Structural mechanism for inhibition of PP2A-B56α and oncogenicity by CIP2A

To characterize the functional relevance of interaction between CIP2A head domain and B56α, we performed mutagenesis spanning the first 40 amino acid residues of CIP2A (Fig.2A). The mutations were introduced to V5-tagged full-length CIP2A(1-905) mammalian expression constructs, and CIP2A protein expression was examined in 22RV1 prostate cancer cells. The chosen strategy was supported by previous data that CIP2A protein stability in cells directly reflects its B56 binding efficiency31. Furthermore, this was the only available approach to study the contribution of these N-terminal head domain amino acids in the context of full-length CIP2A protein because production of recombinant full-length CIP2A protein, in quantities and qualities required for functional experiments, remains elusive.

Notably, strongly indicative for the functional importance of the head domain, all six tested mutants showed an effect on full-length V5-CIP2A(1-905) protein levels, as compared to the wild-type protein (Fig.2B, see Supplementary Fig.5Afor Western blots). In particular, the K8A single point mutation practically abolished, whereas K21A and K40A mutations significantly inhibited full-length CIP2A protein expression. On the other hand, A24E mutation increased full-length CIP2A protein expression when compared to the wild type (WT) (Fig.2Band Supplementary Fig.5A). Acknowledging the functional importance of high CIP2A protein expression in human cancers18,21,23,24,26,33, we explored the potential impact of cancer-derived CIP2A head domain mutations on the protein expression. Based on positioning of the mutated residues and on the anticipated effect of mutation on B56α binding, we selected three CIP2A head mutations (Q16E, S22A, E23K) from the COSMIC and cBioPortal databases. Notably, when tested in MDA-MB-231 TNBC cells, dependent of CIP2A protein expression for their tumour growth34, Q16E showed comparable CIP2A stabilization as with the A24E mutant when compared to the V5-tagged WT protein (Fig.2C).

Corroborating with the decreased full-length CIP2A protein expression in 22RV1 cells (Fig.2B), bacterially produced CIP2A(1-560) K8A mutant demonstrated decreased direct B56α binding as compared to the WT CIP2A(1-560) in in vitro pull-down assay (Fig.2D, E). Also the recombinant CIP2A(1-560) K21A mutant was found to interact less efficiently with B56 than the CIP2A 1-560 WT protein (Supplementary Fig.5B, C). The A24E mutation, when tested in the context of the N-terminal CIP2A(1-330) fragment, instead showed increased direct B56α binding (Fig.2F, G). Due to their instability, the other tested head domain mutations could not be expressed as bacterial proteins. As these results strongly indicate that effects of K8A and A24E mutations on CIP2A protein stability is linked to their B56α-binding propensity, we asked whether the decreased cellular stability of K8A mutant full-length CIP2A could be rescued by increasing its B56α binding affinity via concomitant A24E mutation. Indeed, when MDA-MB-231 cells were transiently transfected with a construct carrying both A24E and K8A mutations, we observed that full-length CIP2A protein expression was rescued to the WT level (Fig.2H,I). As a molecular explanation for the CIP2A protein stability changes, we examined potential ubiquitination sites on the full-length protein. Previously CIP2A protein stability has been shown to be regulated by CHIP-mediated ubiquitination35, but the ubiquitination site on CIP2A has not been characterized. Based on database information (https://www.phosphosite.org/), lysine 647 (K647) is the most prevalently ubiquitinated CIP2A amino acid. Notably, K647A mutation rescued the expression of CIP2A K8A mutant in MDA-MB-231 cells, but did not further increase the expression of already stabilized A24E mutant (Supplementary Fig.5D, E).

Collectively, these data demonstrate that the N-terminal head domain of CIP2A mediating interaction with B56α is a critical determinant of full-length CIP2A protein expression in cancer cells. The results further indicate that decreased stability of N-terminal head mutants is mediated by C-terminal K647, which may be differentially exposed to ubiquitination depending on the tightness of CIP2A-B56α interaction.

