High-dose oncogenic PIK3CA drives constitutive cellular stemness through self-sustained TGFβ pathway activation

Oncogenic PIK3CA mutations activate phosphoinositide 3-kinase-alpha (PI3Kα) and are among the commonest somatic mutations in cancer. We recently demonstrated that the “hotspot” variant PIK3CAH1047R exerts striking allele dose-dependent effects on stemness in human pluripotent stem cells (hPSCs), and found multiple oncogenic PIK3CA copies in a substantial proportion of human cancers. This suggested that the consequences of oncogenic PI3K signaling may differ according to the strength of genetic PIK3CA activation. Here, to identify the stemness-promoting mechanism, we profiled isogenic wild-type, PIK3CAWT/H1047R and PIK3CAH1047R/H1047R iPSCs by high-depth transcriptomics, proteomics and reverse-phase protein arrays (RPPA). We report that the phenotypic switch in homozygous PIK3CAH1047R hPSCs occurs downstream of signaling “rewiring” towards self-sustained TGFβ pathway activation and increased NODAL expression, which was no longer reversible by pharmacological PI3Kα inhibition. Gene expression analysis of PIK3CA-associated human breast cancers in The Cancer Genome Atlas revealed increased expression of NODAL according to tumor stage and PIK3CAH1047R allele dosage. Together with the emerging link between NODAL re-expression and cancer aggressiveness, our data suggest that TGFβ pathway inhibitors warrant investigation in breast tumors stratified by PIK3CAH1047R allele dosage. One-sentence summary Biallelic genetic PI3Kα activation rewires signaling and induces constitutive stemness downstream from self-sustained TGFβ pathway activation.


127
Collectively, these findings corroborate the existence of a threshold of PI3K pathway activity which 128 determines the large majority of gene expression changes in PIK3CA H1047R/H1047R iPSCs in a near-binary manner.

129
While higher depth of sequencing did reveal statistically significant gene expression changes in heterozygous iPSCs, 130 effect sizes were modest and more variable. Similar findings in heterozygous MEFs suggest that this may be 131 generalizable to differentiated cell types, irrespective of species.

132
To assess whether transcriptional changes observed in iPSCs were mirrored in the proteome, we applied 133 label-free proteomics to the iPSC lines used in our first study (5). Around 4,600 protein ratios were obtained for both 134 heterozygous versus wild-type and homozygous versus wild-type iPSC comparisons, as estimated using a novel

145
Altogether, these data consolidate the view that only homozygosity for PIK3CA H1047R results in robust and 146 widespread transcriptional changes in otherwise normal, diploid cells, arguing against the "butterfly" effect of 147 heterozygosity suggested in prior studies using a genetically abnormal breast epithelial cell line (25,26

150
We previously demonstrated a graded increase in AKT (S473) phosphorylation across heterozygous and 151 homozygous PIK3CA H1047R iPSCs (5). To assess in more detail whether the near-binary gene expression difference 152 between heterozygous and homozygous PIK3CA H1047R cells is underpinned by corresponding differences in indices of PI3K pathway activation, we profiled phosphorylation of a wider repertoire of pathway components using reverse 154 phase phosphoprotein arrays (RPPA). To mimic the physiological environment of the pluripotent epiblast, we 155 studied cells in growth factor-replete conditions.

156
Observed changes in protein phosphorylation were surprisingly modest, with the maximal change a two-fold 157 increase in AKT phosphorylation (on S473 and T308) in PIK3CA H1047R/H1047R cells. Contrasting with the near-binary 158 response seen at the transcriptional level, heterozygous and homozygous PIK3CA H1047R expression generally 159 produced graded phosphorylation of PI3K pathway components, with slightly higher levels in homozygous iPSCs 160 ( Fig. 2A).

