Protein phosphatase 2A inactivation induces microsatellite instability, neoantigen production and immune response

Microsatellite-instable (MSI), a predictive biomarker for immune checkpoint blockade (ICB) response, is caused by mismatch repair deficiency (MMRd) that occurs through genetic or epigenetic silencing of MMR genes. Here, we report a mechanism of MMRd and demonstrate that protein phosphatase 2A (PP2A) deletion or inactivation converts cold microsatellite-stable (MSS) into MSI tumours through two orthogonal pathways: (i) by increasing retinoblastoma protein phosphorylation that leads to E2F and DNMT3A/3B expression with subsequent DNA methylation, and (ii) by increasing histone deacetylase (HDAC)2 phosphorylation that subsequently decreases H3K9ac levels and histone acetylation, which induces epigenetic silencing of MLH1. In mouse models of MSS and MSI colorectal cancers, triple-negative breast cancer and pancreatic cancer, PP2A inhibition triggers neoantigen production, cytotoxic T cell infiltration and ICB sensitization. Human cancer cell lines and tissue array effectively confirm these signaling pathways. These data indicate the dual involvement of PP2A inactivation in silencing MLH1 and inducing MSI.

6. Page 14, line 315: The authors claim that all MSI CRC belong to the consensus molecular subtype I. This is an overstatement, as in Guinney et al. 2015 only 75% of MSI CRC were classified as subtype I. Similarly, the MSS tumors are distributed over the subtypes, with only a smaller proportion showing features of subtype IV. The statement needs to be amended. 7. The following section (page 15, line 324) "Therefore, MSI CRC has the tumour microenvironment of type I CMS ..." contains confident statements about combination therapies, claiming for two scenarios that "there is no need to combine Treg inhibition". Those are theoretical considerations, reasonable or not, and should be indicate as such. In this context, studies demonstrating clinical benefit from combination of anti-PD-1 and anti-CTLA4 should be considered.  9. Figure 3g: The difference in E2F1 levels are very small, however, dramatic differences in DNMT and MLH1 levels are observed. How can this be explained?
Reviewer #2: Remarks to the Author: The study shows an interesting connection of PP2A inactivation to MLH silencing and MSI and demonstrates that PP2A knockout promotes the therapeutic response to ICB. The study begain with a mouse model showing depletion of ppp2r1a which encodes the PP2A A scaffold subunit increases immunogenicity, T cell infiltration and induction of MSI status. This piece of data in the mouse model is convincingly demonstrated to support the claim that PP2A indeed has a role to play in cancer immunity. The author went on to examine if the finding obtained from the mouse is clinically relevant. Using TCGA database, they claimed that the mRNA levels of PP2A endogenous inhibitors such as CIP2A and SET are significantly higher in CRC samples with MSI compared to that with MSS, while the PPP2R1A is lower. Protein analysis using CRC tissue array further shows the PPPR1A is indeed lower in MSI tumors while the CIP2A and SET are higher. The subsequent experiment using the mouse model explored the mechanism. It showed that PPP2R1A coimmunoprecipiated the HDAC2 and Rb1 etc and PPP2R1a loss resulted in phosphorylation of HDAC2 and Rb1, leading to increased DNMT expression and MLH1 methylation and silencing. Finally, in a C26 syngenic CRC model, they showed that ppp2R1a knockdown increased response to anti-PD1 and a combination of a chemical inhibitor of PP2A with anti-PD1 also shows some better response in growth inhibition. Major concerns Although the mouse data documenting a role of PP2A ( by PPP2R1a A structural unit) in regulating MLH1 expression is convincing, the clinical data supporting the clinical relevance of PP2A-PPP2R1A loss in relation to MSI status in CRC is questionable. First, as the endogenous inhibitor of PP2A, CIP2A and SET function to regulate PP2A activity, instead of PPP2R1A expression. The authors provided data showing that CIP2A and SET is weakly upregulated in MSI tumors compared to MSS tumors but this does not necessarily warrant a claim that PP2A activity is lower in MSI tumors. Showing PPP2R1a is lower in MSI tumors is strange as PPP2R1a expression in general is not downregulated in CRC. Instead, other B subunits of PP2A are widely silenced by DNA methylation in 90% of CRC. This suggests that the majority of the CRC have PP2A dysfunction as a tumor suppressor. Given that MSI tumor is only found in 10-15% of CRC, it is hard to believe that only the MSI tumors have the PP2A inactivation while most MSS tumors are not. The authors need to address this more carefully. For the mechanistic study, the biochemical evidence showing HDAC2 and Rb1 are PP2A substrates are lacking. Although they are co-immunoprecipitated, it is necessary to demonstate that they are the direct substrates of PP2A. PPP2R1a pulldown for an in vitro phosphatase assay using CT26 cell lysis as a substrate will substantiate the mechanistic claim.