Single amino acid mutation on the N-terminal head domain of CIP2A abrogates TNBC tumorigenicity

To test the cancer relevance of the N-terminal head domain of CIP2A, we created MDA-MB-231 single cell clones carrying K21A mutation by CRISPR/Cas9. K21A mutation was chosen due to its intermediate destabilizing effects on CIP2A protein expression (Fig.2B), as the near complete loss of CIP2A protein expression as observed with K8A mutant, was expected to prevent growth of single cell MDA-MB-231 clones based on previous CRISPR/Cas9 knock-out data (Avana 2020 Q1;https://depmap.org)24. K21 was targeted by two independent guide RNAs (gRNA) using direct ribonucleoprotein (RNP)-mediated transfection36. Both gRNAs resulted in genomic CIP2A targeting (Supplementary Fig.6A) and in decreased CIP2A protein expression in transfected cell pools (Supplementary Fig.6B). Based on more robust impact on CIP2A expression, gRNA #2 cell pool was selected for single cell cloning. Out of sequenced single cell clones, we could verify K21A mutation from two clones (clone1 and clone2) (Supplementary Fig.6C) which were in experiments compared to the control cells that were RNP-transfected without gRNA.

Decreased expression of CIP2A protein in both single cell clones was verified by Western blotting (Fig.3A). CIP2A targeting also inhibited expression of validated PP2A-B56α target MYC (Fig.3A). Importantly, when assessed by proximity ligation analysis (PLA), the K21A mutation on the full-length endogenous CIP2A protein almost entirely abolished the CIP2A-B56α interaction (Fig.3B, C), although there was still approximately 50% of K21A mutant protein expressed in the cells used for this assay (Supplementary Fig.6D). The specificity of the PLA reaction was tested by performing the assay without CIP2A primary antibody (Supplementary Fig.6E). These data further validate critical role of the N-terminal head domain on the full-length CIP2A interaction with B56, and on CIP2A protein stabilityin cellulo.

Interestingly, the colony formation potential in plastic was comparable between the two CIP2A hypomorph clones and the control cells (Fig.3D, E). However, fully consistent with the requirement of PP2A-B56 inhibition for anchorage-independent growth6,7, CIP2A K21A targeted clones were almost completely incompetent to grow as spheres on soft agar (Fig.3F, G). The data above demonstrated decreased interaction between K21A mutant of CIP2A and B56 in vitro andin cellulo. However, it was yet unclear how much the decreased protein expression of K21A mutant of CIP2A contributed to loss of malignant growth on soft agar. To address this, we created two independent CIP2A shRNA clones that have approximately similar levels of CIP2A protein expression than was observed in K21A mutant Crispr clones (Supplementary Fig.6F). We also titrated CIP2A siRNA to the levels that only partially inhibited CIP2A expression (Supplementary Fig.6G). Notably, the both types of CIP2A hypomorph MDA-MB-231 cells were yet fully capable in forming colonies in soft agar (Supplementary Fig.6H–K). This demonstrates that the total loss of malignant growth potential of K21A mutant Crispr clones (Fig.3F, G) cannot be explained only by partial inhibition of CIP2A protein expression in these clones, but it also involves loss of CIP2A-B56 interaction.

Based on the results that single K21 amino acid on CIP2A is imperative for transformed cell growth, the effects of K21A mutation on MDA-MB-231 cell tumorigenicity was tested by orthotopic xenograft assay. Notably, both clones showed significantly impaired tumour growth as compared to control cells (Fig.3H). Especially with clone 1, only 5/8 tumours were palpable, and the tumours grew very slowly reaching the average size of only 113 mm3after 30 days (Fig.3I). As CIP2A K21 was selectively essential for anchorage-independent growth of MDA-MB-231 cells in vitro (Fig.3F, G vs. D, E) and in vivo (Fig.3H, I), we assessed both induction of anoikis (apoptosis due to loss of cell adhesion), and PP2A-regulated phosphotargets related to cell survival in anchorage-independent conditions. To this end, PARP cleavage and phosphorylation of MEK MAPKK and vimentin were studied from clones cultured for 24 h on either plastic or on low-affinity polyHEMA-coated plates. Increased PARP cleavage confirmed the increased sensitivity or K21 targeted clones to anoikis induction under low-affinity conditions (Fig.3J). Interestingly, phosphorylation of pro-survival MEK MAPKK was decreased in K21 targeted clones in both conditions, but the effect was more pronounced in cells from low-affinity plates providing a plausible explanation for increased anoikis37(Fig.3J). Further, phosphorylation on PP2A target vimentin38, supporting anchorage-independent growth39, was strongly reduced in both K21A clones in both conditions.