161
None of the mutant genotypes showed consistently increased mTORC1-dependent phosphorylation of 162 P70S6K or its downstream substrate S6 ( fig. S3A), perhaps reflecting saturation of this level of the pathway due to 163 the presence of the additional stimuli in the medium (e.g. amino acids). When deprived of growth factors for 1 h 164 prior to RPPA profiling, both heterozygous and homozygous mutant did exhibit increased P70S6K phosphorylation, 165 whereas S6 phosphorylation remained similar to wild-type cells (Fig. 2B). Inhibition of PI3Kα activity with low-166 dose BYL719 for 24 hours fully reversed canonical PI3K signaling-related changes in phosphorylation of 167 downstream proteins including AKT, GSK3, FOXO1, TSC2 and P70S6K (Fig. 2B). Other changes in protein 168 phosphorylation remained unaffected in PIK3CA H1047R/H1047R hPSCs, including increased SMAD2 and ERK1/2 169 phosphorylation as well as increased expression of c-MYC and IGF1R (Fig. 2B, fig. S3B). This suggests signaling 170 rewiring that is, unexpectedly, resistant to acute inhibition of the primary inducing event.

172
Pathway and network analyses implicate TGF signaling in PIK3CA H1047R dose-dependent stemness 173 Pathway and network analyses were next applied to proteomic and transcriptomic data in an attempt to 174 identify candidate mechanism(s) mediating PIK3CA H1047R dose-dependent stemness. Consistent with our previous 175 study, in which we showed strong induction of the TGF family member NODAL (5), TGF1 was again the most 176 significant predicted upstream activator according to Ingenuity ® Pathway Analysis (IPA) of the top 2000 upregulated 177 and top 2000 downregulated transcripts in PIK3CA H1047R/H1047R iPSCs (Fig. 3A). TGF1 was also the most significant 178 predicted upstream activator revealed by analysis of the PIK3CA H1047R/H1047R proteomic dataset (Fig. 3B). These data 179 strongly suggest activation of the TGF pathway in homozygous PIK3CA H1047R iPSCs.
Although PIK3CA WT/H1047R iPSCs showed around 10-fold fewer differentially-expressed genes than 181 homozygous iPSC cells, IPA for upstream activators in the heterozygous iPSCs also revealed multiple TGF 182 pathway-related stimuli among the most significant predicted activators (Fig. 3C). Moreover TGF1 was predicted 183 as one of only two differential upstream activators when analysis was performed on genes that were concordantly 184 differentially expressed (n = 180) in PIK3CA H1047R mutant iPSCs versus wild-type controls ( Fig. 3C and table S8).

185
The other significant upstream regulator common to both heterozygous and homozygous PIK3CA H1047R was 186 MAPK1 (encoding ERK2), which is consistent with RPPA as well as immunoblot evidence of increased ERK 187 kinase phosphorylation in PIK3CA H1047R mutant iPSCs (Ref. (5) and schematic in Fig. 2A). Although TGF 188 signaling-related predictions held in heterozygous PIK3CA H1047R iPSCs, the significance of the effect (overlap p-189 value = 1.74e-05) was much lower than in homozygous (overlap p-value = 4.25e-21) mutants. This points towards 190 a critical role for the TGF pathway in mediating the allele dose-dependent effect of PIK3CA H1047R in human iPSCs.

191
To complement the IPA analysis, which is based on highly curated, proprietary datasets, we undertook 192 non-hypothesis-based Weighted Gene Correlation Network Analysis (WGCNA)a network-based data reduction 193 method that seeks to determine gene correlation patterns across multiple samples, irrespective of the function of 194 individual genes (30). Using all transcripts expressed across wild-type and both heterozygous and homozygous 195 PIK3CA H1047R iPSCs (Fig. 4A), this analysis returned a total of 43 clusters (or modules) of highly interconnected 196 genes (Fig. 4B). Of the two modules with the highest correlation to the homozygous trait, one showed enrichment 197 for several KEGG pathway terms of relevance to the stemness phenotype of PIK3CA H1047R/H1047R iPSCsmost 198 notably "Signaling pathways regulating pluripotency in stem cells" (Fig. 4C). Given prior predictions of strong 199 activation of TGF signaling in homozygous mutants, we next constructed the minimal network of differentially 200 expressed genes in PIK3CA H1047R/H1047R iPSCs that linked the pluripotency, PI3K and TGF signaling pathways 201 within this network module (Fig. 4D). The resulting network exhibited high interconnectivity, with multiple shared 202 nodes across all three pathways, suggesting close crosstalk between PI3K and TGF signaling in stemness 203 regulation. Finally, the fact that most of the network nodes represented genes with increased expression in 204 homozygous mutants strengthens the notion that strong oncogenic PI3Kα activation stabilizes the pluripotency 205 network in human iPSCs.

