Limiting Self-Renewal of the Basal Compartment by PKA Activation Induces Differentiation and Alters the Evolution of Mammary Tumors

Summary

Differentiation therapy utilizes our understanding of the hierarchy of cellular systems to pharmacologically induce a shift towards terminal commitment. While this approach has been a paradigm in treating certain hematological malignancies, efforts to translate this success to solid tumors have met with limited success. Mammary-specific activation of PKA in mouse models leads to aberrant differentiation and diminished self-renewing potential of the basal compartment, which harbors mammary repopulating cells. PKA activation results in tumors that are more benign, exhibiting reduced metastatic propensity, loss of tumor-initiating potential and increased sensitivity to chemotherapy. Analysis of tumor histopathology revealed features of overt differentiation with papillary characteristics. Longitudinal single cell profiling at the hyperplasia and tumor stages uncovered an altered path of tumor evolution whereby PKA curtails the emergence of aggressive subpopulations. Acting through the repression of SOX4, PKA activation promotes tumor differentiation and represents a possible adjuvant to chemotherapy for certain breast cancers.

eTOC Blurb

The mammary basal compartment harbors the majority of repopulating cells in gland. Ognjenovic et al. demonstrate that activation of PKA signaling inhibits the self-renewing potential of these basal cells, leading to aberrant mammary development. Functioning through the inactivation of Sox4, PKA curtails the emergence of multiple aggressive tumor subpopulations.

Introduction

The process of differentiation involves a complex multi-step journey of cells from a state of multipotency towards one of commitment to a cell state that is, typically, post-mitotic. The process of tumorigenesis involves the acquisition of limitless proliferative capacity, the evasion of growth suppression, and the enabling of replicative immortality (Hanahan and Weinberg, 2011), all of which are characteristics of the circumvention of terminal differentiation. Differentiation therapy was first championed in hematological malignancies in which the use of all-trans-retinoic acid led to the terminal differentiation of acute promyelocytic leukemias (APL) (Huang et al., 1988; Warrell et al., 1993). Despite these advances, barring a few notable examples (Saha et al., 2014; Storm et al., 2016), differentiation therapy in solid tumors has largely been either unsuccessful or exploratory.

3',5'-cyclic adenosine monophosphate (cAMP) is a widely used second messenger that regulates multiple downstream signaling cascades upon G-protein coupled receptor activation and subsequent Gα_s or Gα_i-mediated regulation of adenylate cyclase (Tasken and Aandahl, 2004). One of the major routes of cAMP action is via activation of the ubiquitous enzyme Protein Kinase A (PKA). While in certain contexts, mutant Gα_s acts in an oncogenic fashion (Coles et al., 2020; Patra et al., 2018; Wu et al., 2011), multiple lines of evidence suggest a role for Gα_s-cAMP-PKA signaling in tumor-suppressive properties that could be mediated through the induction of differentiation (He et al., 2014; Iglesias-Bartolome et al., 2015; Pattabiraman et al., 2016).

The presence of heterogeneous tumor subpopulations has been a major hurdle in the clinical management of various cancer types. Indeed, recent studies have pointed to specific epigenetic mechanisms by which alterations in lineage restriction can contribute to tumor heterogeneity and progression of disease (Flavahan et al., 2017; LaFave et al., 2020; Pomerantz et al., 2020). Breast cancers are a heterogeneous group of diseases that have been classified based on their histopathological features, or molecularly, based on their gene expression profiles (Perou et al., 2000). These modes of classification have been useful in highlighting the cellular and molecular composition of tumors, while also providing prognostic information that can be leveraged for therapeutic decision-making. While these analyses have been instrumental in our ability to comprehend inter-tumoral heterogeneity by binning tumors into subtypes, they have not been able to shed light on the specific subpopulations within tumors that drive tumor growth, their differentiation state, and the relationship between the composition of cells in a tumor and the histopathological features that they manifest.

Here we show that activation of PKA impairs mammary differentiation by specifically reducing the self-renewal potential of the basal compartment. Mammary tumors with constitutively active PKA exhibit features of papillary differentiation with a reduced propensity for metastasis and increased susceptibility to chemotherapy. Longitudinal single cell profiling of mammary tumors at early and late stages of tumorigenesis revealed their inherent heterogeneity through the emergence of multiple, transcriptionally-distinct, subpopulations. Firstly, the emergence of a lumino-basal subpopulation that co-expresses both luminal and basal markers in MMTV-PyMT tumors, marking the relinquishing of their developmental lineage-of-origin. Additionally, the emergence of an EMT-like sub-population that signals loss of key epithelial markers.

Both these subpopulations are severely diminished in PKA-activated tumors, which exhibit reduced metastatic capacity and increased therapeutic sensitivity. These observations imply that the induction of a more differentiated cell state may favorably alter the representation of tumor cell subpopulations that are responsible for progression. Our single-cell analyses revealed Sox4 as a potential candidate in the regulation of the lumino-basal subpopulation, whereby its direct phosphorylation by PKA at S56 and S103 impair its ability to act as a transcription factor and promote tumor progression.

Results

Constitutively Active PKA Signaling Impairs Mammary Differentiation

Previous studies have shown the ability of PKA to induce MET in some breast cancer cell lines (Pattabiraman et al., 2016). These data do not, however, reveal the biological effects of PKA activation in normal mammary development and its impact on the subsequent in vivo development of tumors. To address this, we employed a mouse model harboring a floxed constitutively active mutant allele of the catalytic subunit of PKA (Niswender et al., 2005), and expressed this allele in the mouse mammary epithelium by crossing it to a mouse mammary tumor virus-driven Cre (MMTV-Cre; line D) model (Wagner et al., 2001)(Fig. 1A). The mammary glands of these Prkaca^CαR/fl mice at eight weeks consistently contained fewer epithelial cells than littermate Prkaca^fl/fl controls, with a reduction of at least 50% in the EpCAM^medCD49f^hi basal/myoepithelial cells (Fig. 1B,C), the compartment that harbors the highest proportion of mammary repopulating units. This reduction was not a result of differences in Cre recombination between the luminal and basal lineages as 70-85% of cells from both compartments underwent recombination when crossed with a R26-mTmG reporter (Fig S1A,B). Postnatal mammary gland development in the Prkaca^CαR/fl mice was aberrant displaying a severe defect in ductal morphogenesis (Fig. 1D). Histological analyses of the mammary ducts revealed intact luminal and basal layers (Fig 1E), however, the overall number of terminal ductal lobular units (TDLUs) per gland was 3-fold lower in the Prkaca^CαR/fl mice compared to Prkaca^fl/fl controls (Fig 1F,S1C), likely a result of impaired ductal outgrowth. In order to test whether the decrease in basal cells was merely quantitative, or whether activation of PKA led to a quantitative alteration of their self-renewal ability, we sorted for EpCAM^medCD49f^hi basal cells from Prkaca^CαR/fl mice and control Prkaca^fl/fl littermates, and carried out organoid-forming assays (Guo et al., 2012). EpCAM^medCD49f^hi cells from Prkaca^CαR/fl mice were impaired in their organoid-forming potential, indicating a loss of self-renewing potential (Fig. 1G,H). Additionally, when transplanted into the cleared fat pads of recipient mice, organoids from Prkaca^CαR/fl mice were at least 100-fold less efficient at forming ductal trees (Fig. 1I), revealing a loss of stem-like repopulating ability. This lack of efficiency in repopulation was also observed upon transplantation of sorted EpCAM^medCD49f^hi basal cells (Fig S1D). Notably, the basal cells from Prkaca^CαR/fl mice were severely deficient in serial transplantation assays (Fig S1D), suggesting that the activation of PKA leads to the exhaustion of the repopulating ability of this compartment. This phenotype is reminiscent of previous observations whereby activation of Gα_s-PKA signaling led to stem cell exhaustion and depletion (Iglesias-Bartolome et al., 2015).

Figure 1: Activation of PKA impairs mammary development and repopulating ability.

(A) PKA-CαR mice were crossed to MMTV-Cre mice to activate the CαR constitutively active allele. (B) FACS plots showing proportions of EpCAM^hiCD49f^med luminal and EpCAM^medCD49f^hi basal compartments upon constitutive activation of PKA and a summary of relative proportions of these subpopulations across multiple mice (C). Carmine alum stained whole mounts of mammary glands from and Prkaca^CαR/fl control mice displaying ductal outgrowth (D). Bar, 3mm. (E) Fluorescence imaging of Krt8 (luminal) and Krt14 (basal) layers in mammary glands from Prkaca^CαR/fl and Prkaca^fl/fl mice. Bar, 150μm. (F) Hematoxylin-eosin staining of FFPE sections of mammary glands depicting terminal ductal lobular units from Prkaca^CαR/fl and Prkaca^fl/fl mice. Bar, 150μm. Organoid assays were carried out to assay for differences in self-renewal potential upon activation of PKA (G, H). Data are shown as mean ± SD (n =10 mice). p value was determined by Student’s two-tailed t test (unpaired). *p < 0.01. Limiting dilution transplantation into cleared fat pads estimated the frequency of mammary repopulating units in Prkaca^CαR/fl and Prkaca^fl/fl controls (I). Schematic representation of breeding strategy to generate mammary-specific Gα_s active mutants (J). FACS was carried out to capture differences in representation of luminal and basal subpopulations (K). Organoid assays carried out to test for differences in self-renewal potential (L) (n = 5 mice). Statistical significance was calculated by a Student t-test (two-tailed) to compare two groups (P < 0.05 was considered significant).