The final data showing PP2A inhibitor sensitizing PD1 therapy is not impressive. I find it is hard to appreciate the claim of using PP2A inhibitor as a strategy to induce MSI and then sensitize ICB. As PP2A is widely inactivated in human CRC, it is not convincing that inhibition of PP2A can be a strategy for cancer therapy. Also, inhibiting PP2A as a tumor suppressor can cause lots of oncogenic signaling, including RB E2F1 and many other known oncogenes. I am afraid that the therapeutic implication of using PP2A inhibitor in CRC has limited potential (though it has been reported before). Also, the study focuses on colorectal cancer and it is not clear why the in vivo studies also included the triple-negative 4T1 and pancreatic model.

Reviewer #3:
Remarks to the Author: The antigenicity and hence the immunogenicity of tumours is likely to be a major limiting factor in response to immunotherapy. Altering immunogenicity is a major challenge and this study is potentially important because it attempts to rise to this challenge by evaluating a pathway which may give rise to new antigens.
Whilst the experiments examining the impact of targeting PP2A on MSI status are compelling, the experiments examining effects on the immune response fall short of providing definitive answers. As such, the data as it stands is over-interpreted.
The histology shown in Figure 1 needs some improvement. It is difficult to understand why no CD4+ cells are observed when Foxp3+ cells are seen? What are the cells stained in the lower right panel?
In studies of human CRC (and other cancers) CD8+ T cells and Tregs are normally reported as positively correlating? This is because Tregs are induced when there is an immune response to suppress. This does not appear to be the case in the analysis shown here. Could the authors comment?
The tumour growth curves in Figure 4 show significant differences however these are only assessed for a short period of time (up to day 21). Did the study extend beyond this time-point? The comparison of lymphocyte numbers by IF staining must have been carried out on very small tumours where PP2A is absent. Smaller tumours often have more lymphocytes / g tumour compared to larger tumours thus the authors should provide more details and/or normalise for tumour size.
It is impossible to conclude that there are more neoantigens generated in these tumours without exome sequencing. This is the key missing piece of data. The number of TCRs alone is not sufficient evidence of neoantigen-driven clonal expansions.
The data with the small molecule inhibitor is less compelling. Also, despite the authors' claim that there is an effect beyond an impact of Tregs; this is not proven by the experiment carried out as it does not include the use of the PI-3065 plus anti-PD1 alone as a control. In addition, the authors should note that PI-3065 has effects beyond just direct effects on Tregs. It also affects effector T cells directly as well as monocytes. This experiment needs a re-think to include all necessary controls as well as a "cleaner" method of targeting Tregs.
1. In the legend to Figure 5, the authors state that treatment with LB-100 was performed for 2 days. Length differences for BAT25 between LB-100-treated cells and untreated controls are extremely pronounced with more than 10 basepairs difference. Acknowledging previous studies and estimations for basepair deletion rates per cell division, this cannot be explained. Previous studies suggested mononucleotide peak patterns (precisely of the markers used in the present study) as molecular clocks. These previous studies consistently reported the emergence of MSI-indicating peaks only months after the onset of MMR deficiency, which is in stark contrast to the data presented here. As this experiment is a crucial to convincingly demonstrate the functional connection between PP2A inactivation and MSI, thorough re-analysis is required. To assess this point, the authors should perform standardized time scale experiments, quantitatively examining changes of the peak patterns in parallel after knockout and LB-100 treatment. Response: Thank you very much for your great comments. We have now provided more data to prove the causal link between PP2A loss with MSI induction and its relevance to human CRC. Western blotting analysis showed that treatment with different shRNAs against PPP2R1A or with different PP2A inhibitors (LB100 and LB102) for 2 or 7 days decreased MLH1 protein levels in SW620 and another human MSS CRC cell line, HT29 (Fig. 5d). Western blot analysis of SW620; HT29 transfected with indicated shRNAs, and treated with vehicle control or LB100/LB102. We have also now used commonly recognized method and definition to define whether the tumour was MSI 1 . The panel of markers included D2S123, D5S346, D17S250 and BAT25. Compared with control cells, HT29 treated with PP2A inhibitors for 2 or 7 days or with different shRNAs against PPP2R1A showed changes in the length of all markers used ( Fig. 5e and Fig. S11, S12). Moreover, the profiles of marker length changes caused by treatment with PP2A inhibitors (LB100) for 2 or 7 days or the use Page 2 of different shRNA treatments against PPP2R1A were almost similar. These data suggest that MSI induction caused by PP2A inhibition or knockdown occurs very rapidly and shares the similar profiles of MSI marker length changes. These findings were supported by MSI induction caused by hypoxia 2 and chemical agents 3 , when MSI induction happened 2 and 3 days after the treatment, respectively. Together, these data suggest the causal link between PP2A loss with MSI induction and its relevance to human CRC.
The reason that we did not use knockout to perform this experiment is that complete Aα loss has no transformative properties. Additionally, in cancer patients, Aα is found to be inactivated in a haploinsufficient manner. CRISPR/Cas9-mediated homozygous Aα deletion resulted in decreased colony formation and tumour growth across multiple colorectal and endometrial cancer cell lines. This study further uncovered a mechanism by which PP2A Aα regulates Aβ protein stability and activity and suggests why homozygous loss of Aα is rarely seen in cancer patients 28 .