Collectively, these results validate the cancer relevance of the N-terminal head domain of CIP2A. They also identify targeting of the head domain as an efficient strategy to block malignant growth of CIP2A-driven TNBC cells24(Fig.3K).

CIP2A hijacks B56α and PP2Ac from the PP2A-B56α heterotrimer

Although the results above clearly demonstrate the relevance of the N-terminal head domain on oncogenicity by CIP2A, it is still totally unknown how CIP2A mechanistically affects PP2A-B56α holoenzyme function. To address this, we mapped the CIP2A(1-560) cross-link sites to B56α structure in the PP2A-B56α trimer configuration. Interestingly, comparison of the hypothetical surface area which CIP2A(1-560) could cover on B56α based on the cross-link sites (Fig.4A, red), to the known sites involved in B56α interaction with the A-subunit (Fig.4A, purple)40,41, revealed that these surface areas are highly overlapping (Fig.4A, right panel, shaded circle). Also, one of the CIP2A cross-linking sites, K358, is adjacent to the PP2Ac interaction surface on B56α (Fig.4A, right panel). These observations provoked us to hypothesize that CIP2A(1-560) binding to B56α might interfere with PP2A-B56α holoenzyme assembly.

To study this intriguing possibility, we incubated GST-fused CIP2A(1-560) with the pre-assembled PP2A-B56α holoenzyme (A-B56α-PP2Ac) in vitro and analyzed reconstitution of protein interactions from glutathione-bead immunoprecipitates by Coomassie gel stain. Supporting the structure-based hypothesis above, CIP2A was found to capture B56α and PP2Ac, but not the scaffolding A subunit from the PP2A trimer (Fig.4B). These results were confirmed by Western blot analysis of the glutathione-bead immunoprecipitates from independent experiments (Fig.4C). Consistently with other data, no direct interaction between CIP2A(1-560) and A-subunit could be detected in a GST pull-down experiment (Supplementary Fig.7A). These results were substantiated by demonstration of CIP2A concentration-dependent displacement of the scaffold subunit, and extraction of B56 and PP2A-c from the holoenzyme (Fig.4Dand Supplementary Fig.7B). Importantly, the data also demonstrated higher capability of CIP2A(1-560) to extract B56α from the holoenzyme than CIP2A(1-330) (Fig.4D). This is consistent with the role of C-terminal interaction region in stabilizing the CIP2A-B56 interaction31(Fig.1C, D). Interestingly, while CIP2A-PP2Ac interaction has not been previously reported, their direct interaction was confirmed by GST pull-down experiment (Supplementary Fig.7C).

The results above imply a model where CIP2A can potentially function as a pseudo A-subunit for a trimeric CIP2A-B56α-PP2Ac complex. To demonstrate that such a CIP2A-B56α-PP2Ac complex can be reconstituted, we incubated the pre-assembled PP2A holoenzyme (A-B56α-PP2Ac) with either GST only, or with GST-CIP2A(1-560), and subjected the samples for analytical gel filtration analysis (Fig.4E, Fand Supplementary Fig.7B). Reassuringly, in the control samples co-incubated with GST (Fig.4E, middle panel), the A-B56α-PP2Ac trimer components co-eluted in fractions 3 and 4, and without GST, whereas in the samples incubated with GST-CIP2A(1-560) (Fig.4F, middle panel), CIP2A co-eluted with B56α and PP2Ac without the A-subunit in the fractions 2-4. We also conducted gel filtration to mimic conditions that would exist following exclusion of A-subunit from the CIP2A-B56α-PP2Ac trimer. Therefore we incubated either GST alone, or GST-CIP2A(1-560), with B56α and PP2Ac, which do not constitute a stable dimer without any scaffolding protein40,41. Consistent with this previous data, B56α and PP2Ac did not co-elute when incubated with GST (Fig.4E, lower panel). However, in the presence of GST-CIP2A(1-560), PP2Ac and B56α co-eluted in high molecular weight complexes 2 and 3 containing also GST-CIP2A(1-560) (Fig.4F, lower panel).

These results reveal reorganization of PP2A-B56α holoenzyme upon CIP2A binding, resulting in formation of unforeseen trimeric CIP2A-B56α-PP2Ac complex (Fig.4G).