In order to ensure that the effects observed could be attributed specifically to activation of PKA signaling, we used a second model, Gα_s^R201C/fl mice (Iglesias-Bartolome et al., 2015), that express constitutively-active Gα_s under the control of a doxycycline-inducible promoter. Upon expressing this allele in the mammary gland (Fig. 1J), doxycycline treatment was carried out to activate the Gα_sR201C mutant from 3 weeks post-partum, at which point postnatal mammary development is initiated, to full maturity at 8 weeks. As observed in the Prkaca^CαR/fl mice, the Gα_s^R201C/fl mice also exhibited a reduction in the EpCAM^medCD49f^hi basal cells, which were severely impaired in their organoid-forming potential (Fig. 1K,L) while also displaying severe defects in mammary ductal outgrowth (Fig. S1E). Importantly, organoid-forming potential of basal cells from control MMTV-Cre mice was unaffected by doxycycline treatment (Fig. S1F). Put together, these results indicate a role for activated PKA signaling in limiting the self-renewal and expansion of the mammary basal lineage.

Basal lineage-specific impairment upon activation of PKA

To further confirm the lineage-specific effects of activation of PKA, the CαR allele was expressed specifically in either luminal or basal lineages of the mammary gland. We crossed the Prkaca^fl/fl mice to a R26-mTmG reporter mouse (Muzumdar et al., 2007), and F1 mice were subsequently crossed to a Krt5-CreERT2 mouse to express the CαR allele in a basal-specific manner (Van Keymeulen et al., 2011)(Fig. 2A). Upon tamoxifen administration at 4 weeks of age and observation of mammary glands at 8 weeks, we noted an absence of GFP-positive basal cells in the Prkaca^CαR/fl mice, present only in controls that lacked the CαR allele, or those that did not receive tamoxifen (Fig. 2B-D). Initial recombination following administration of tamoxifen led to 76% of basal cells undergoing Krt5-Cre recombination (Fig S2), whereas in 8-week old Prkaca^CαR/fl mice, less than 1% of basal cells were GFP positive, compared to >30% in Prkaca^fl/fl controls (Fig 2B). These observations suggest that the expression of the Prkaca^CαR allele inhibited the expansion of the basal compartment, leading to the observed defects in ductal outgrowth. We subsequently expressed the CαR allele specifically in luminal cells using the Krt8-CreERT2 mouse (Van Keymeulen et al., 2011)(Fig 2E). Upon tamoxifen administration at 5 weeks of age and observation of mammary glands at 8 weeks, the percentage of GFP-expressing luminal cells in Prkaca^CαR/fl mice was comparable to those of Prkaca^fl/fl controls (Fig 2F), and indeed, the presence of GFP+ ducts was observed upon whole mount imaging and staining of sections (Fig 2G,H), indicating that activation of the PKA allele did not affect the luminal lineage. These results reiterate the ability of PKA signaling in inducing a qualitative and quantitative alteration of the mammary basal/myoepithelial compartment. We further explored the impact of these alterations on the development and progression of mammary tumors that arise from these aberrant glands.

Figure 2: Specific activation of PKA in basal, but not luminal cells, impairs ductal outgrowth.

Schematic of breeding strategy to generate Krt5-Prkaca^CαR/fl mice that specifically express activated PKA in basal cells (A). FACS plots (B) and fluorescence imaging of ducts (C) show the extent of GFP-expressing cells within the basal compartment and ducts that have undergone Cre recombination in Krt5-Prkaca^CαR/fl mice and Krt5-Prkaca^fl/fl controls at 8 weeks. Bar, 500μm (D) Fluorescence imaging of a cross-section of a mammary duct of Krt5-mTmG-Prkaca^CαR/fl mice to assess histology and extent of tdTomato vs GFP recombination. Bar, 150μm. Schematic of breeding strategy to generate Krt8-Prkaca^CαR/fl mice that specifically express activated PKA in luminal cells (E). FACS plots (F) and fluorescence imaging of ducts (G) show the extent of GFP-expressing that have undergone Cre recombination in Krt8-PrkacaCαR/fl mice and Krt8-Prkaca^fl/fl controls at 8 weeks. Bar, 500μm. (H) Fluorescence imaging of a cross-section of mammary ducts of Krt8-mTmG-Prkaca^CαR/fl mice to assess histology and extent of tdTomato vs GFP recombination. Bar, 150μm.

Human Breast Cancers Harbor Recurrent Genomic Alterations in the PKA locus

Given the previously identified role for PKA in the induction of mesenchymal-epithelial transition (MET)(Pattabiraman et al., 2016), we interrogated the clinical significance of the PKA genes in human breast cancer. Analysis of the breast cancer genomic data from The Cancer Genome Atlas (TCGA)(Cancer Genome Atlas, 2012) revealed amplifications in the genomic loci encoding PKA subunits in 11.2% (121/1085) of breast cancers (Fig. 3A). PRKAR1A encodes one of four negative regulatory subunits that sequester the catalytic subunits, including that encoded by PRKACA, which carry out enzymatic functions downstream (Skalhegg and Tasken, 2000). The frequency of amplifications in the other regulatory subunits PRKAR1B, PRKAR2A and PRKAR2B were significantly lower than that of PRKAR1A at 1.1%, 0.1% and 0.6%, respectively (Fig 3B). We, thus, focused on the effects of PRKAR1A amplifications and resulting PRKACA catalytic subunit activity for further interrogation. Both of these subunits were also the highest to be expressed across breast tumors (Fig 3C). Mining the METABRIC datasets (Curtis et al., 2012; Pereira et al., 2016) also revealed PRKAR1A amplifications in 11.9% (82/684) of tumors, with amplifications being more frequent in the Luminal B subtype of tumors across both TCGA (13.03%) and METABRIC (14.32%) datasets (Fig. 3D). PRKAR1A amplification was correlated with higher expression levels of its mRNA (Fig. 3E), while also inversely correlating with expression levels of the catalytic subunit PRKACA (Fig. S3A,B) in a subtype-independent fashion. Higher PRKAR1A expression conferred better prognosis in tumors of the basal-like (Fig. 3F), but not other subtypes (Fig S3C). These observations were, however, only observed in TCGA, but not other, datasets (Saal et al., 2015)(Fig. S3D,E), or for other or for other PKA subunits (Fig 3G).

Figure 3: Recurrent genomic amplification of the PKA locus in human breast cancer.

Analysis of 1085 breast tumors from the TCGA dataset revealing amplifications in the regulatory (PRKAR) and catalytic (PRKAC) genes that encode PKA subunits (A, B). (C) Box plots show the mRNA expression levels of genes encoding each PKA subunit and their relative expression levels across human breast cancer subtypes. Summary of PRKAR1A and PRKACA amplifications in the TCGA and METABRIC datasets in basal and luminal B subtypes (D). mRNA expression of PRKAR1A, the most amplified PRKA gene, across tumors with and without amplifications in the PRKAR1A locus (E). Kaplan-Meier curves outline the ability of high or low levels of PRKAR1A (F) or PRKACA (G) mRNA levels to stratify prognosis of patients over ten years. PRKACA activation was measured by immunohistochemistry (IHC) using a p-PKA substrate antibody and quantifying the percentage of cells with nuclear signal in primary breast cancer patient samples that harbor amplifications in PRKAR1A compared to controls (H). Statistical significance was calculated by a Student t-test (two-tailed) to compare two groups (P < 0.05 was considered significant) (n=35 patient tumors). IHC using p-PKA substrate antibody showing samples that express different levels of PRKACA activation (I). Bar, 25μm.

These recurrent genomic alterations in the PKA loci point to possible clinical implications for breast tumorigenesis, particularly of the luminal B subtype. We explored the direct connection between PRKAR1A amplification in human breast tumors and the corresponding activation of the PRKACA catalytic subunit signaling. In line with a prognostic role for the PKA locus, tumors that harbored amplifications in the PRKAR1A locus had lower percentages of cells that exhibited PKA activation compared to controls that harbored no amplifications (Fig 3H,I). This result indicates that genomic amplifications in the PRKAR1A locus play a direct role in regulating the activation status of PKA catalytic subunit signaling in human breast cancer. We, thus, proceeded to further study the role of PKA catalytic subunit activation and its downstream effectors in mammary tumorigenesis.