2. Results of MSI analyses are presented poorly. Response: Thank you very much for the great comments. We have now provided clear MSI data in Fig. 2d. We divided the data of each marker into three panels and used different colors to clearly indicate the size change according to the colors used to mark the type of treatment. In addition, the scale unit has been clearly labelled on the tops of each marker.
3. Abstract, Page 2, line 32: "however, the mechanism of MSI status development is unclear". This statement disregards the history of MSI cancer research and the fact that genetic and epigenetic alterations responsible for the MSI phenotype have been largely clarified; according to some recent studies, epigenetic silencing of MLH1, somatic biallelic MMR gene mutations, and the combination of first and second hit of the same MMR gene in Lynch syndrome together can explain the vast majority of MSI cancers. The authors only present correlation data suggesting a link between PP2A and MSI, no evidence is presented that PP2A inactivation is responsible for the natural occurrence of the MSI phenotype in human cancers. The increased proportion of MSI tumours among endometrial cancers harboring PP2A, SET, and CIP2A mutations may be coincidental or related to the fact that MMR deficiency is generally associated with high mutation burden (so that MMR deficiency is cause, not consequence). Response: Thank you very much for the great comments. 1. We have revised the abstract as "Microsatellite-instable (MSI) tumours are one of the few cancers that respond to immune checkpoint blockade (ICB) with genetic and epigenetic alterations well clarified; however, the mechanism of MSI status development is not well understood." Please refer to Page 2, line 32 to 35 in the revised version. 2. Protein phosphatase 2A (PP2A) is a tumour suppressor that regulates many signaling pathways [29][30][31] , and its loss of function has been associated with cell transformation 32 . PP2A has been directly implicated in the negative regulation of double-strand break DNA repair proteins 33 . Consistent with the idea those protein phosphatases are not just negative regulators of DNA repair signaling, selective inhibition of PP2A activity impairs DNA repair 34,35 . PP2A has been suggested or confirmed to dephosphorylate over 300 substrates including MLH1, PMS1, and PMS2 36 . We also use experiments to prove that pp2a inhibitor can reduce MLH1 expression and MSI status (Fig. 5e, Fig. S11 and Fig. S12).
3. We further demonstrated the positive correlation between mutation count and CIP2A and SET mRNA levels (Unshown Figure. 1). The correlation coefficients were r = 0.13 (P = 0.003) and r = 0.18 (P < 0.00005) for CIP2A and SET, respectively.
Unshown Figure. 1. The mRNA levels of CIP2A and SET mRNA level (Y-axis) positively correlated with total mutation count (mean) (X-axis) for TCGA-colorectal cancer samples.
We used the classification and regression trees (CART), a powerful approach optimizing the cutoff point of independent variables for predicting dependent variable used in medical data sets 37 , to divide the CIP2A or SET data into high-level and low-level subgroups. The cut-off values of CIP2A and SET data were calculated as 440.953 and 11716.08, respectively. The percentages of the subgroups with high and low levels of CIP2A were 16.22% and 83.78%, respectively. In addition, the mutation count of the subgroup with high CIP2A was significantly higher than that of the subgroup with low CIP2A (P = 0.015) (Unshown Table 1). Similarly, the percentages of the subgroups with high and low levels of SET were 3.42% and 96.58%, respectively. Moreover, the mutation count in the high SET subgroup was higher than that in the low SET subgroup, although not significant. (Unshown Table 1). In the MSK-IMPACT cohort, including the clinical and genomic data of 1,661 advanced cancer patients treated with ICB 38 , tumours with PPP2R1A mutation accounted for 1.4%, which was associated with increased tumour mutation burden score and mutation count and better overall survival status (Fig. S15). Together, these data indicate that PP2A is not widely inactivated in colorectal tumours, and inhibition of PP2A may be a strategy for colorectal cancer treatment. The conclusion that PP2A mutation status may help to predict ICB therapy response based on these data is not justified. Response: Thank you very much for the great comments.

Unshown
A previous study analyzed the clinical and genomic data of 1,661 advanced cancer patients treated with ICB, whose tumours underwent targeted next-generation sequencing (MSK-IMPACT) 38 , and showed that higher somatic tumour mutational burden (highest 20% in each histology) can predict survival after immunotherapy across multiple cancer types. Given its clinical and practical importance, we have undertaken a reanalysis of these data and showed that PPP2R1A mutation (1.4%) was associated with an increased tumour mutation burden score and mutation count, and a better overall survival status ( Fig. S15a, b). Moreover, the median survival time and the univariate Cox regression hazard ratio of patients with PPP2R1A-mutated tumours were much better than those of patients with PPP2R1A-non-mutated tumours (Fig. S15c). The pan-cancer nature of this biomarker probably reflects the fundamental mechanisms by which ICB functions. These data also support the hypothesis that PPP2R1A, SET, and CIP2A mutations or altered mRNA levels are associated with higher mutation burden and MSI status and help predict response to ICB.