CIP2A mutes LxxIxE motif-dependent substrate binding to B56α

The results above identify a mechanism by which CIP2A hijacks B56α and PP2Ac from the PP2A-B56α complex. However, if CIP2A would merely function as a pseudo-A subunit, it is unclear whether the newly formed CIP2A-B56α-PP2Ac heterotrimer could still recognize B56α targets. Excitingly, we made a notion that the CIP2A(1-560) cross-linking sites on B56α (aa. 171, 181, 217, and 227), were lining the groove which recognizes the LxxIxE motif in B56α substrates (Fig.5A). On the other hand, the recently identified additional determinant for high affinity binding of LxxIxE substrates to B56α, an acidic patch between B56α amino acids E301-E34142, is sandwiched between the LxxIxE motif, and the C-terminal CIP2A cross-link sites in B56α (aa. 347, 354, and 358) (Supplementary Fig.8A). These results indicate that CIP2A could mute B56α in the CIP2A-B56α-PP2Ac trimer by sterically preventing recognition of the LxxIxE motif containing substrate proteins. To study this possibility, we used a LxxIxE motif-containing protein fragment from BubR1, which is the prototypic B56α substrate15,43,44. In the experiment, GST-tagged BubR1(647-720) was pre-incubated with B56α prior to the addition of V5-tagged recombinant CIP2A(1-560), and binding efficiency of B56α to BubR1 was assessed from the glutathione bead pull-down samples. Notably, whereas CIP2A was found not to interact with BubR1 (Fig.5B, lane 2), it efficiently out-competed B56α from BubR1 (Fig.5B, C, lane 1 vs lane 3).

To directly link these results to structural requirements of CIP2A-B56α interaction, we tested whether the head domain peptide of CIP2A (aa.18-40) could alone block B56α-BubR1 interaction. In the experiment, B56α was pre-incubated with the CIP2A head peptide, after which the complexes were co-immunoprecipated with GST-BubR1. Fully consistent with our model, BubR1 binding to B56α was reduced in the presence of CIP2A head peptide (Fig.5D, E). We further compared CIP2A binding to the B56α variants with groove mutations that abolish binding of the LxxIxE motif targets10. Notably, whereas both of the B56α variants tested, Y215Q and R222E, exhibit total loss of BubR1 binding10, they had only slight impact on CIP2A binding: Y215Q did not influence CIP2A binding at all, and R222E had about 40% inhibitory effect as compared to B56α wild-type (Fig.5F, G). These results demonstrate that CIP2A inhibits B56α binding to LxxIxE motif substrate, but is itself not a classical LxxIxE motif substrate.

To address whether the cross-linked residues neighbouring the LxxIxE binding groove on B56 have a role on either CIP2A binding, or on PP2A holoenzyme assembly, we mutated the K181, K217 and K227 to alanines. In in vitro GST pull-down assay, these mutations did not impact direct CIP2A-B56 interaction (Supplementary Fig.8B). This could be explained by multiple other CIP2A-B56 cross-link sites potentially stabilizing the interaction enough under these experimental conditions. However, when testedincellulo, these B56 mutations had an instrumental role in stability of PP2A trimer, as demonstrated by very potent inhibition of both B56-A and B56-PP2Ac interaction, and loss of catalytic PP2A activity (Fig.5Hand Supplementary Fig.8C) from the B56 mutant protein pull-down samples. These results demonstrate that in addition to its critical role in substrate recognition, the LxxIxE groove region on B56 has also an important role in PP2A trimer stability. This is best explained by polar interactions between D177 of the PP2A-A subunit and K227 of the B56α (Fig.5I; PDBID 6NTS). Consistent with the conserved interaction between CIP2A and both B56α and B56γ31(Supplementary Fig.5B, C), D177 of PP2A-A interacts in a similar fashion with K202 of B56γ (Fig.5I; PDB: 2IAE41).

These results provide mechanistic explanation for displacement of the PP2A-A subunit from B56 upon CIP2A binding to the LxxIxE region of B56.