PKA-induced Differentiation leads to Attenuation of Metastatic Progression and Improved Response to Therapy

Given the high rate of recurrent PRKAR1A and PRKACA amplifications in the luminal B and basal subtypes of breast cancer (Fig. 3D), we chose to model PKA activation in the MMTV-PyMT mouse model of mammary tumorigenesis (Guy et al., 1992), which exhibit luminal B features in the early stages of tumor development, but acquire a more basal-like phenotype as tumors progress (Lin et al., 2003). Upon crossing the Prkaca^CαR/fl mice to the MMTV-PyMT mice (Fig. 4A), we observed that tumors in the control Prkaca^fl/fl PyMT mice took, on average, 142 days to reach the tumor volume endpoint of 2cm², whereas those in the Prkaca^CαR/fl PyMT were slower to develop, taking 173 days (Fig. 4B), although overall tumor volume was comparable between the two genotypes (Fig. 4C). As expected, activation of the PKA signaling pathway was higher in the Prkaca^CαR/fl PyMT tumors compared to the Prkaca^fl/fl PyMT controls (Fig. S4A,B).

Figure 4: Induction of tumor differentiation upon activation of PKA signaling.

(A) Schematic of breeding strategy to generate MMTV-PyMT mice with active PKA. Differences in survival (B; n = 24 mice) and primary tumor volumes (C; n = 11 mice) of Prkaca^CαR/fl PyMT mice compared to Prkaca^fl/fl PyMT controls. Differences in lung metastatic burden was captured from H&E stained FFPE lung sections (D, E), red arrows indicate macrometastases. Bar, 2mm (n = 13 mice). Treatment with Adriamycin showing differences in response to chemotherapy (F, G) between Prkaca^CαR/fl PyMT mice and Prkaca^fl/fl PyMT controls (n = 20 tumors). Tumor initiating potential of Prkaca^CαR/fl PyMT mice and Prkaca^fl/fl PyMT controls was assessed by limiting dilution transplantation analyses (H). Fluorescence imaging of tumor sections stained with antibodies against E-cadherin, Vimentin, Krt8 and Krt14 revealing epithelial-mesenchymal (I,J) and luminal-basal heterogeneity (K,L). Yellow arrows highlight Vimentin (I) - and Krt14 (K)-expressing tumor cells. Bar, 10μm and 4μm. Error bars represent ± standard deviations of the mean. Statistical significance was calculated by a Student t-test (two-tailed) to compare two groups (P < 0.05 was considered significant) except for survival analyses where significance was determined using a Log-rank (Mantel-Cox) test (P < 0.05 was considered significant) and chemotherapy treatment where a Wald Z-test was used to compute the p-value for the difference of slopes in two treatment groups.

In order to quantify the metastatic burden of the Prkaca^CαR/fl PyMT, we enumerated the number of lung macrometastatic foci from H&E stained sections and normalized this value to the primary tumor volume at the time of sacrifice. Thus, despite harboring primary tumors of comparable volumes (Fig. 4C) and having spent longer periods developing in the mice, the Prkaca^CαR/fl PyMT mice contained only half the metastatic burden of the Prkaca^fl/fl PyMT controls (Fig. 4D,E). Importantly, Prkaca^CαR/fl PyMT tumors were also more sensitive to treatment with adriamycin (Fig. 4F), a commonly used chemotherapeutic agent in the neoadjuvant treatment of human triple negative breast cancer. Treatment with Adriamycin significantly improved the survival of mice bearing Prkaca^CαR/fl PyMT when compared to controls (Fig. 4G).

Observation of hematoxylin-eosin (H&E) stained tumor sections revealed significant differences in their histopathology. The control Prkaca^fl/fl PyMT tumors resembled human invasive ductal carcinomas (IDC) of high grade comprised of solid sheets of pleomorphic cells with abundant mitotic activity (Fig. S4C). In contrast, tumors from Prkaca^CαR/fl PyMT mice exhibited a more differentiated phenotype with gland formation and blander nuclear characteristics, akin to an intermediate grade invasive ductal carcinoma. Additionally, these tumors contained a rich vascularized stroma reminiscent of papillary differentiation (Fig. S4C). Activation of PKA also resulted in a large reduction in tumor cells co-expressing Vimentin and E-cadherin (Fig. 4I,J) indicating a partial loss of EMT traits. The wild-type tumors, also harbored a significant number of Krt14-positive cells basal cells as previously reported (Cheung et al., 2013), a feature that was less prominent in Prkaca^CαR/fl PyMT tumors (Fig. 4K,L).

Upon transplantation in limiting dilution into secondary hosts, Prkaca^CαR/fl tumors exhibited a 40-fold reduction in their tumor-initiating potential when compared to control tumors (Fig. 4H). To further understand the reduction in tumor-initiating potential we assessed the loss of stemness markers that have previously been uncovered in MMTV-PyMT tumors. We carried out immunofluorescence for Zeb1, which has been shown to enrich for tumor-initiating cells in this model (Ye et al., 2015) and observed that Prkaca^CαR/fl PyMT tumors exhibit reduced expression of Zeb1 compared to Prkaca^fl/fl PyMT controls (Fig S5A,B). In another study that characterized the heterogeneity of EMT states in MMTV-PyMT tumors, the CD106^neg CD51^neg subpopulation was found to contain the most metastatic subpopulation (Pastushenko et al., 2018). We observe that Prkaca^CαR/fl PyMT tumors have a 50% reduction in this compartment compared to Prkaca^fl/fl PyMT controls (Fig S5C). Together, these data point to a reduction in the EMT traits and decreased stemness potential in mammary tumors as a result of PKA activation.

Activation of PKA induces a more differentiated histology in multiple models

The histological differences in Prkaca^CαR/fl PyMT tumors compared to Prkaca^fl/fl PyMT control tumors, specifically that of increased differentiation, could have arisen as a result of aberrant development of the mammary gland prior to oncogenic transformation. To test whether induction of PKA in already formed tumors would have a similar effect, we used the inducible Gα_s^R201C/fl model in which activation of PKA signaling would occur following treatment with doxycycline (Fig. 5A). Upon doxycycline induction at 8-9 weeks of age until sacrifice, Gα_s^R201C/fl tumors also exhibited characteristics of papillary differentiation (Fig. 5B) indicating that the altered histological features were a result of activation of Gα_s-PKA signaling and not disproportionate cell-type representation in the pre-transformed mammary glands.

Figure 5: Differentiated histology induced by PKA across tumor models.

(A) Schematic of mouse crossing strategy to express the doxycycline-inducible Gα_s^R201C/fl allele in the mammary glands of MMTV-PyMT mice following which tumors were harvested and FFPE sections were stained with hematoxylin and eosin (B) to study their histopathology and differentiation status. (C) Schematic of mouse crossing strategy to express the PKA-CαR allele in the mammary glands of C3(1)-Tag mice following which tumors were harvested and FFPE sections stained with hematoxylin and eosin to study their histopathology (D). Fluorescence imaging of Prkaca^CαR/fl C3(1)-Tag and control Prkaca^fl/fl C3(1)-Tag tumors using Vimentin and E-cadherin to assess EMT status (E). Kaplan-Meier curve comparing the survival of Prkaca^CαR/fl C3(1)-Tag and control Prkaca^fl/fl C3(1)-Tag mice (F). Bar values as shown. Statistical significance was determined using a Log-rank (Mantel-Cox) test (P < 0.05 was considered significant; n = 22 mice).

The induction of differentiation was also observed in a second model of mammary tumorigenesis, C3(1)-Tag (Maroulakou et al., 1994)(Fig. 5C), which develops basal-like tumors with features of metaplasia. Prkaca^CαR/fl C3(1)-Tag tumors also show signs of differentiation with a reduction in metaplastic and sarcomatoid features, accompanied by a loss of Vimentin and gain in E-cadherin, when compared to control Prkaca^fl/fl C3(1)-Tag tumors (Fig. 5D,E). Importantly, Prkaca^CαR/fl C3(1)-Tag tumor-bearing mice also exhibit extended survival compared to Prkaca^fl/fl C3(1)-Tag controls (Fig S5F). Thus, the ability to induce differentiation is observed across both Luminal B and basal-like models of breast cancer upon induction of PKA signaling. The resulting tumor pathology that manifests following PKA activation appears to depend on the nature of the tumor and its cellular origins.