Minor points: 4. The authors use the cell lines SW-480 and SW-620 as examples for human CRC cell lines. Both cell lines are derived from the same patient and tumor, but derived at different time points. Therefore, one can expect that both lines will behave similarly regarding treatment response. What was the reasoning behind this selection?
Response: Thank you very much for your great comments. We apologize for this mistake and further explain the reason for using these two cell lines. Although SW480 (primary) and SW620 (lymph node) cell lines derived from primary tumour and metastasis from the same patient and carried identical mutation profiles, but had epigenetic differences 39 . According to your comments, we have also now repeated the use of PP2A inhibitors, not only LB-100 but also LB-102, to induce MSI in another MSS human CRC cell line, HT29. As shown in Fig. 5e, Fig. S11 and S12, treatment with LB100 or LB-102 induced MSI both at 2 and 7 days. (including CTLA-4, PD-1/PDL-1 and Combo group), whose tumours underwent targeted next-generation sequencing (MSK-IMPACT) 38 , and showed that higher somatic tumour mutational burden (highest 20% in each histology) can predict survival after immunotherapy across multiple cancer types. Given its clinical and practical importance, we have undertaken a reanalysis of these data and showed that PPP2R1A mutation (1.4%) was associated with an increased tumour mutation burden score and mutation count, and a better overall survival status (Fig. S15a, b). Moreover, the median survival time and the univariate Cox regression hazard ratio of patients with PPP2R1A-mutated tumours were much better than those of patients with PPP2R1A-non-mutated tumours (Fig. S15c). The pan-cancer nature of this biomarker probably reflects the fundamental mechanisms by which ICB functions. These data also support the hypothesis that PPP2R1A, SET, and CIP2A mutations or altered mRNA levels are associated with higher mutation burden and MSI status and help predict response to ICB. In response to your question about the combination of LB-100 and anti-CTLA4, we believe that the combination of LB-100 and anti-CTLA may improve the therapeutic effect.

Reviewer #2 (Remarks to the Author):
The study shows an interesting connection of PP2A inactivation to MLH silencing and MSI and demonstrates that PP2A knockout promotes the therapeutic response to ICB. (1) This study "found that PPP2R2B, encoding B55β, is the only subunit that is consistently downregulated or silenced in all examined CRC cell lines, but not in the normal colon mucosa samples". Specifically, Figure 1B in the Cancer Cell paper shows that B55β (PPP2R2B) is one of the few genes upregulated in primary tumour compared with normal tissue. However, PPP2R1A A subunit, PPP2CA C subunit, and other B subunits, such as PPP2R2A (B55α), PPP2R1B (PR65β), PPP2R3B (PR70), PPP2R5C (B56γ), PPP2R5D (B56δ), are obviously upregulated in primary tumour compared with normal tissue 4 . These data show that compared with normal tissues, the main A and C subunits and most of the B subunits of PP2A are up-regulated in primary tumours. Because RNAi against Aα of PP2A decreased total PP2A activity 5,6 , these data may imply an increase in PP2A activity in tumours 4 . However, this comparison was made between the primary tumour and each corresponding normal tissue, and has nothing to do with individual tumour differences. For example, SET was found overexpressed in 13 out the 21 tumour samples compared with corresponding normal tissue 7 , however, SET overexpression was only detected in 15.4% of 247 CRC patients without metastatic disease at diagnosis 8 . Need to be noted, the observed PP2A B subunit (PPP2R2B) silencing in the Cancer Cell paper was a result of comparison between the primary tumour and each corresponding normal tissue 4 .
(2) Instead, we used TGCA data to examine the correlations of different immune cell infiltration with PPP2R1A and endogenous PP2A inhibitors in different individual tumours. We found that the expression of endogenous PP2A inhibitors, CIP2A, and SET correlated positively with the infiltration of CD8+ T cells and CD20+ B cells and negatively with FOXP3+ Treg cells (Fig. 1c). Contrastingly, the expression of PPP2R1A correlated negatively with CD8+ T cells but positively with FOXP3+ Treg cells (Fig. 1c). Similar findings were observed in human rectal cancers (TCGA) (Fig. 1d).