Validation of the “hijack and mute” model of PP2A-B56α inhibition by CIP2Ain cellulo

The above in vitro results reveal an unprecedented hijack and mute mechanism of PP2A holoenzyme inhibition. In this model, CIP2A both expels the A-subunit from the PP2A-B56α trimer and subsequently inhibits B56α binding to LxxIxE motif substrates. To validate these two inhibitory modesin cellulo, we first analyzed the impact of CIP2A on B56α interactome by affinity purification coupled to MS (AP-MS) approach45. For that purpose, we used murine NIH3T3 cells stably expressing either human CIP2A(1-905)V5, or empty vector control (Fig.6Aand Supplementary Fig.8D). NIH3T3 cells do not express endogenous CIP2A protein, but CIP2A over-expression increases their proliferation potential21. These features make them a perfect cellular model for studying PP2A-B56α inhibition resulting from CIP2A over-expression. These stable cells were transiently co-transfected with YFP-tagged B56α, and GFP-Trap immunoprecipitation followed by MS analysis was used to identify B56α interacting complexes (Fig.6A). The results of AP-MS experiment with three replicate samples confirmed association of CIP2A with B56α in CIP2A(1-905)V5 overexpressing cells only (Fig.6Band Supplementary Data2). Further, fully supportive of the model that CIP2A can expel A-subunit from the A-B56α-PP2Ac trimer, A-subunit (PPP2R1A) association with B56α was significantly decreased in CIP2A over-expressing cells (Fig.6B). On the other hand, validating the function for CIP2A in preventing B56α-substrate interactions, among all B56α co-precipitating proteins for which CIP2A over-expression had a significant impact (> Log2 fold change 1,p < 0.05), 80% of proteins showed decreased B56α binding (Fig. 6B, C). Notably, 63% of these proteins had a candidate LxxIxE motif (Fig.6Cand Supplementary Data2) which is consistent with our in vitro results.

ANIH3T3 cells, stably expressing human CIP2A(1-905) V5, or empty vector control (see Supplementary Fig.8D), were transiently transfected with YFP-B56α WT, followed by immunoprecipitation using GFPTrap and analysis by mass spectrometry.BProteins differentially interacting with B56α in CIP2A(1-905)V5 over-expressing cells (N= 3) vs. control (N= 3). Indicated proteins exceed the threshold of 2-fold difference withp < 0.05 (two-sided t-test using log2 of protein abundances, normalized to the protein abundance of B56α in the CTRL_1 sample, as an input). CPie-chart of share of proteins displaying either decreased or increased binding to B56α upon CIP2A over-expression (excluding PPP2R1A and CIP2A). Bar graph indicates percentual share of proteins with putative LxxIxE motif among the proteins which interaction with B56α was decreased by CIP2A.DPie-chart presentation of percentual share of proteins dephosphorylated (FDR < 0.05) in CIP2A K21A clone1 as compared to control clone in relation to putative LxxIxE motif found from the dephosphorylated protein (in orange), or whether dephosphorylated protein is an interactor of a protein with validated LxxIxE motif (in red).ESignificantly enriched kinase motifs based on phosphopeptides dephosphorylated in CIP2A K21A clone1. Background indicates the number of motifs found from all phosphopeptides identified by MS from the same samples.FReactome analysis image capture of the process Senescence enriched in targets dephosphorylated in CIP2A K21A clone1 as compared to control clone. Schematic presentation of CIP2A-mediated PP2A-B56α inhibition by the identified hijack and mute mechanism. In normal cells with low CIP2A expression, PP2A-B56α binds to its substrate protein (light blue) via LxxIxE groove and the adjacent acidic patch (orange) resulting in substrate dephosphorylation. In cancer cells with highCIP2Agene transcription, CIP2A binding to PP2A-B56α trimer results in expel of A and the formation of CIP2A-B56α-PP2Ac trimer. In this alternative trimer CIP2A shields the LxxIxE groove and the adjacent acidic patch from B56α substrates and thereby the substrate remains phosphorylated (dark blue). Binding of CIP2A to B56α via its head domain further stabilizes CIP2A protein.

To substantiate these conclusions, we validated the impact of N-terminal K21A mutation on phosphorylation of LxxIxE-motif containing proteins. Comparison of LC-MS analyzed phosphoproteomes between the control clone and CIP2A K21A clone 1 (both in triplicates) resulted in identification of 78 phosphopeptides from 63 individual proteins that were significantly (FDR < 0.05) downregulated in CIP2A K21A cells (Supplementary Data3). The entire MS/MS identification data can be found from supplementary Data4and from GEO database (PXD035179).