207
Inhibition of TGF signaling destabilizes the pluripotency gene network in PIK3CA H1047R/H1047R iPSCs 208 TGF signaling plays a critical and well-established role in pluripotency regulation (20,23,31), and a 209 differentiation-resistant phenotype has previously been reported in NODAL-overexpressing hPSCs (22). These 210 observations, together with increased NODAL expression in homozygous PIK3CA H1047R iPSCs and computational 211 identification of enhanced TGF pathway activity in PI3K-driven "constitutive" stemness, led us to hypothesize that 212 strong PI3Kα-dependent activation of NODAL expression underlies the establishment of the differentiation-resistant 213 phenotype of homozygous mutants through stabilization of the pluripotency gene network. Specifically, we 214 hypothesized that NODAL elicits autocrine enhancement of TGF signaling in PIK3CA H1047R/H1047R iPSCs, with 215 downstream promotion of NANOG expression ultimately "locking" the cells in a state of perpetual stemness.

216
Testing this hypothesis in hPSCs is challenging for both biological and technical reasons. These include 217 lack of specific pharmacological NODAL inhibitors and the difficulty in detecting the subtle early phenotypic 218 consequences of partial destabilization of the hPSC pluripotency gene regulatory network. Moreover, the widely-219 adopted maintenance medium and coating substrate we used for cell culture both contain TGF ligands (32, 33), 220 which may mask effects of NODAL repression by PI3Kα-specific inhibition. To minimize these confounders, we 221 prepared maintenance medium with and without recombinant NODAL supplementation, and assessed expression 222 of NODAL itself and NANOG as surrogate markers of stemness over 72 h of culture.

223
Within 48 h, exclusion of NODAL from the medium resulted in the expected downregulation of NODAL 224 and NANOG expression in wild-type iPSCs, and this was greater still at 72 h ( Fig. 5 and fig. S5A). In contrast,

225
NODAL removal had no effect on the increased NODAL and NANOG expression in PIK3CA H1047R/H1047R iPSCs ( 228 consistent with NODAL's known ability to control its own expression through a feedforward loop (34). Despite a 229 55% reduction in NODAL mRNA after 72 h, however, little effect on NANOG expression was seen (Fig. 5). The

235
Thus, to confirm that TGF signaling is required for maintenance of stemness in PIK3CA H1047R/H1047R iPSCs, 236 we treated cells with SB431542a potent and specific inhibitor of TGF and NODAL type I receptors (35). This 237 resulted in acute and complete repression of NODAL expression within 24 h, accompanied by downregulation of 238 NANOG expression (Fig. 5). A similar effect was observed on POU5F1 (OCT3/4) expression, consistent with 239 destabilization of the pluripotency gene regulatory network in PIK3CA H1047R/H1047R iPSCs (Fig. 5). Confirming this,

245
Collectively, these data strongly suggest that the stemness phenotype of PIK3CA H1047R/H1047R iPSCs is 246 mediated by self-sustained TGF signaling, most likely through PI3K dose-dependent increase in NODAL 247 expression, and is amenable to reversal through pharmacological inhibition of the TGF pathway but not of PI3Kα.

249
Stage II human breast cancers with multiple PIK3CA H1047R alleles exhibit increased NODAL expression 250 It has long been known that dedifferentiating tumor cells re-express embryonic markers (37), and that the 251 TGF pathway promotes cancer stem cell maintenance, metastasis and drug resistance (38-40). The strong link 252 between homozygosity for PIK3CA H1047R and TGF signaling in human iPSCs prompted the question of whether 253 this mechanism may be of relevance in human PIK3CA-associated cancers. Specifically, NODAL re-expression has 254 previously been linked to aggressive features such as invasion and metastasis in a range of cancers (reviewed in Refs.