Differentiation Results from Altered Evolution of Tumor Subpopulations

In order to understand the molecular and cellular basis for the histopathological properties of the Prkaca^fl/fl PyMT and Prkaca^CαR/fl PyMT tumors, we longitudinally harvested cells at the early and late stages of tumorigenesis. At 9 weeks of age, the mammary glands of Prkaca^fl/fl PyMT mice are hyperplastic, while still largely maintaining their ductal architecture. We surgically resected one of two inguinal (no. 4) glands without euthanizing the animals, allowing them to develop tumors, which were later resected and sequenced. Using single-cell mRNA sequencing (scRNA-Seq) at both these time points, we were able to track the evolution of tumors at the hyperplasia and tumor stages (Fig. 6A). Based on expression of well-known marker genes, clusters were annotated into basal, mature luminal, luminal progenitor, lumino-basal and EMT-like (Fig. S6A-D).

Figure 6: Altered evolution of tumor cell subpopulations upon activation of PKA.

(A) Schematic of experimental set up to carry out sequential scRNA-Seq of hyperplastic and overt tumor samples from matched mice. tSNE plots show similar cellular constituents in hyperplastic glands of Prkaca^CαR/fl PyMT mice and Prkaca^fl/fl PyMT controls (B, D) but divergent evolution of subpopulations during the transition from hyperplasia to tumor stages of development (C, E). (F) tSNE plots highlighting the expression levels of the Lalba (left) and Sox4 (right) genes in tumors from Prkaca^fl/fl PyMT (top) and Prkaca^CαR/fl PyMT tumors (bottom) The lumino-basal subpopulation is highlighted by the red ellipse. Pseudotime analyses enabled the capture of the directionality of evolution from hyperplasia to tumor in the Prkaca^fl/fl PyMT tumors (G) that is altered upon PKA activation in Prkaca^CαR/fl PyMT tumors (H).

Dimensionality reduction and clustering of cells from Prkaca^fl/fl PyMT and Prkaca^CαR/fl PyMT mice yielded similar populations at the hyperplastic stage, with both exhibiting an expansion of the luminal progenitor compartment, while still retaining a portion of their normal basal and mature luminal subpopulations (Fig. 6B,D). There were, however, significant differences in the patterns of evolution from the hyperplasia to tumor stage. The Prkaca^fl/fl PyMT tumors harbored a luminal progenitor (LP) population that constituted the largest subpopulation within the tumor, but also contained two additional subpopulations. Firstly, lumino-basal (LB) subpopulation that expressed high levels of Sox4, which co-expressed LP and basal markers (Fig. 6C,E). Secondly, an EMT-like subpopulation, expressed relatively lower levels of a number of cytokeratins (Krt8, Krt14, Krt5, Krt79, Krt80), claudins (Cldn4, Cldn6, Cldn10), and markers of mammary differentiation (Prlr, Pgr, Lalba) (Fig. S6B,D). Notably, both these subpopulations were evident only at the tumor stage, indicating that they evolved between the hyperplasia and tumor stages (9-22 weeks of age). Importantly, the Prkaca^CαR/fl PyMT tumors harbored significantly reduced populations of both the lumino-basal and EMT-like clusters (Fig. 6F).

Pseudotime kinetics (Trapnell et al., 2014) of Prkaca^fl/fl PyMT tumors point to evolution from a hyperplastic signature to eventually acquiring a tumor signature (Fig. 6G). In contrast, the Prkaca^CαR/fl PyMT tumors exhibited an altered evolution trajectory whereby gene kinetics for tumor cells preceded that of hyperplastic cells in Prkaca^CaR/fl PyMT, with the tumor displaying a more differentiated profile than the hyperplasia (Fig. 6H). This provides a transcriptional basis for the Prkaca^CαR/fl PyMT tumors adopting a more differentiated histopathology, which likely drives the observed phenotypic and functional differences. The scRNA-Seq profiles reveal that the major cellular differences between the Prkaca^CαR/fl PyMT and controls lie in the evolution of the transcriptionally-distinct lumino-basal and EMT-like subpopulations. The presence of these cells, ostensibly, drives the metastatic and therapy-resistant traits of the wild-type tumors, which are diminished upon activation of PKA signaling. The emergence of these two subpopulations could mark the point at which wild-type tumors initiate metastatic spread.

Sox4 is a Key Downstream Target of PKA-induced Differentiation

To understand the mechanism by which PKA could induce tumor differentiation, we assessed the activation of the Hedgehog (Hh) pathway, which has previously been shown to act downstream of PKA (Iglesias-Bartolome et al., 2015). Upon treatment with forskolin for 30 minutes, nuclear translocation of the PKA catalytic subunit was observed, whereas no change in nuclear localization of Gli1 was observed (Fig S7B). Interestingly, a decrease in nuclear localization of β-catenin was observed after 6 hours of PKA activation, suggesting that repression of Wnt is likely an indirect downstream effect. We thus resorted to a more objective approach to uncovering possible mechanistic insights into pathways acting downstream of PKA.

To further understand the emergence of the lumino-basal subpopulation, we focused on Sox4, one of the key transcription factors that is highly expressed in this cluster (Fig. 6F). Other Sox family members have been shown to have profound impacts on mammary development and tumorigenesis (Dravis et al., 2015; Guo et al., 2012), whereas Sox4 has previously been demonstrated to play a role in EMT (Tiwari et al., 2013). Based on these studies, we hypothesized that PKA could be repressing Sox4 activity and affecting its transcriptional output. We first tested whether Sox4 loss could recapitulate the effects of PKA activation by carrying out CRISPR-Cas9 mediated knockout in PB3, a cell line derived from an MMTV-PyMT tumor (Dongre et al., 2017) (Fig S7A). Loss of Sox4 induced a partial MET in PB3 cells, resulting in the acquisition of a more epithelial-like morphology (Fig 7A) as well as upregulation of the epithelial marker E-cadherin and loss of the mesenchymal marker Vimentin (Fig. 7B). Importantly, mammary fat pad transplantation of Sox4^−/− PB3 cells resulted in the formation of tumors that exhibited characteristics of increased differentiation, with the presence of more epithelioid cells in comparison with controls that exhibit a more sarcomatoid profile (Fig. 7C). In line with these observations, Prkaca^CαR/fl PyMT tumors exhibit reduced levels of Sox4 nuclear localization when compared to wild-type controls (Fig 7D).

Figure 7: Suppression of Sox4 activity by PKA leads to tumor differentiation.

CRISPR-Cas9 mediated knockout of Sox4 in PB3 cells to assess differences in cell morphology as captured by bright-field images (A) and the expression of epithelial-mesenchymal markers by immunofluorescence (B). Bar, 5, 10μm. H&E stained FFPE sections of transplanted PB3 tumors were carried to estimate changes in histopathology which would reflect altered differentiation state (C). Bar, 50μm (D) Brightfield images of immunohistochemical staining of Sox4 in FFPE sections from Prkaca^fl/fl PyMT (top) and Prkaca^CαR/fl PyMT (bottom) tumors. Bar, 25μm. PKA phosphorylation sites on Sox4 identified by in-vitro kinase assays (E) including sites in the HMG-box containing DNA binding domain that are conserved across evolution (F). Electrophoretic mobility shift assays carried out in the presence of purified PKA-Cα and Sox4 (right) or PKA-Cα, Sox4 and ATP (left) to test for retardation of mobility upon phosphorylation of Sox4 by PKA (G).

To test whether Sox4 is a direct substrate of PKA, we carried out an in vitro kinase assay with purified PKA catalytic subunit and Sox4. A number of residues were found to be directly phosphorylated upon addition of ATP to the reaction, including S56, S103, S136 and S266 (Fig 7E, Table S1). The first two phosphorylated residues reside in the highly conserved HMG box-containing DNA binding domain (van de Wetering et al., 1993)(Fig 7F), suggesting that their post-translational modifications might have an effect on the DNA-binding capability of the protein. Indeed, as observed in an electrophoretic mobility shift assay (EMSA), the presence of Sox4 and PKA leads to a retardation of mobility of a fluorescence-labeled consensus oligo sequence, ostensibly due to the formation of Sox4-DNA complexes (Fig 7G). Addition of ATP to the reaction relieved the retardation, suggesting that PKA phosphorylation could impair the DNA binding ability of Sox4. These results implicate Sox4 as being a direct downstream substrate of PKA, phosphorylation of which impairs its ability to act as a transcription factor. Given the role that Sox4 plays in EMT and promoting tumor metastasis, it is likely to be activated during the progression of tumorigenesis as a result of PRKAR1A locus amplifications that result in a reduction in PKA catalytic subunit activity.