(3) Moreover, long-term culture may induce changes in DNA methylation 9 , even though the cells studied here are human mesenchymal stem cells. Because PPP2R2B hypermethylation causes acquired apoptosis deficiency in activated T cells of systemic autoimmune diseases 10 , it is possible to induce cloned cells after long-term culture. Actually, this argument is supported by the data shown in the Cancer Cell paper 4 , where obvious PPP2R2B hypermethylation has been observed in CRC cell lines (Fig. 1E: HCT116, RKO….that only exhibited methylated band) compared to the primary CRC tumour samples (Fig. 1F: 2T, 4T, 12T…that exhibited both unmethylated and methylated bands). Therefore, the increase in PPP2R2B DNA methylation levels of CRC cell lines compared with the primary tumour cells can easily be explained as the contribution of long-term culture. As expected, frequent aberrant DNA methylation of PPP2R2B was observed in primary tumour tissues of ductal carcinoma in situ and early invasive breast cancer. PPP2R2B DNA has been shown to be hypomethylated in cancers with TP53 mutations 11 . These data suggest long-term culture induces changes in DNA methylation, which may help select cells with acquired apoptosis resistance Page 10 through PPP2R2B hypermethylation. (4) The survival and proliferation of established CRC cells (SW620 and HT29 lines) and primary human colon cancer cells can be suppressed by LB-100 that inhibited PP2A activity and activated AMPK signaling both in vitro and in vivo 12 . In addition, the self-renewal and sphere formation HT29 cell line and primary human colon cancer cells can be suppressed by silibinin 13 that inhibited the PP2Ac/AKT Ser473/mTOR pathway. Similarly, we used PP2A inhibitors, LB100 and LB102 to suppress PP2A activity (Fig. S10) and thereby induced MSI status (Fig. 5e and Fig.  S11, S12). Together, these data indicate that PP2A activity and its downstream pathways of CRC cell lines and primary human CRC cells can still be manipulated by agents that inhibit PP2A activity.
For the mechanistic study, the biochemical evidence showing HDAC2 and Rb1 are PP2A substrates are lacking. Although they are co-immunoprecipitated, it is necessary to demonstrate that they are the direct substrates of PP2A. PPP2R1a pulldown for an in vitro phosphatase assay using CT26 cell lysis as a substrate will substantiate the mechanistic claim. Response: Thank you very much for your great comments. We have now demonstrated that HDAC2 and Rb1 are the direct substrates of PP2A. CT26 cells were treated without (CTR) and with LB100, small-molecule inhibitor of PP2A, followed by PPP2R1a pulldown of the cell lysates for an in vitro phosphatase assay and western blotting. The data showed that the PPP2R1A pulldown in LB100-treated cell lysate exhibited decreased PP2A activity (Fig. S5a) and increased pRb1 and pHDAC2 levels (Fig. S5b). The biochemical evidence showing that HDAC2 and Rb are direct PP2A substrates. We have also revised manuscript as "To further demonstrate that Rb and HDAC2 were the direct substrates of PP2A, CT26 cells were treated without (CTR) and with LB100, a small molecule inhibitor of PP2A 14 , followed by Ppp2r1a pulldown of the cell lysates for an in vitro phosphatase assay and western blotting. The data showed that the Ppp2r1a pulldown in LB100-treated cell lysate exhibited decreased PP2A activity (Fig. S5a) and increased pRb and pHDAC2 levels (Fig. S5b). The biochemical evidence showing that HDAC2 and Rb1 are direct PP2A substrates." Please refer to Page 7 in line 159-164 in the revised version.
The final data showing PP2A inhibitor sensitizing PD1 therapy is not impressive. I find it is hard to appreciate the claim of using PP2A inhibitor as a strategy to induce MSI and then sensitize ICB. As PP2A is widely inactivated in human CRC, it is not convincing that inhibition of PP2A can be a strategy for cancer therapy. Response: Thank you very much for your great comment. We have used TCGA colorectal cancer samples (n=594) to demonstrate that compared with MSS cancers, MSI cancers have increased CIP2A and SET mRNA levels, but reduced PPP2R1A mRNA levels (Fig. 2i). We further demonstrated the positive correlation between mutation count and CIP2A and SET mRNA levels (Unshown Figure. 1). The correlation coefficients were r = 0.13 (P = 0.003) and r = 0.18 (P < 0.00005) for CIP2A and SET, respectively.
Unshown Figure. 1. The mRNA levels of CIP2A and SET mRNA level (Y-axis) positively correlated with total mutation count (mean) (X-axis) for TCGA-colorectal cancer samples.