259
comparing tumors with multiple PIK3CA H1047R copies to those with a single copy.
PIK3CA H1047R was detected in 108 BRCA samples, 57 of which had more than one copy of the mutant 261 allele. Of the remaining 51 samples with a single PIK3CA H1047R allele, 10 harbored a second PIK3CA mutation and 262 were thus grouped with cancers with multiple PIK3CA H1047R copies (giving rise to a total of 67 samples classified as 263 'multiple'). MYC mRNA levels were robustly detected in both sample groups (Fig. 6A), consistent with ubiquitous 264 expression. In contrast, mRNA expression of the embryonic markers NODAL, NANOG and POU5F1 was low (< 265 1 count per million) (Fig. 6A), as expected if breast cancer stem cell-like cells only comprise a small proportion of 266 the bulk tumor tissue (46). The low expression notwithstanding, there was evidence of increased expression of MYC, 267 POU5F1 and NODAL in several samples with multiple PIK3CA H1047R copies (Fig. 6A).

268
Given that re-expression of NODAL has been linked with aggressive breast cancer (42-45), we next 269 stratified the samples according to tumor stage and assessed the expression of the four stemness markers (Fig. 6B).

270
Stage I denotes breast cancer that has not spread outside the breast and exhibits little or no involvement of proximal 271 lymph nodes, whereas stage II and III tumors exhibit progressively higher spreading to lymph nodes (47). Stage IV 272 denotes breast cancer that has spread to other organs (47). We excluded stage IV samples from the analysis due to 273 low sample size (2 and 1 in the "single" and "multiple" PIK3CA H1047R groups, respectively). Notably, in stage II 274 breast tumors, for which the sample size was sufficient, NODAL expression increased as a function of PIK3CA H1047R 275 allele dosage (Fig. 6B). This trend was apparent despite the genetically heterogeneous nature of the samples and

282
The so-called "hotspot" variant PIK3CA H1047R is the most common activating PIK3CA mutation in human 283 cancers and in a group of largely benign overgrowth disorders known as PROS (6). We recently found that PIK3CA-284 associated cancers often harbor multiple mutant PIK3CA copies, and demonstrated that homozygosity but not 285 heterozygosity for PIK3CA H1047R leads to self-sustained stemness in human pluripotent stem cells (hPSCs) (5). Using 286 a combination of computational analyses and targeted experiments, the current study adds further support for the 287 existence of a PI3K signaling threshold in cells with oncogenic PI3Kα activation. We provide evidence for self-sustained TGF pathway activation as the main mechanism through which PIK3CA H1047R homozygosity "locks" 289 hPSCs in a differentiation-resistant state that becomes independent of the driver mutation and the associated PI3K 290 pathway activation.

291
High-depth transcriptomics confirmed that heterozygosity for PIK3CA H1047R fails to induce widespread 292 substantial transcriptional remodeling, whether chronically modelled in CRISPR-edited iPSCs or acutely induced in 293 mouse embryonic fibroblasts (MEFs) by Cre expression, despite induction of canonical PI3K pathway activation in 294 both cases (current study and Ref. (5,28,48)). Similarly, iPSCs with heterozygous expression of PIK3CA E418K , a 295 "non-hotspot" mutation, were transcriptionally indistinguishable from their isogenic wild-type controls. In contrast 296 to the mild transcriptional consequences of heterozygous PIK3CA H1047R expression, homozygosity for PIK3CA H1047R 297 was associated with differential expression of nearly 1/3 of the hPSC transcriptome. This observed near-binary 298 response is not a consequence of a similar quantitative difference in PI3K pathway activation assessed by 299 phosphoprotein profiling, which instead showed a relatively modest and graded increase in homozygous versus 300 heterozygous PIK3CA H1047R hPSCs. This suggests that the apparent sharp PI3K signaling threshold that determines 301 the cellular response in hPSCs hPSCs is "decoded" distal to the canonical pathway activation.