Discussion

This study reveals a role for PKA as a differentiation-promoting pathway in some breast tumors. The ability of PKA to promote stem or progenitor cell differentiation has previously been shown in epidermal and hair follicle stem cells (Iglesias-Bartolome et al., 2015), in the adrenal cortex (Drelon et al., 2016), and granule neural progenitors (He et al., 2014). Wnt and Hh pathways have been previously implicated in stem cell self-renewal and maintenance in different tissue types (Briscoe and Therond, 2013; Clevers et al., 2014). Although PKA has previously been shown to inhibit β-catenin (Drelon et al., 2016), our results point to a reduction in nuclear translocation only after 4 hours of PKA activation (Fig S7B), suggesting a secondary mechanism of repression of the Wnt pathway. In line with this assumption, Sox4 has previously been shown to play a role in stabilizing β-catenin (Lee et al., 2011; Sinner et al., 2007), pointing to an alternate mechanism by which the latter could be de-stabilized and inactivated upon PKA activation. The reduction in nuclear localization of β-catenin also correlates with increased cytoplasmic accumulation of Sox4 (Fig S7B).

Previous studies identified a subpopulation in MMTV-PyMT tumors that expresses basal markers and is responsible for collective invasion (Cheung et al., 2013). Our scRNA-Seq results also point to the emergence of a subpopulation of cells, the lumino-basal (LB) fraction, that is correlated with an increase in metastatic burden. We observe that these cells still retain the expression of luminal markers, including Krt8 and CD14, while also expressing basal markers. Our scRNA-Seq analyses reveal that activation of PKA suppresses the acquisition of basal traits that appear to be necessary to generate LB cells, mirroring phenotypes observed in the normal mammary gland. In drawing parallels between normal development and tumorigenesis, Sox4 expression has also been correlated with the process of mammary gland involution (Shibayama et al., 2019). Assessing the activation state of PKA during involution could provide further clues on how this kinase-substrate duo regulate the balance between self-renewal and differentiation. We observed features of papillary differentiation in tumors that arise in two separate models of PKA activation, one from glands that expressed constitutive activation of PKA (Fig. S4C), as well as upon activation of PKA following initiation of tumorigenesis (Fig. 5B). Interestingly, the lack of an intact myoepithelial layer appears to be a feature of malignant papillary carcinomas (Pal et al., 2010), indicating that the resulting papillary features could be a direct effect of a decrease in basal/myoepithelial cells. Moreover, the activation of Gα_s-PKA signaling appears to confer papillary features in other tumor-types (Wu et al., 2011), although the overall tumorigenic or tumor suppressive effects appear to be tissue-type dependent.

The presence of recurrent genomic amplifications in PRKAR1A suggests a role for suppression of this pathway in tumorigenesis. Our analyses looked at genomic alterations in each regulatory and catalytic subunit and we decided to pursue PRKAR1A amplifications owing to their presence in greater proportions of Luminal B and basal-like breast cancers (Fig 3B,D). Indeed, the major downstream effectors of the PKA complex are the catalytic subunits, whose enzymatic activity is known to regulate numerous cellular processes (Skalhegg and Tasken, 2000; Tasken and Aandahl, 2004). Our efforts were, thus, focused on PRKACA, for which have uncovered a critical role in the induction of differentiation. This suggests that higher levels of PRKACA could confer better prognosis in breast cancer patients, whereas the negative regulators of the catalytic subunit, encoded by PRKAR1A and other genes, could confer poorer prognosis.

It must also be noted that, while the effects of PKA reported here, and by others (Drelon et al., 2016; He et al., 2014; Iglesias-Bartolome et al., 2015; Klutzny et al., 2018; Nadella et al., 2008; Pattabiraman et al., 2016), may be more akin to those of a tumor suppressor, there are contexts in which PKA signaling acts in a growth-promoting fashion. It is known to play oncogenic roles in pancreatic and small cell lung cancers (Coles et al., 2020; Patra et al., 2018) induces EMT in certain contexts (MacPherson et al., 2010; Shaikh et al., 2012), and promotes cell proliferation through downstream activation of Src and MAPK (Beristain et al., 2015; Stork and Schmitt, 2002). These seemingly inconsistent observations are reflective of highly context-specific roles for PKA in different tissue types. Moreover, shifts in balance of proteins that maintain the tightly regulated network of Gα_s-PKA signaling including adenylyl cyclases, phosphodiesterases, A-kinase anchoring proteins (AKAPs) and PRKAR proteins all play a role in dysregulation of this pathway. In the subset of breast tumors that we focus on here, the amplifications and increased expression of PRKAR proteins are likely what triggered the suppression of the catalytic subunit and its downstream effects.

Activation of PKA results in increased chemosensitivity of PyMT tumors to adriamycin, likely a result of a more differentiated histopathology. We have previously observed increased sensitivity to both adriamycin and paclitaxel in-vitro upon PKA-induced mesenchymal-epithelial transition (Pattabiraman et al., 2016). Here we extend these observations to show that in-vitro effects of MET induction manifest in vivo in the form of enhanced differentiation. Given that the effects are observed with multiple chemotherapeutic drugs acting via different mechanisms, it is likely that the increased sensitivity is a result of differentiation rather than any specific interactions between adriamycin and PKA. Indeed, along these lines, others have previously shown that inducing a more differentiated cell state renders tumor cells more sensitive to a range of chemotherapeutic drugs in breast cancers (Bhola et al., 2013; Gupta et al., 2009; Qu et al., 2014).

Basal cells in mammary development, while contributing to the regenerative capacity of the gland during pre-pubertal development, also have other heterotypic signaling functions. Specifically, basal-to-luminal cell signaling has been shown to be critical for luminal cell maturation (Forster et al., 2014). Upon activation of PKA, which impairs basal cell self-renewal, we observe no overt effects on the proportions (Fig 1B), the histological phenotype (Fig 2D), or maturation (Fig 6E) of luminal cells, suggesting that basal-to-luminal cell heterotypic signaling are still active. This is conceivable since the basal cells that persist following PKA activation, while defective in self-renewal, may still carry out other normal functions. Alternatively, it is plausible that PKA signaling may also diminish the ability of basal cells to signal to luminal cells, ultimately preventing their maturation. Currently available Cre models are, however, insufficient to distinguish these two scenarios as the percentage of recombination invariably leaves a portion of cells unlabeled or in the wild-type state. Along these lines, previous studies have highlighted the polyclonal nature of metastasis in the PyMT model (Cheung et al., 2016), suggesting a role for basal-luminal heterotypic interactions in the establishment of metastatic colonies. We observe significantly reduced metastases in Prkaca^CaR/fl PyMT tumors (Fig 4D), which we interpret as a result of reduction in the lumino-basal subpopulation of the tumor. The reduction in metastasis could, however, also occur as a consequence of impaired basal-luminal cell interactions.

This study provides rationale for the targeting of PKA signaling in at least certain breast tumor types, although there are a number of challenges associated with the translation of such a strategy. Molecular therapeutics have traditionally relied on inhibition of enzymatic activity e.g., the use of kinase inhibitors, whereas our results point to a requirement for activation, rather than inhibition of PKA signaling, which would present a unique set of challenges. Additionally, since PKA is a ubiquitous kinase that carries out numerous context-specific roles in different tissue types, it is likely that a strategy to target it will be met with issues of toxicity and untoward side-effects. On the flip side, identifying key downstream molecules could help alleviate some of these concerns i.e., targeting Sox4 could be a more attractive avenue, especially since its expression pattern appears to be confined only to certain tissue types and lacking in others such as the liver, pancreas and small intestine (Dy et al., 2008). Our studies show that PKA directly phosphorylates Sox4 at S56 and S103, both residues falling within its HMG box that is critical for its DNA binding activity (Dy et al., 2008). Strategies that interfere with Sox4 DNA-binding or transactivation might have clinical utility in the progression of Luminal B and basal-like breast cancers that harbor PRKAR amplifications.

Our study provides insights into how the induction of differentiation could help derive therapy responsiveness in solid tumors, while curtailing their aggressive traits and metastatic spread. The ability of constitutively active PKA signaling to deplete self-renewal of basal cells in multiple tissue types suggests that careful calibration of its activity is essential for the maintenance of the stem cell state and the regulation of cellular differentiation. While targeted and immunotherapies have shown immense success in a number of cancer types, the inherent cellular heterogeneity of some breast cancers are a major barrier to the identification of targetable recurrent genomic alterations. For such cancer types, a more feasible approach could be the reversal of a trait that is common to all cancers: the evasion of terminal differentiation.

STAR Methods

Resource Availability

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Diwakar Pattabiraman (raman@dartmouth.edu).

Materials Availability

All unique/stable reagents generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

Data Availability

scRNA-Seq data generated in this study have been deposited in GEO and are accessible under the accession number GSE158257.

