analysis suggests that pS6 upregulates both ER-α66 and ER-α46 (Fig 7G).

pAKT is associated with ER-α46 upon TAM treatment in MDA-MB-231
In hormone-insensitive cancer, TAM induces apoptosis by inhibiting pAKT in a dose independent pathway [60]. We examined whether TAM modulates the expression level of pAKT via the ER-α46 associated non-canonical ER signaling pathway (Fig 8). In contrast to MCF-7 cells, we observed no change in the mean expression of pAKT with or without TAM treatment. The observation was the same in both the ER-α46 + \ pAKT + subpopulation and the overall population (Fig 8A, S6 Fig, S2 Table in S1 File). Interestingly, we observed that the presence of pAKT leads to a greater mean ER-α46 expression level under TAM (Fig 8B, S4 Table in S1 File). Spearman's correlation suggested median correlation between ER-α46 and pAKT (ρ = 0.61, Fig 5B). Accordingly, we suspect a strong interaction between ER-α46 and pAKT in hormone-insensitive MDA-MB-231 cells upon TAM treatment (Fig 8C).

CD44 is the downstream protein target in the non-canonical ER signaling pathway
We next sought to scrutinize the non-canonical ER signaling pathway in the hormone-insensitive MDA-MB-231 cells. While we did not observe mean CD44 expression level changes upon TAM treatment at the population level (S2 Table in S1 File), we did observe that TAM increases the mean CD44 expression level by 60% in the ER-α46 + population (Fig 8D, S3 Table in S1 File). Further, the mean CD44 expression level in the ER-α46 + \ CD44 + subpopulation is 53% greater than the mean CD44 expression level in the ER-α46 -\ CD44 + subpopulation (Fig 8E, S4 Table in S1 File). On the other hand, the CD44 + subpopulation did not see ER-α46 influenced by TAM (Fig 8D, S3 Table in S1 File). Taken together, ER-α46 induces CD44, while CD44 does not appear to regulate ER-α46 (Fig 8F).

Principal component analysis suggests dominant ER signaling targets in BCa cell lines
After detecting interactions between the ER-α isoforms and ER signaling targets, we applied PCA with K-means clustering to distinguish the BCa subpopulations responding to TAM.

PLOS ONE
Single-cell immunoblotting resolves estrogen receptor-α isoforms in breast cancer

Discussion
Here, single-cell immunoblotting reports ER-α isoform heterogeneity in both hormone-sensitive and hormone-insensitive BCa. We classify and compare ER-α isoform expression among cell subpopulations in cell lines to understand the BCa cell lines as ER-signaling models. With PCA explaining 48.6% of the variance of 7 ER-signaling targets, we find similar ER signaling expression levels between the cell lines and the tissue specimens (Fig 2B) [62,63].
In bulk assays, stimulation of BCa cells with TAM is known to modulate ER-α66 mRNA [51,53,64]. By single-cell immunoblot, we observed a decrease in ER-α66 protein expression after TAM treatment (Fig 3D). While TAM is understood to stabilize ER-α66 protein expression [65,66], these studies use different tamoxifen metabolites (i.e. 4-hydroxytamoxifen), and concentrations (i.e., < 1 μM). In contrast, TAM at > 1 μM was reported to degrade ER-α66 by the proteasome [67]. The effect of 1 > μM non-metabolized TAM on ER-α66 protein expression has not been investigated previously. Measuring ER-α isoforms at a single cell level with various concentrations of ER modulators from derived and different ER model cell lines would provide more insight on pharmacological research.
In hormone-sensitive MCF-7, the ER-α66 protein expression level is lower in the same cells expressing ER-α46 (Fig 3C), mirroring with the repressive estrogenic activity of ER-α66 by ER-α46 at the transcription level as reported [15,17]. However, the positive correlation between ER-α66 and ER-α46 under no treatment and the loss of the correlation under TAM suggest that ER-α46 may not directly inhibit ER-α66 (Fig 5A). Indeed, in the MCF-7 subpopulation expressing both ER-α66 and ER-α46, we did not observe a further decrease of mean ER-α66 expression under TAM (Fig 3F).
Although a few studies have reported that ER-α46 inhibits cell growth in the presence of the ER modulator (TAM/E2) [27,42], the TAM effect on ER-α46 expression has not been investigated in BCa models. The truncated ER-α46 possesses the ligand binding domain that interacts with TAM/E2 [68]. In human macrophages, the E2 treatment increases ER-α46 transcription by inducing the promoter F of the ER-α gene (ESR1) [54]. Thus, one possible mechanism is a change in promoter activity in which E2 or TAM mediates alternative splicing to generate ER-α46 [42,54]. Although ESR1 recruitment in a promoter F region is known to increase the ER-α46 expression level in the MCF-7 cells [69], the TAM regulation of the ESR1 promoter activity is unknown. Our data indicate that TAM/E2 treatment increases ER-α46 expression level in the MCF-7 cells (Fig 3D and 3F, S5A Fig in S1 File). Because the TAM effect in the ligand-dependent AF-1 domain varies with specific cell and promoter types, further examination of transcription and translation is needed to understand TAM mechanisms at the level of individual cells.
Single-cell multiplexing of ER signaling provides detailed examination of cell-to-cell variation in canonical and non-canonical ER-α actions (Fig 1A). Looking at subpopulations expressing specific ER signaling proteins, pair-wise comparison of ER-α isoforms and ER signaling proteins shows TAM enhancing the correlation between ER-α isoforms and ER signaling proteins (Fig 5). Strong ER-α isoform correlations with pS6, CD44, and Cyclin A in TAM treated cells implicates the non-canonical ER signaling pathway (Fig 5).
With a focus on validation and application of precision single-cell protein measurement tools, we scrutinize unmodified, endogenous protein isoforms in signaling pathways using two model BCa cell lines. We verify the isoform selectivity, analytical sensitivity, throughput, and monitoring of response to drug treatment. Looking forward, the integration of single-cell immunoblotting and gene knockout/overexpression of other BCa cell lines would offer a deep dive into the signaling cascades [49,80,81]. Further, subcellular analysis of ER isoforms and signaling proteins would tease apart the role of membrane and nuclear protein forms to boost understanding of membrane-bound ER-α [16,17,82]. Given the importance of truncated oncoprotein isoforms in the development of drug resistance and as potential therapeutic targets, high-selectivity and multiplexed cytometry tools-such as that described here-are a critical component for advancing personalized therapies to benefit each individual patient.