Notably, 53% of the proteins dephosphorylated CIP2A K21A cells had a candidate LxxIxE motif (Fig.6Dand Supplementary Data4). It was previously shown that LxxIxE containing proteins can act as scaffolds for the recruitment of other proteins for PP2A-B56α mediated dephosphorylation46. Using the PINA.3 database47, we found that an additional 15% of proteins identified in our screen are interactors of proteins with validated or predicted LxxIxE docking motifs46,48(Fig.6Dand Supplementary Data4). Thus, the 68% of proteins dephosphorylated in CIP2A K21A clones are either LxxIxE containing proteins, or their interactors, strongly supporting the notion that CIP2A inhibits substrate recognition of PP2A-B56. We also analyzed enrichment of kinase target motifs among the dephosphorylated peptides from CIP2A K21A cells using all quantified phospho-peptides from the same MS analysis as the background (Fig.6E). Notably, among the enriched motifs, 14-3-3 protein binding motif is a firmly established PP2A-B56α target in many signal transduction proteins49. Further, recent studies indicate functional relationship between PP2A-B56 and CLK2 in different physiological and pathological context50,51. Lastly, it was recently reported that RSK2-mediated phosphorylation of Emi2 promotes its well-known interaction with PP2A-B5652,53. Finally, in a pathway analysis, CIP2A-regulated phosphoproteins were enriched among senescence-associated proteins (Fig.6Fand Supplementary Data5), which with the previous evidence linking CIP2A to senescence evasion18, provides a plausible mechanistic explanation for the potent tumour growth inhibition in CIP2A K21A mutant clones.

Collectively, these results provide comprehensivein cellulovalidation that CIP2A inhibits PP2A-B56α function by hijack and mute mechanism: by expelling the A-subunit from the trimer, and by shielding B56α from binding to its LxxIxE motif substrate proteins (Fig.6F).

Discussion

During the past two decades, there has been a major advancement in understanding how protein serine/threonine-specific protein phosphatases (PPPs), historically considered promiscuous enzymes lacking substrate specificity, can selectively dephosphorylate their substrates1,3. The key to the precision of most PPPs in substrate recognition is formation of dimeric, or trimeric, complexes between the catalytic subunit and different regulatory subunits; such as B-subunits for the trimeric PP2A complexes. Thereby, and especially as different B-subunit containing PP2A complexes can have even opposite cellular functions, such as tumour suppression or oncogenesis54, understanding the function and regulation of individual PP2A B-subunits is essential. Amino acid sequence determinants for selective recognition of target proteins have recently been discovered for some PPP regulatory subunits1,10,14,42,49,55. However, whether the regions of regulatory subunits mediating the substrate interactions. are themselves subject to regulation by endogenous mechanisms, is poorly understood. The regulation at this level would provide a sophisticated cellular mechanism to selectively impact PPP sub-complex functions, without causing deleterious effects via unselective modulation of PPP catalytic activities. Identification of such substrate recognition regulatory mechanism could also facilitate selective therapeutic PPP modulation approaches.

In this work we discovered structural mechanism inhibiting PP2A-B56 complex assembly and substrate recognition by a human oncoprotein CIP2A. We demonstrated that CIP2A can hijack B56α and PP2Ac from the PP2A-B56α trimer, function as a scaffold for the newly formed CIP2A-B56α-PP2Ac trimer, and subsequently mute substrate recognition by B56α (Fig.6F). While the majority of the experiments were performed by using B56α protein, we validated the impact of N-terminal head mutant K21A also by using B56γ (Supplementary Fig.5B, C). Together with the notion that the B56 family proteins are structurally very conserved on the regions implicated here in mediating CIP2A interaction, we postulate that the presented results can potentially be generalized over the B56 protein family. Based on the results, it can be envisioned how binding of CIP2A to the proximity of K227 on B56α would interfere with B56-PP2A-A interaction and thereby hijack B56 from PP2A-A (Fig.5H, I). In this model, the local protein concentrations, and the active repulsion of PP2A-A from B56 upon CIP2A binding would be more critical than pure competition based on protein interaction affinities. Although the mechanism described here is clearly different from the mechanism previously described for TIPRL-mediated PP2A inhibition56, they resemble each other regarding formation of an alternative trimer in which the inhibitor protein (CIP2A or TIPRL) substitutes for the canonical PP2A complex components. Whereas CIP2A replaces the A subunit from PP2A trimer, TIPRL, via several steps, hijacks the C-subunit to form a TIPRL-PP2Ac-α4 complex56. Together these studies indicate that hijacking of the PP2A complex components might be a commonly utilized mechanism for PP2A inhibition by endogenous inhibitor proteins.