We used the classification and regression trees (CART), a powerful approach optimizing the cutoff point of independent variables for predicting dependent variable used in medical data sets 37 , to divide the CIP2A or SET data into high-level and low-level subgroups. The cut-off values of CIP2A and SET data were calculated as 440.953 and 11716.08, respectively. The percentages of the subgroups with high and low levels of CIP2A were 16.22% and 83.78%, respectively. In addition, the mutation count of the subgroup with high CIP2A was significantly higher than that of the subgroup with low CIP2A (P = 0.015) (Unshown Table 1). Similarly, the percentages of the subgroups with high and low levels of SET were 3.42% and 96.58%, respectively. Moreover, the mutation count in the high SET subgroup was higher than that in the low SET subgroup, although not significant. (Unshown Table 1). In the MSK-IMPACT cohort, including the clinical and genomic data of 1,661 advanced cancer patients treated with ICB 38 , tumours with PPP2R1A mutation accounted for 1.4%, which was associated with increased tumour mutation burden score and mutation count and better overall survival status (Fig. S15). Together, these data indicate that PP2A is not widely inactivated in colorectal tumours, and inhibition of PP2A may be a strategy for colorectal cancer treatment. including RB E2F1 and many other known oncogenes. I am afraid that the therapeutic implication of using PP2A inhibitor in CRC has limited potential (though it has been reported before). Response: Thank you very much for your great comment. Although the safety of LB100 has partly addressed in the phase I-II clinical trial 14 , there is a concern that inhibiting PP2A, a tumour suppressor, can cause lots of oncogenic signaling. We have addressed the limitations of current study in the discussion section and proposed some tumour-targeting and controlled-release drug carrier systems, such as liposome or nanocages, can solve these limitations. We have now further addressed this issue in the discussion section. "To be noted, inhibiting PP2A, a tumour suppressor, can cause lots of oncogenic signals in normal tissues, thereby limiting the therapeutic potential of PP2A inhibition in cancer treatment. Similarly, this problem can be solved by controlled delivery of therapeutic agents to tumours." Please refer to Page 16 in line 368-370.

Unshown
Also, the study focuses on colorectal cancer and it is not clear why the in vivo studies also included the triple-negative 4T1 and pancreatic model. Response: Thank you very much for your great comment. Both of triple-negative breast cancer 32 and pancreatic cancer 49 have low incidence of dMMR/MSI-H tumours and poorly respond to monotherapy with antibodies against PD1 or PD-L1 50,51 .
However, immune checkpoint blockade is a FDA-approved tissue-agnostic drug for the treatment of MSI-high solid tumours 52 . Pancreatic cancer was also included in 12 tumour types in the cohort of 86 MSI-high tumour patients and responded to anti-PD1 monotherapy 53 . We therefore expanded the application of current results to the treatment of triple-negative 4T1 and pancreatic model. Our results show that the combined use of LB100 in 4T1 and Pan-18 tumour models increased the Page 13 sensitivity of ICB by increasing the mutation burden and inducing MSI status ( Fig. 5a and Fig. S8).

Reviewer #3 (Remarks to the Author):
The antigenicity and hence the immunogenicity of tumours is likely to be a major limiting factor in response to immunotherapy. Altering immunogenicity is a major challenge and this study is potentially important because it attempts to rise to this challenge by evaluating a pathway which may give rise to new antigens.
Whilst the experiments examining the impact of targeting PP2A on MSI status are compelling, the experiments examining effects on the immune response fall short of providing definitive answers. As such, the data as it stands is over-interpreted. Response: Thank you very much for your great comments. We have answered the following questions, which will help solve the problems you raised here.
The histology shown in Figure 1 needs some improvement. It is difficult to understand why no CD4+ cells are observed when Foxp3+ cells are seen? What are the cells stained in the lower right panel?
Response: Thank you very much for your great comments. We apologize for this mistake. The original figure for Foxp3 was misplaced with picture taken from distal small intestine that was filled with Treg attracted by CCR5 15 . The other figures in Fig.  1a were from colon (large intestine). In the revised Fig. 1a, all pictures were taken from colon. The new figures show that very few CD4+ or Foxp3+ cells were observed in the histology of control tissues (Fig. 1a). The cells stained in the lower right panel are CD20+ B cell aggregates that were only observed in the tissue of ppp2r1a loss tumours. Tumour-infiltrating B cells have recently been identified as cellular components of tertiary lymphoid structures (TLSs) in the tumour, which are associated with better respond to immunotherapy 19 . The CD20+ B cell aggregates have been identified in melanoma and sarcoma (three papers side by side published in Nature), which are associated with an increased chance that patients' tumours would respond to immunotherapy [16][17][18] . We further chose some figures published in the Nature papers and merged into the following figure for your reference. In studies of human CRC (and other cancers) CD8+ T cells and Tregs are normally reported as positively correlating? This is because Tregs are induced when there is an immune response to suppress. This does not appear to be the case in the analysis shown here. Could the authors comment?
Response: Thank you very much for the great comments. We used TCGA colorectal cancer data that mainly include the transcriptome of the tumour per se. In other word, we analyzed the intratumoural CD8+ T cells and Tregs. In fact, the ratio of intratumoural CD8+ T/FOXP3+ cells in different colorectal tumours is very different, which is a predictive marker for the survival of colorectal cancer patients [54][55][56] . Intraepithelial lymphocytes were defined as lymphocytes located within tumour cell nests that may be the area used for transcriptome analysis. It has been noted that CD8 expression is detected in the epithelium in 100% of the cases, while FOXP3 expression is not or only sporadically present in the tumour epithelium 55 . Therefore, the lack of a positive correlation between CD8+ T cells and Tregs in our data may be due to the specimens used for analysis.