302
The increased expression of several proteins in homozygous PIK3CA H1047R iPSCs, and their resistance to 303 downregulation in response PI3Kα-specific inhibition with BYL719, further suggested that the phenotype of these 304 cells may have become partially uncoupled from the oncogenic trigger event. In systems biology terms, the near-305 binary response observed with two versus one copy of PIK3CA H1047R , and the inability to reverse the phenotype 306 upon inhibition of the primary PI3K signaling defect, is consistent with a non-linear network topology characterized 307 by bistability and an "all-or-nothing" response. This is usually mediated by indirect or direct positive feedback loops 308 (49) such as those known to occur between the TGF pathway and the pluripotency network (9, 21, 34).

309
The well-known dose-dependent effects of TGF signaling in a developmental context (34) helps explain 310 our previous observations of allele dose-dependent stemness effects of PIK3CA H1047R in hPSCs (5). Specifically, our 311 data suggest a model in which homozygosity but not heterozygosity for PIK3CA H1047R promotes NODAL expression 312 and thus increases TGF pathway activity to a level that is sufficient for increased NANOG expression and 313 stabilization of the stem cell state (Fig. 6C), but insufficient to tip the balance towards mesendoderm differentiation 314 (5). Such an indirect effect of PI3K activation on the pluripotency network in hPSCs is in line with the lack of warranted to determine whether PI3K-dependent regulation of NODAL expression contributes to its developmental 317 functions in vivo, which include roles in gastrulation and establishment of the body axes (34). It is notable that 318 patients with a germline activating AKT2 mutation exhibit overgrowth that is predominantly left-sided (50), yet it 319 remains unclear whether this is linked to differential regulation of NODAL expression.

320
Since specific pharmacological inhibition of NODAL is not possible, our strongest evidence for its role 321 in the stemness phenotype of PIK3CA H1047R/H1047R hPSCs comes from the ability of these cells to sustain increased 322 expression of key stemness genes without exogenous NODAL supplementation. Moreover, the ability of SB431512 323 to collapse the stemness gene signature in PIK3CA H1047R/H1047R hPSCs unequivocally demonstrates that sustained 324 TGF pathway activation is required to maintain their stemness phenotype. Exactly how PI3K activation regulates 325 NODAL expression remains to be determined. A potential mechanism involves increased expression of the stem cell 326 reprogramming factor MYC, which was observed both at the mRNA and protein level in homozygous but not 327 heterozygous PIK3CA H1047R iPSCs. Furthermore, MYC was the only node in the WGCNA-based network of 328 pluripotency, PI3K and TGF pathway components that was classified as a member of all three pathways. Finally,

329
increased MYC expression has previously been shown to exert oncogenic effects that depend on a sharp threshold 330 of MYC expression, reminiscent of the effects we observe for PIK3CA activation (51). As the first specific 331 pharmacological inhibitor of MYC was recently reported (52), its use in the PIK3CA H1047R hPSC system may provide 332 additional insight into the potential role of MYC as a molecular link between oncogenic PI3Kα activation, NODAL 333 expression and stemness.

334
The close relationship between PI3K and TGF pathway-driven stemness has potentially important 335 implications for understanding PIK3CA-driven cancer. Breast cancers are enriched for PIK3CA mutations, and 336 studies in mice have revealed that PIK3CA H1047R induces multipotency in differentiated mammary cells (53, 54).

337
Conversely, advanced cancers typically exhibit increased phenotypic plasticity and stemness, which is closely linked 338 to drug resistance and the ability of cancer cells to undergo epithelial-mesenchymal-transition (EMT) and

347
We now report that increased NODAL expression is discernible even in the context of cellularly 348 heterogeneous human breast cancers, with the degree of upregulation related to the number of oncogenic 349 PIK3CA H1047R alleles present. This suggests that our findings in hPSCs have potential clinical relevance (Figure 6c).