Very importantly, we provide in cellulo validation for all major conclusions of the study. First, single point mutants of CIP2A head domain amino acids mediating B56α binding dramatically impacted CIP2A protein expression in cancer cells (Fig. 2). It is important to note that these results validate the relevance of the mutated N-terminal residues on the function of full-length CIP2A also containing the unstructured C-terminal tail. Second, AP-MS analysis of the B56α interactome confirmed that CIP2A over-expression expel A-subunit from B56α (Fig. 6B). Third, combination of two types of proteomics analysis with bioinformatics analysis confirmed that CIP2A inhibits B56α binding to LxxIxE motif substrates in cancer cells (Fig. 6 and Supplementary Data 4). Fourth, mutation of only one amino acid from the N-terminal head of CIP2A was sufficient to abrogate tumour growth of aggressive TNBC cell line (Fig. 3). Together, the discovered two-headed strategy by which CIP2A both disrupts functionality of the PP2A-B56α tumour suppressor complex, and simultaneously is itself protected from degradation, is likely to explain notable potency of CIP2A as a human oncoprotein. However, while PP2A-B56α obviously is an important target for CIP2A´s oncogenic activity, our data do not exclude other potential mechanisms by which CIP2A promotes oncogenesis.

The co-distribution of CIP2A cross-links with amino acids known to be critical for A subunit interaction with B56α provide mechanistic basis for competition between CIP2A and A subunit for B56α (Fig. 4A). We also provide substantial evidence to support the B56α muting function for CIP2A. The cross-linking data, mutagenesis, and peptide competition experiments all support the conclusion that CIP2A N-terminal head binding to B56α sterically hinders LxxIxE motif substrate binding. However, unlike the bonafide B56α LxxIxE substrates, CIP2A binding to B56α is not dependent on the actual LxxIxE binding groove. Thereby CIP2A is not merely a pseudo-substrate for PP2A-B56α, but a steric inhibitor protein (Fig. 6F). Our results further implicate an important function for the LxxIxE groove region on stabilizing the PP2A trimer. Very interestingly, the recently identified CIP2A variant NOCIVA (NOvel CIP2A VAriant), that consists of the N-terminal CIP2A amino acids 1-545 fused to a unique C-terminal tail57, contains all structural features identified here critical for CIP2A-B56α interaction. Therefore, it is likely that results of this study are relevant also for the understanding of the mechanism by which NOCIVA promotes myeloid cancer therapy resistance57. It is, however, evident that additional structural studies would be required to understand the molecular-level details of both CIP2A and NOCIVA interactions with PP2A-B56.

These results are likely to advance development of PP2A reactivating therapies. Our results reveal the CIP2A head domain as a critical structural region mediating PP2A-B56α inhibition, but also high CIP2A protein expression, as observed across human cancers19,33. Therefore, the results would pave the way for therapeutic strategies by targeting CIP2A. Based on near complete inhibition of tumorigenic properties of MDA-MB-231 cells by mutation of single amino acid K21 (Fig. 3), we propose that therapeutics preventing CIP2A head domain binding to B56α would constitute an efficient cancer therapy strategy by simultaneously preventing PP2A-B56α inhibition and by causing CIP2A protein destruction. Notably, the benefits of drug-induced target protein degradation are clear. Such an approach would not only inhibit CIP2A functions towards PP2A-B56α, but would remove any other CIP2A protein functions such as direct TopBP1 binding24,26, and results in longer pharmacodynamic effects that are predicted to remain even after drug has been metabolized. Although the functional relevance was validated for TNBC cells here, therapeutic impact of inhibition of CIP2A protein expression has been validated by numerous studies across human cancer types18,20,21,23,33,34. CIP2A inhibition also sensitizes cancer cells to large number of cancer therapies27,28,30.

In summary, this work describes a hijack and mute mechanism of multiprotein complex regulation that mediate inhibition of tumour suppressor PP2A-B56α. The results also provide clear cues for the development of novel targeted cancer therapeutics by revealing structural determinants behind oncogenicity of CIP2A. As CIP2A is also strongly implicated in Alzheimer´s disease via regulation of both Tau and APP32,58, these results may have broad relevance to human diseases also beyond cancer.