The tumour growth curves in Figure 4 show significant differences however these are only assessed for a short period of time (up to day 21 to cells transfected with control shRNAs (WT CT26) (Fig. 4c). Although WT CT26 did not respond to anti-PD1 treatment (Fig. 4d), CT26 with Ppp2r1a knockdown responded to anti-PD1 treatment (Fig. 4e), suggesting Ppp2r1a knockdown sensitises CT26 to anti-PD1 treatment (Fig. 4e). Moreover, increased levels of CD8+ tumour-infiltrating T cells were found in tumours formed by CT26 cells with Ppp2r1a knockdown compared to those formed by WT CT26 cells (Fig. 4f).
We further provided the detailed data of CD8+ tumour-infiltrating T cells in tumours formed by CT26 with Ppp2r1a knockdown at different time-points from 21 to 35 days, when the tumours were found to become larger. After immunofluorescence staining, we processed the images to analyse the numbers of positive signals using TissueQuest software (TissueGnostics) 20 . In order to normalise for tumour size, we also divided the numbers of CD8+ tumour-infiltrating T cells by the tumour weights ( Fig. 4f Right) 21,22 . However, these data did not show that smaller tumours had more lymphocytes/ g tumour compared to larger tumours.
It is impossible to conclude that there are more neoantigens generated in these tumours without exome sequencing. This is the key missing piece of data. The number of TCRs alone is not sufficient evidence of neoantigen-driven clonal expansions.
Response: Thank you very much for the great comments. Although there is a plethora of neoantigen discovery pipelines based on genetic information to predict epitopes, the current pipelines are human-centered and are therefore mainly designed for clinical use. Recently, NAP-CNB 27 , a novel bioinformatic pipeline, has been developed to directly estimate H-2 peptide ligands from murine tumour samples, and its area under the curve (AUC) is equal to or better than the state-of-the-art methods. Moreover, this pipeline also has a neural network model of the binding affinity prediction function. We therefore used the NAP-CNB 27 pipeline to identify potential tumour neoantigens. The detailed method has been added in the "Materials and methods" section. We have also modified the manuscript as below: "To demonstrate that Ppp2r1a knockdown converted cold tumours into hot tumours by increasing neoantigen, we submitted the RNA-seq data of CT26-shppp2r1a and CT26-scr tumour samples, integrated in the fastq.gz files, and applied the NAP-CNB 27 to predict neoantigens. A total of 270 missense transcripts, corresponding to 220 genes, shared by three CT26-shppp2r1a tumours but not found in the CT26-scr tumour were identified (Fig. 4g). The software also generated a ranking of putative neoantigens that are common in the three CT26-shppp2r1a tumours samples. The 30 top-scoring putative neoepitopes are shown in Table S5." Please refer to Page 9 in line 193-200 in the revised version.

Page 16
The data with the small molecule inhibitor is less compelling. Also, despite the authors' claim that there is an effect beyond an impact of Tregs; this is not proven by the experiment carried out as it does not include the use of the PI-3065 plus anti-PD1 alone as a control. In addition, the authors should note that PI-3065 has effects beyond just direct effects on Tregs. It also affects effector T cells directly as well as monocytes. This experiment needs a re-think to include all necessary controls as well as a "cleaner" method of targeting Tregs. Response: Thank you very much for the great comments. We have now added the use of PI-3065 plus anti-PD1 alone as a control. The data showed that there is no significant difference between PI-3065 and PI-3065 plus anti-PD1 (Fig. 5b).
We have also strengthened the rationales of using PI-3065 to block mouse Treg-mediated immunosuppression.
(1) The reason we did not use Foxp3-mutant scurfy mice or Foxp3-null mice t mice for this study is that these mice suffered from the lethal lymphoproliferative autoimmune syndrome and become moribund at approximately 4 weeks of age 23 . Although depletion of Treg cells by neonatal thymectomy, adoptive transfer of naive T cell samples depleted of Treg cells into lymphopenic hosts or treatment of mice with antibodies specific for CD25, results in a much milder and more slowly progressing disease 24 , however, Treg cells are also critical in self-tolerance and prevent catastrophic autoimmunity throughout the lifespan of mice. We therefore gave up using the methods mentioned above. (2) Instead, we chose to use p110δ inactivation that has been successfully demonstrated to block Treg-mediated immune suppression in mice carrying solid tumours 25 . Notably, long-term administration of PI-3065, a small molecule inhibitor with selectivity for p110δ, to mice was well tolerated and did not induce weight loss 25 . (3) There are concerns that the inactivation of p110δ in Treg cells will indirectly release CD8 cytotoxic T cells and induce tumour regression, and the inactivation of p110δ will also block the intrinsic immunosuppression of PMN-MDSCs (Ly6G high ), leading to reduced tumour growth. However, it has also been reported that inhibiting p110δ in cancer might impair cytotoxic T cells and negatively impact on cancer immune surveillance 26 . Previous data 25 show that although p110δ blockade reduces the effectiveness of cytotoxic T cells, it also overrides Treg-and probably also MDSC-mediated suppression of anti-tumour immune responses, enabling even weakened CTLs to successfully attack tumours. Thus, p110δ is apparently more essential for Treg rather than effector T-cell responses Page 17 against cancer cells.