350
Monotherapy with pan-or isoform-specific PI3K inhibitors has had modest success in cancer (60), and it is thus 351 noteworthy that PI3Kα-specific inhibition with BYL719 was unable to reverse the stemness phenotype of 352 PIK3CA H1047R/H1047R iPSCs. Systematic studies in a larger number of breast cancer samples and breast cancer cell 353 lines with defined PIK3CA H1047R copy number are needed to confirm the phenomenon that we have observed in 354 human iPSCs. Nevertheless, based on our current findings of increased NODAL expression in some human breast 355 cancers with multiple PIK3CA H1047R copies, we believe that TGF pathway inhibition is worthy of further 356 exploration in this setting. Moreover, as induction of pro-tumorigenic TGF signaling has been reported in a range 357 of human cancers (24, 61), the allele dose-dependent effects of PIK3CA H1047R may extend beyond breast cancer.

358
Accordingly, it has previously been reported that PIK3CA amplification leads to TGF pathway-dependent EMT 359 in a mouse model of oral carcinogenesis, with additional evidence that the degree of PIK3CA amplification correlates 360 with tumor stage in human head and neck squamous cell carcinoma (62).

361
Finally, our study demonstrates the power of network-based approaches to study the highly context-362 dependent complexity of cell signaling and to identify potential therapeutic targets beyond those suggested by simply 363 considering the presence/absence of a specific genetic defect (63-65). Given the dose-dependent effects of both 364 TGF and PI3K pathway activation, a comprehensive understanding of their concerted actions will hinge upon 365 adoption of network-based analyses and mathematical modeling capable of capturing context-dependent 366 relationships in response to qualitative and quantitative signaling differences. This may result in tailored treatment 367 strategies that take into account PIK3CA allele dose-dependent signaling reconfiguration in individual tumor cells.

381
All cell lines were tested negative for mycoplasma and genotyped routinely to rule out cross-382 contamination during prolonged culture. Short tandem repeat profiling was not performed. All 383 experiments were performed on cells within 10 passages since thawing.

385
Collection for RNA sequencing and total proteomics 386 For RNA sequencing and total proteomics, subconfluent cells were fed fresh E8/F 3 h prior to 387 snap-freezing on dry ice and subsequent RNA or protein extraction. Relative to the results in Ref. (5), 388 the current transcriptomic data of PIK3CA H1047R were obtained more than 6 months following the first 389 study, with cells at different passages, and were thus independent from one another. Moreover, sample 390 collection for the second transcriptomics experiment was conducted over three days according to a 391 block design, thus allowing us to determine transcriptional differences that are robust to biological 392 variability.

394
Cell lysate collection for RPPA For RPPA in growth factor-replete conditions, cells were fed fresh E8/F 3h before collection.

396
To assess variability due to differences in collection timing, clones from each iPSC genotype were 397 collected on each one of three days according to a block design, giving rise to a total of 22 cultures. To 398 test the effect of the PI3Kα-specific inhibitor BYL719, cells were treated with 100 nM drug (or DMSO 399 only as control treatment) for 24 h and exposed to growth factor removal within the last hour before

444
Correlations between corresponding transcriptomics and/or proteomics data were calculated 445 using Spearman's rank-order correlation test for non-normally distributed data.

493
In order to find non-reference, mutated peptides in the MS data, we increased the search FASTA 494 file with mutations affecting the protein sequence, as detected by WES with a high sensitivity filter:

503
Methionine oxidation, protein N-terminal acetylation and serine/threonine/tyrosine phosphorylation 504 were set as variable modifications and cysteine carbamidomethylation was set as a fixed modification.

505
False discovery rates were set to 1% and the "match between runs" functionality was activated. We 506 filtered out peptides that were associated with multiple identifications in the MaxQuant msms.txt file, 507 had a score < 40, were identified in the reverse database or came from known contaminants. Analysis 508 of the observed peptides passing these filters was performed using a Monte Carlo Markov Chain model 509 as described previously (29). Briefly, the model predicted the average ratio (sample versus control) of

522
The protein lysates were processed for slide spotting and antibody incubations as described 523 previously (77). Briefly, a four-point dilution series was prepared for each sample and printed in 524 triplicate on single pad Avid Nitrocellulose slides (Grace Biolabs) consisting of 8 arrays with 36x12 525 spots each. Next, slides were blocked and incubated in primary and secondary antibodies. The processed 526 arrays were imaged using an Innopsys 710 slide scanner. Non-specific signals were determined for each slide by omitting primary antibody incubation step. For normalization, sample loading on each array 528 was determined by staining with Fast Green dye and recording the corresponding signal at 800 nm.