Experimental Model and Subject Details

Plasmids

pLentiCRISPRv2 was a gift from the Feng Zhang via Addgene (#52961, http://n2t.net/addgene:52961; RRID:Addgene_52961). sgRNAs for Sox4 were cloned into the construct using protocols from the Zhang lab. The Sox4 sgRNA sequences that were used were:

sgSox4_2: CAACAACGCGGAGAACACTG

sgSox4_4: CGACAAGATTCCGTTCATCC

sgLacZ: TGCGAATACGCCCACGCGAT

Cells

PyMT tumor-derived PB3 cells were obtained from the Robert A. Weinberg lab (Whitehead Institute for Biomedical Research) and were cultured in DMEM/F12 medium containing 5% adult bovine serum with 1× penicillin–streptomycin and 1× nonessential amino acids. For the Sox4 knockout studies, cells were cultured for 14 days in the presence of puromycin for complete CRISPR-Cas9 mediated loss of Sox4, following which they were used for subsequent studies including immunoblotting, immunofluorescence and transplantation.

Animal studies

Research involving animals complied with protocols approved by the Dartmouth College Committee on Animal Care. All animals were housed in specific pathogen-free (SPF) barrier facility in ventilated microisolator cages containing 5 mice each. The PKA-CαR (Niswender et al., 2005) mice were a gift from Stanley McKnight (UWashington). Tg(MMTV-cre)^4Mam/J line D (Wagner et al., 2001) was obtained from JAX (Stock No. 003553). The Gα_sR201C mice (Iglesias-Bartolome et al., 2015) were a gift from J. Silvio Gutkind. The B6.Cg-Gt(ROSA)26Sorr^{tm1(rtTA,EGFP)Nagy/J} mice (Belteki et al., 2005) (Stock No. 005670), mTmG reporter mice - B6.129(Cg)-Gt(ROSA)26Sorr^{tm4(ACTB-tdTomato,-EGFP)Luo/J} mice (Muzumdar et al., 2007) (Stock No. 007676), Tg(KRT5-cre/ERT)^Blh/J mice (Stock No. 029155)(Van Keymeulen et al., 2011), Tg(KRT8-cre/ERT)^Blpn/J mice (Stock No. 017947)(Van Keymeulen et al., 2011), Tg(MMTV-PyVT)^634Mul/LellJ mice (Guy et al., 1992) on a C57Bl/6J background (Stock No. 022974) and Tg(C3-1-TAg)^cJeg/JegJ mice (Maroulakou et al., 1994) (Stock No. 013591) were obtained from The Jackson Laboratory. For transplantation studies, cells suspended in DMEM containing 30% Matrigel (GFR)/PBS mix (BD Biosciences; 356230) were injected into the inguinal mammary gland fat pads of age-matched female NOD/SCID or C57Bl6/J mice (Jackson Laboratory). Mice were euthanized after 10 weeks or when tumors reached a diameter of 2 cm. For chemotherapy studies, mice were administered adriamycin intraperitoneally at 2mg/kg twice a week for up to 40 days or until tumors reach the endpoint of 2cm in diameter. For studies of mammary development, female mice at 3 weeks, 6 weeks and 8 weeks were used to study the initiation of postnatal development, puberty and adulthood, respectively. For tumor studies, 9 week-old female mice were used for the hyperplasia stage and 18-23 week-old mice were used for the tumor stage. For the study of metastasis and therapy resistance, female tumor-bearing mice that were 23-40 weeks-old were used.

Method Details

Mammary Gland Preparation

Mouse mammary fat pads (no. 3 and 4) were harvested and dissociated as per previously published protocols (Prater et al., 2013) with minor modifications. Briefly, glands were digested with Collagenase A and Hyaluronidase for 2-3 hours at 37°C in a rotator followed by mild trituration with a 10ml pipette. Following red cell lysis, cells were treated with trypsin, and subsequently with dispase to aid dissociation. Cells were then stained with antibodies (see table below) to analyze and sort luminal and basal populations

Carmine Alum Staining

Harvested mammary glands are spread out on a slide and allowed to dry for 1 hour at RT. Slides are fixed overnight in Carnoy’s fixative (6 parts 100% EtOH, 3 parts CHCL3, 1 part glacial acetic acid) at RT. On day 2 slides were rehydrated by incubating in 70% ethanol twice for 10 min each, twice in 50% ethanol for 10 min each, twice in 30% ethanol for 10 minutes each, twice in 10% ethanol for 10 minutes each and in distilled water for 5 min. Slides are then stained for 1-2 days in Carmine Alum (Stem Cell Technologies; Catalog No. 07070) at RT. Slides are dehydrated by incubating in 70% ethanol twice for 10 min each, twice in 95% ethanol for 10 min each and twice in 100% ethanol for 10 minutes each. Slides were then cleared in Xylene for 3-4 days, with the solution being changed every day after which they are rehydrated and dehydrated again as above before mounting using Permount (Fisher Chemical; SP15). Slides were scanned on a Perkin Elmer^© Vectra 3 slide scanner and analyzed on Phenochart^©.

Organoid Assays & Transplantation

Dissociated mammary epithelial cells were dissociated into single cells and cultured with advanced DMEM (Gibco; 12491015) containing 5% Matrigel, 5% heat-inactivated FBS, 10 ng/ml EGF, 20 ng/ml bFGF, 4 mg/ml heparin, and 5 mM Y-27632. Cells were seeded at 1,000 per well in 96-well ultralow attachment plates (Corning; 29443-034). The number of organoids was counted 7–14 days after seeding.

1000 or 500 cells from dissociated organoids were resuspended in 10 μl 75% DMEM/ 25% matrigel were injected into the number 4 glands of 4w old NSG female mice that had been cleared of endogenous epithelium. Recipient glands were harvested at 12wk, dissected, fixed and stained with carmine alum. An outgrowth was defined as an epithelial structure comprising ducts and branching and the percentage of outgrowth was quantified by proportion of ducts and branches in relation to the fat pad.

Tumor Dissociation

Tumors from MMTV-PyMT mice and C3(1)-Tag mice were resected and minced using a razor blade in DMEM containing 2 mg/mL collagenase and 100 U/mL hyaluronidase (Fischer Sci) in a rotator at 37°C for 30 minutes. Following incubation, tumors were minced further and re-incubated for another 30 minutes. Dissociated cells were washed twice in PBS and filtered through a 70- and 40-μm cell strainer to obtain single-cell suspensions.

Immunofluorescence (cultured cells)

Cells were cultured on dishes containing coverslips for 2 to 3 days, after which coverslips were washed in cold PBS, fixed in 5% Neutral Buffered Formalin for 10 min at 4°C and permeabilized in 0.2% TritonX in PBS for 2 min. Cells were then washed in PBS, blocked for 1 hour at room temperature in PBS containing 3% normal horse serum (Vector Labs, USA; S-2000). Fixed cells were then incubated with the primary antibody in PBS containing 1% bovine serum albumen (BSA) solution overnight at 4°C. Cells were washed in PBS three times, and secondary antibody was added in PBS containing 1%BSA solution for 1 to 2 hours at room temperature in the dark. Cells were washed three times in PBS and were incubated for 2min in DAPI solution, after which they were washed in PBS and mounted with a drop of Prolong Diamond antifade reagent (Thermo; P36962) and placed on coverslips. Slides were viewed on a Zeiss^© LSM800 microscope and analyzed using the Zen^© Digital Imaging software.

Immunofluorescence (FFPE tissue sections)

Slides were rehydrated by incubating in Histoclear solution twice for 5 min each, followed by incubation in 100% ethanol twice for 5 min each, in 95% ethanol twice for 5min each, 70% ethanol twice for 5 min each, once in 35% ethanol for 5 min, and in water for 5 min. Pressure cooker– mediated heat-induced epitope retrieval was carried out in 250 ml of unmasking buffer containing sodium citrate at pH 6. After retrieval, slides were blocked for 30 min in PBS containing 3% normal horse serum after which they were incubated with primary antibody in blocking solution overnight at 4°C. Slides were washed twice with PBS and incubated with secondary antibody at room temperature for 1 hour in the dark. After two PBS washes, 20 ml of mounting medium was added, then slide contents were topped with coverslips, and stored in the dark for 24 hours before imaging on a Zeiss LSM800 microscope and analyzed using the Zen Digital Imaging software.