(4) To show specific blocking of Treg by using the p110δ inhibitor PI-3065, we first demonstrated that p110δ was only expressed in Treg (Foxp3+), but not expressed in CD8+ or PMN-MDSCs (Ly6G high ) in the CT26 tumour microenvironment. We then used PI-3065 to block regulatory T cell-mediated immune suppression in mice 25 , and thereby we could study the effect of anti-PD1 plus PP2A inhibition on tumour killing without Treg interference. "To prove that LB100 sensitised tumour cells to ICB therapies regardless of its Treg inhibitory activity, we first showed the expression of p110δ in Treg (Foxp3+), but not in CD8+ or polymorphonuclear myeloid-derived suppressor cell (PMN-MDSC) (Ly6G high ) in the CT26 tumour microenvironment (Fig. S9). We then used the p110δ inhibitor PI-3065 to block Treg-mediated immune suppression in mice 25 , and showed that the therapeutic effects of the combination of LB100 and anti-PD1 on reducing tumour growth and enhancing survival were also observed in the presence of PI-3065 (Fig. 5b)." Please refer to Page 10 in line 222-229 in the revised version.
again (that is what we mean "double-checked") and found out that when we extended the exposure time from 4 sec to 20 sec (data were shown below), we found that E2F1 knockdown has a more pronounced effect on the change in E2F1 level, and it has a stronger correlation with changes in downstream gene levels. The left panel was shown in the revision version, while the right panel was shown in the original version.
5. Prior to the use of immune checkpoint blockade, MSI-H/dMMR was historically associated with poorer outcomes in advanced (stage IV) colorectal cancer (which is inverted compared to its association with favorable outcomes in early stage I-III disease). In light of this, the statements on page 3 line 54-55 and 64 ("High incidence of somatic mutations can lead to MSI tumours of a less aggressive nature") are not entirely accurate and should be revised.
Response: Thank you very much for your great comments. We have now revised the Main Text as "MSI is associated with better stage-adjusted prognosis in early stage I-III colorectal cancer 6 and response to immune checkpoint blockade (ICB) 7 than microsatellite-stable (MSS) tumours, leading to the urgent need to investigate the mechanisms causing MSI tumour development." Please refer to Page 3 in Line 56-59.
We have also deleted "High incidence of somatic mutations can lead to MSI tumours of a less aggressive nature".
6. The authors claim that PPP2R1A, SET, and CIP2A mutations "help to predict responses to ICB" is not strongly supported by the results. Namely, univariate Cox regression analysis of survival differences between PPP2R1A-mutated and PPP2R1A-non-mutated tumor is insufficient and lacks adjustment for other prognostic / predictive factors. MSI status and tumor mutation burden (TMB) are themselves predictive biomarkers, but the authors haven't demonstrated that PPP2R1A-mutation status is independent of these.
Response: Thank you very much for your great comments. Based on your comment, we have checked the correlation between patient survival and all parameters of the cohort. Among them, total mutation burden (TMB) and PPP2R1A mutation were found to significantly increase the survival rate of patients and reduce the hazard ratio (HR). We then compared the numbers of TMB between PPP2R1A-mutated and PPP2R1A-non-mutated tumors. Besides, we further compared the numbers of TMB between tumors with and without mutation in several driver mutations, such as TP53, PIK3CA and KRAS. We found that TMB was significantly higher in PPP2R1A-mutated than PPP2R1A-non-mutated tumors (p = 0.00026). TMB was also significantly higher in TP53 (p = 2.92x10 -6 ) and PIK3CA (p = 0.007) mutation groups than non-mutation groups (Unshown Table 1). There was a tendency for TMB in KRAS mutated tumors to be higher than in KRAS non-mutated tumors. These data indicate that tumors with high TMB are more likely to have some key driver mutations than tumors with low TMB. When performed univariate Cox regression analysis of survival differences between gene-mutated and gene-non-mutated tumors, we found that only PPP2R1A mutation reduced the HR to 0.4296 (p =0.03), while TP53 and PIK3CA mutations significantly increased the HR to 1.473 and 1.31, respectively (Unshown Table 2).
There was a tendency for KRAS mutation to increase HR to 1.31. These data indicate that, except for PPP2R1A, most single-gene mutations did not reduce HR, and cannot be used as "favorable prognostic markers" to help predict the response to ICB (Unshown Figure 1). We further performed adjustment for PPP2R1A mutation and other prognostic / predictive factors. Nevertheless, we also found that the PPP2R1A mutation reduced the HR to 0.6142, although the p-value is not significant (Unshown Table 3).
In addition, we showed in mouse tumor models that loss of PPP2R1A led to increased TMB and MSI status, and in mouse and human tumor cell models, PP2A inactivation also led to loss of MLH1 and induces MSI status (Fig. 5). Therefore, this manuscript can suggest a causal relationship between the PPP2R1A mutation and TMB.