529
Details for all primary and secondary RPPA antibodies are included in Supplementary Key Resources 530 Table. 531 Data analysis 532 Slide images were analyzed using Mapix software (Innopsys), with the spot diameter of the 533 grid set to 270 μm. Background signal intensity was determined for each spot individually and 534 subtracted from the sample spot signal. A test for linearity was performed from the four-point dilution 535 series, according to a flag system where R 2 > 0.9 was deemed good, R 2 > 0.8 was deemed acceptable 536 and R 2 < 0.8 was poor (excluded from subsequent analyses). Median values from the four-point dilution 537 series were calculated for each technical replicate and normalized to the corresponding Fast Green value 538 to account for differences in protein loading. For each sample and protein target, a mean expression 539 value was calculated from the remaining technical replicates and normalized to the corresponding mean 540 of the wild-type group. All phosphoprotein signals were also normalized to the corresponding total 541 protein values.

542
A statistical test for differential expression was performed on datasets with more than three 543 samples per group, using the limma package to apply the limma-trend method with lmFit() and 544 eBayes(), specifying collection time as blocking factor (67). Phosphoprotein and total protein lists were 545 processed separately. The associated p-value for assessment of differential gene expression was 546 adjusted for multiple comparisons with the Benjamini-Hochberg method at FDR < 5% (70). The 547 function duplicateCorrelation() was applied to correct for the use of replicate iPSC clones on the same 548 day. Heatmaps were generated using the heatmap.2() function within the gplots package in R, using

571
All qPCR data were acquired on a Quant Studio™ 5 Real-Time PCR System (Thermo Fisher 572 Scientific). The thermocycling conditions (SYBR Green reactions) were as follows (ramp rate 1.6C/s 573 for all): 50C for 2 min, 95C for 10 min, 40 cycles at 95C for 15 sec and 60C for 1 min, followed by 574 melt curve analysis (95C for 15 sec, 60C for 1 min, and 95C for 15 min with ramp rate 0.075C/sec).

575
The TaqMan hPSC Scorecard thermocycling conditions were as specified by the manufacturer in the 576 accompanying template.

577
All relevant primer sequences are included in Supplementary Key Resource Table. 578

580
The list of differentially expressed total proteins (PIK3CA H1047R/H1047R vs wild-type) was 581 subjected to IPA (build version: 448560M; content version: 36601845) against the Ingenuity 582 Knowledge Base, considering only relationships where confidence was classified as "Experimentally 583 Observed". Following exclusion of chemicals and drugs, the Upstream Regulators list was used for 584 generation of Volcano plots of the respective activation z-scores and overlap p-values.

585
IPA was also used to analyze the lists of differentially expressed genes in both heterozygous

598
The IPA Upstream Regulator Analysis is based on the proprietary Ingenuity Knowledge Base 599 which is used to compute two scores based on user-specified data: an enrichment score (Fisher's exact 600 test p-value) that measures overlap between observed and predicted regulated gene sets; a z-score that 601 assesses the match between observed and predicted up/down regulation patterns (78). The results of the 602 Upstream Regulators Analysis were extracted for downstream Volcano plotting of overlap p-values and 603 associated activation z-scores. Note that for heterozygous PIK3CA H1047R iPSCs, a bias-corrected 604 activation z-score was used for plotting to take into account any bias arising from a larger number of 605 upregulated vs downregulated genes in these cells.

607
Weighted Gene Correlation Network Analysis (WGCNA) 608 RNA sequencing counts from all 12 samples were converted to reads per kilobase million 609 (RPKM). A threshold of 10 RPKM was used to filter out low-expression genes, followed by removing

614
To identify modules associated with homozygosity for PIK3CA H1047R , we used the correlation between a 615 gene's module membership (eigengene) and significance for differential expression in homozygous

662
Table S1. List of differentially expressed genes in PIK3CA WT/H1047R vs wild-type hPSCs after applying an absolute 663 fold-change cut-off of minimum 1.3.                   1235 Key Resources Table. 1236 1237 1238