Antibody	Vendor	Catalog No.	Application	Dilution
CD45 PECy7	BioLegend	103114	FACS	1:100
CD31 PECy7	BioLegend	102418	FACS	1:100
EpCAM APC	BioLegend	118214	FACS	1:100
CD49f PE	BioLegend	313612	FACS	1:100
Vimentin	Cell Signaling	5741	IF/IHC	1:100
E-cadherin	BD	610182	IF/IHC, WB	1:200
Keratin 8	Thermo Fisher	PA5-29607	IF/IHC	1:400
Keratin 14	Thermo Fisher	MA5-11599	IF/IHC	1:1000
Sox4	Thermo Fisher	MA5-31424	WB	1:1000
p-PKA substrate	Cell Signaling	9624	IHC	1:200

Analysis of publicly available datasets

The Cancer Genome Atlas (TCGA) (Cancer Genome Atlas, 2012) and METABRIC (Curtis et al., 2012; Pereira et al., 2016) breast cancer gene expression, copy number, clinical and survival data, were accessed from the cBioPortal database (www.cbioportal.org). The Swedish Cancerome gene expression and clinical data were publicly available in the Gene Expression Omnibus database (www.ncbi.nlm.nih.gov/geo/) under accession GSE96058 (Saal et al., 2015). All statistical analyses were performed in the R statistical computing environment (v3.5.1). Specifically, the correlation between mRNA expression of two genes were assessed by ordinary least-square regression. The univariate Kaplan-Meier analyses and related data visualizations were implemented in R packages survival (v.2.43.3) and survminer (v.0.4.3)(Therneau and Grambsch, 2000).

Inferring copy number alterations

Copy number alteration profiles were inferred relative to normal breast samples (n=5, Oltra et al., 2018). MethylationEPIC array intensities were used to estimate segmented copy number alterations, defined as the base 2 logarithmic ratio of tumor-to-control copy number (log2R) (Feber et al., 2014). The frequency of copy number gains or losses was determined using a log2R threshold of +0.1 and −0.1, respectively.

Single-cell RNA-seq library construction and sequencing

Single Cell Capture and Library Preparation: Immediately following dissociation, single cell suspensions were placed on ice and counted on a Luna automated cell counter. Cell concentrations from each sample were be normalized to 1000 cells/ul and loaded onto a Chromium Single Cell A Chip (10x Genomics Inc.) targeting a capture rate of 5,000 cells per sample. Single cell RNA-seq libraries were be prepared using the Chromium Single Cell 3’ v2 kit (10x Genomics) following the manufacturer’s protocol. Libraries were quantified by qubit and peak size determined on a fragment analyzer instrument. All libraries were pooled and sequenced on an Illumina NextSeq500 High Output 26bp x 98bp run to generate an average of 50,000 reads/cell. Data Analysis: Raw sequencing were processed using the 10x Genomics Cell Ranger to generate quality metrics and primary data visualizations as well as gene expression matrices for downstream analysis in R using Seurat and other open-source packages.

Single-cell RNA-seq data processing

To investigate the transcriptional evolution of tumors at single cell level, we performed of matched hyperplasia and tumor samples was performed on the 10X genomics platform, and generated data at an average of ~78M reads per sample with cell numbers ranging from ~2600 to 5000. Single-cell transcriptome sequencing raw reads were quality filtered using Fqtrim (v0.9.7) tool [10.5281/zenodo.1185412]. Reads were trimmed and filtered for low quality bases, poly-A/T tails and N bases while retaining the paired-end integrity of the reads. Read1 of the pair containing barcodes was not considered for trimming but allowed to be filtered to maintain paired-end integrity. Sequencing and PCR errors in cell barcodes can convolute the process of differentiating reads per cell barcode; hence we used UMItools (v0.5.4) to distinguish the reads per cell (barcode) accommodating for technical errors(Smith et al., 2017). A knee density method-based approach is used in UMItools to estimate the number of acceptable cell barcodes and then reads were assigned for each cell.

Clean and barcode classified reads were aligned against Mouse genome reference (GRCm38) using STAR aligner (v2.5.3a)(Dobin et al., 2013) and output was restricted to uniquely aligned reads. De-duplication of transcripts with the same UMIs arising from PCR amplification were removed. Reads were assigned with position based annotation of genomic features using featureCounts module(Liao et al., 2014) in the subread package (v1.6.0). Taking advantage of the UMI information, read counts were extrapolated to quantify molecular level count for each transcript using directional-adjacency method-based count module in UMItools. Reads were grouped per cell (based on barcode) and then a gene expression matrix was generated with RNA molecule count of genes in rows for each cell in columns represented in GxC matrix format (where G is gene in rows and C is cells in columns).

Cell-type classification and Pseudotime analysis

Gene expression matrix was filtered for cells with <500 genes expressed and <500 total UMI counts genes and >0.25 percentage of reads aligned to Mitochondrial genome using Monocle2 (v2.6.4) R package(Qiu et al., 2017a; Qiu et al., 2017b; Trapnell et al., 2014). Further, outliers of total mRNAs count for each cell was removed from the downstream analysis. Expression counts were normalized using negative binomial distribution of library sizes. Genes with mean expression value of 0.1 across cells were used for PCA based dimensionality reduction and then clustered using an unsupervised densityPeaks algorithm in Monocle2 and projected using t-SNE method.

To classify the population into various cell-types, we investigated the expression of curated and well-established cell-type specific marker genes’ expression in our data and selected a specific list of markers that are expressed and/or not expressed for each cell-type. Following are the cell-type specific markers used for the classification in this study,

Cell types	Higher expression
Basal cells	Krt14, Krt5, Snai2
Luminal Mature cells	Pgr, Prlr, Foxa1, Gpx3, Esr1
Luminal Progenitor cells	CD14, Kit, Lalba
Lumino-basal	Krt14, Krt8, CD14
EMT-like	Trim29, Vim, Cldn6, Sox4

We identified and classified cell-types using Monocle2’s classifyCells function. First, each cell was annotated into a cell-type based on the presence and absence of positive and negative cell-type marker genes respectively. Then cells were re-clustered using added cell-type information to bring cells with similar gene-expression pattern and cell-type annotation into closer clusters, and to segregate un-annotated cells. Ambiguous and Un-annotated clusters were investigated manually for other known marker genes using find_cluster_markers function in Monocle3 alpha (v2.99.3) and re-annotated manually into different sub-types based on GO enrichment analysis of selected marker genes and expression of other known cell-type specific marker genes.

Pseudotime analysis attempts to reconstruct the transcriptional transitions of cells during the process of tumorigenesis. By predicting the transcriptional state of each cell, cells are ordered into an estimated pseudotime indicating their cell-state and transition across a time-series. Transition of gene expression kinetics was performed by integrative Pseudotime analysis using Monocle2. Given that the hyperplasia and tumor stage cells were extracted from the same mice, the hyperplasia and tumor stage samples of same mice were merged using Seurat R package (v2.3.0)(Butler et al., 2018) while retaining the cell-type and sample annotation. Merged data was processed again for dimensionality reduction using DDRTree algorithm and then pseudotime trajectories were calculated. Plots were generated using different color codes based on cell-type, sample and pseudotime to illustrate the gene expression kinetics transition across stages and samples. Data plots were generated using Monocle2 and Monocle3 alpha packages.

In-vitro kinase assay

Replicate samples from in vitro kinase assays were prepared for LC-MS/MS analysis by SP3 isolation and on-bead digestion exactly as described (Hughes et al., 2014; Hughes et al., 2019) and trypsin digestion overnight in 50mM ammonium bicarbonate buffer. In brief, kinase assay samples were diluted to 100ul final volume with 1% SDS/50mM Tris pH 8.1, followed by addition of DTT to 2mM and reduction at 45C for 15 minutes. Samples were allowed to cool to room temperature and alkylated with 7mM iodoacetamide (final concentration) for 1h at room temperature in the dark, followed by quenching with 2mM DTT (final concentration). SP3 was performed by addition of 1ul SP3 magnetic beads and vortexing thoroughly to mix, precipitation with 100ul 200 proof ethanol, shaking in a thermomixer at 24C for 5 minutes at 1000 RPM, and isolation of the beads in a magnetic tube rack. The supernatant was removed to waste, and the beads were washed 3 times with 80% ethanol/water solution with vortexing in between. After the final wash, the supernatant was removed, and the beads were additioned with 50ul 50mM ammonium bicarbonate solution containing 250ng sequencing grade (Promega) trypsin and digestion overnight at 28C in a thermomixer with shaking at 2000 RPM. Finally, the resulting peptides were separated from beads by magnetic tube rack, desalted over homemade STAGE-TIPs and loaded on to the LC-MS/MS system. Digested peptides were separated from paramagnetic beads by magnetic isolation and desalted on an OASIS uHLB (2mg) desalting plate (Waters-Millipore). The resulting desalted peptides were dried by vacuum centrifugation, resuspended in 1% formic acid/5% methanol/94% ultrapure water, transferred to limited volume inserts and injected on to a Proxeon EasyLC-1200 UPLC system couple to an Orbitrap Fusion Lumos mass spectrometer. Samples were analyzed by data-dependent analysis using a linear gradient from 0 - 39% B (Buffer A: 0.125% formic acid, 3% acetonitrile; Buffer B: 0.125% formic acid, 80% acetonitrile) over 45 minutes on an in-house manufactured microcapillary reverse-phase column (100um ID, 34cm long). Parent ion scans (R = 120,000, AGC target = 200,000 ions, charge states 2 - 4, dynamic exclusion +/− 10ppm for 20s) were collected for precursors from 350 to 1500 m/z, followed by child ion scans (R = 15,000, AGC target = 25,000 ions, quadrupole isolation = 1.0 m/z, HCD collision = 28%, maximum injection time = 50ms), all as centroided Orbitrap scans, for a 1.5s duty cycle (top speed mode). The resulting data were then data-searched using the COMET algorithm (Eng et al., 2013) against the human proteome (UniProt) using dynamic modifications for oxidation on methionine and phosphorylation on serine, threonine and tyrosine, and filtered to a 1% FDR using the target-decoy strategy (Elias and Gygi, 2010). These peptide identifications were then quantified in a "label-free" mode using in-house software across all six replicates. Quantified phosphopeptides were then collapsed into single identifiers as a function of modified residue ("ModSites") as done previously (Kettenbach et al., 2011). Quantification values were then averaged for each ModSite across the three replicates for control and + ATP samples; these averaged quantification values were then compared for significance using Student's T-test (Excel). ModSites that were identified in at least two of three replicates and exhibited a p-value < 0.05 were included in Table S1.

Preparation of oligonucleotides duplexes

Oligonucleotides end-labeled with or without IRDye 800 (IDT) infrared dye were annealed by combining the oligonucleotides at a 1:1 molar ratio with complementary oligonucleotides for probes or competitors, respectively in 25 mM HEPES, 1 mM MgCl2 and 50 mM NaCl. The duplex mixtures were heated at 95°C for 5 minutes and then at 70°C for 10 minutes, and samples ramped down to 25°C for 45 minutes to dissociate any intra-strand duplexes.

Electrophoresis Mobility Shift Assay

In vitro-kinase assays were performed with 500ng recombinant human cAMP Protein Kinase Catalytic subunit (Abcam, ab56268) and 1000ng recombinant Human SOX4 (Novus Biologicals, H00006659-Q01) proteins with and without ATP in kinase assay buffer (20mM HEPES pH 7.7, 5mM MgCl2, 0.1mM EGTA, 0.2mM DTT, 0.5mM ATP) for 6 hours. Labeled probes (50 fmol) were incubated with the in vitro kinase assay reaction in 20 mL of binding buffer (10 mM Tris-HCl, pH 7.5, 50 mM KCl, 3.5 mM DTT, 0.25% Tween 20, 5% glycerol, 5 mM MgCl2, and 50 mM EGTA) and 1 mg poly (dI-dC), in the presence or absence of non-labeled competitor (50 pmol) for 30 min at room temperature and then run on to 6% polyacrylamide gels (Thermo Fisher Scientific, EC6265BOX) in 0.5X TBE buffer at 4°C. The infrared dye signal was detected and digitized using an Odyssey CLx imaging system (LI-COR Biosciences, Lincoln, NE, USA).

Elf5_F_sense_1

CAAGCTCCTCATCTATCAGGGAACAAAGGCTAAACATGGCTGTCCACCAT

Elf5_R_antisense_1

GTTCGAGGAGTAGATAGTCCCTTGTTTCCGATTTGTACCGACAGGTGGTA

Elf5_F_sense_Mutant

CAAGCTCCTCATCTATCAGGGATCAGAGGCTAAACATGGCTGTCCACCAT

Elf5_R_sense_Mutant

GTTCGAGGAGTAGATAGTCCCTAGTCTCCGATTTGTACCGACAGGTGGTA

Quantification and Statistical Analysis

Data are presented as means ± SD. A Student’s t test (two-tailed) was used to compare two groups (P < 0.05 was considered significant) unless otherwise indicated. Numbers of animals (biological replicates) are indicated in figure legends.

Supplementary Material

Table S1

Table S1: Identification of PKA phosphorylation sites on Sox4 by Mass Spectrometry. Related to Figure 7. LC-MS analysis of PKA-Sox4 in-vitro kinase assay. Table outlines raw data obtained from LC-MS showing peptides on Sox4 that are modified upon addition of ATP. Fold change is shown under the “Ratio” tab and significance was calculated by Student T-test.

KEY RESOURCES TABLE

REAGENT or RESOURCE		SOURCE		IDENTIFIER
Antibodies
CD45 PECy7		BioLegend		Cat# 103114
CD31 PECy7		BioLegend		Cat# 102418
EpCAM APC		BioLegend		Cat# 118214
CD49f PE		BioLegend		Cat# 313612
Vimentin		Cell Signaling Technology		Cat# 5741
E-cadherin		BD Biosciences		Cat# 610182
Keratin 8		Thermo Fisher		Cat# PA5-29607
Keratin 14		Thermo Fisher		Cat# MA5-11599
Sox4		Thermo Fisher		Cat# MA5-31424
p-PKA Substrate		Cell Signaling Technology		Cat# 9624
Chemicals, Peptides, and Recombinant Proteins
Prolong Diamond antifade reagent		Thermofisher Scientific		Cat# P36962
Normal Horse Serum		Vector Labs		Cat# S-2000
Collagenase A		Sigma		Cat# 10103586001
Hyaluronidase		Fisher Scientific		Cat# ICN10074091
Carmine Alum		Stem Cell Technologies		Cat# 07070
Permount		Fisher Chemicals		Cat# SP15
Recombinant PRKACA		Abcam		Cat# ab56268
Recombinant SOX4		Novus Biologicals		Cat# H00006659
Deposited Data
scRNA-Seq data of cells from PrkacaCαR/fl PyMT and Prkacafl/fl PyMT mice at the hyperplasia and tumor stages.		GEO		GSE158257
Experimental Models: Cell Lines
PB3 cell line		Weinberg lab (WIBR)		Dongre et al. 2017
Experimental Models: Organisms/Strains
PKA-CαR mice		Stanley McKnight (UW)		Niswender et al. 2005
GαsR201C mice		J. Silvio Gutkind (UCSD)		Iglesias-Bartolome et al. 2015
Tg(MMTV-cre)4Mam/J line D		The Jackson Laboratory		Stock No. 003553
B6.Cg-Gt(ROSA)26Sortm1(rtTA,EGFP)Nagy/J		The Jackson Laboratory		Stock No. 005670
B6.129(Cg)-Gt(ROSA)26Sortm4(ACTB-tdTomato,-EGFP)Luo/J		The Jackson Laboratory		Stock No. 007676
Tg(KRT5-cre/ERT)Blh/J		The Jackson Laboratory		Stock No. 029155
Tg(KRT8-cre/ERT)Blpn/J		The Jackson Laboratory		Stock No. 017947
Tg(MMTV-PyVT)634Mul/LellJ		The Jackson Laboratory		Stock No. 022974
Tg(C3-1-TAg)cJeg/JegJ		The Jackson Laboratory		Stock No. 013591
Oligonucleotides
CAACAACGCGGAGAACACTG	This paper		sgSox4_2
CGACAAGATTCCGTTCATCC	This paper		sgSox4_4
TGCGAATACGCCCACGCGAT	This paper		sgLacZ
CAAGCTCCTCATCTATCAGGGAACAAAGGCTAAACATGGCTGTCCACCAT	This paper		Elf5_F_sense_1
GTTCGAGGAGTAGATAGTCCCTTGTTTCCGATTTGTACCGACAGGTGGTA	This paper		Elf5_R_antisense_1
CAAGCTCCTCATCTATCAGGGATCAGAGGCTAAACATGGCTGTCCACCAT	This paper		Elf5_F_sense_Mutant
GTTCGAGGAGTAGATAGTCCCTAGTCTCCGATTTGTACCGACAGGTGGTA	This paper		Elf5_R_sense_Mutant
Software and Algorithms
Fqtrim (v0.9.7)		https://ccb.jhu.edu/software/fqtrim/		10.5281/zenodo.1185412
UMItools (v0.5.4)		https://github.com/CGATOxford/UMI-tools		Smith et al., 2017
STAR aligner (v2.5.3a)		https://github.com/alexdobin/STAR		Dobin et al., 2013
subread package (v1.6.0), featureCounts module		http://subread.sourceforge.net		Liao et al., 2014
Monocle2 (v2.6.4) R package		http://cole-trapnell-lab.github.io/monocle-release/		Qiu et al., 2017a

Highlights.

• Activation of PKA attenuates self-renewal and expansion of the mammary basal compartment

• This activation induces tumor differentiation, curtailing metastasis and chemoresistance

• PKA activation attenuates the evolution of multiple aggressive tumor subpopulations

• PKA directly phosphorylates and inactivates Sox4, contributing to tumor differentiation