key: cord-0264758-ol7isvg8
authors: Baldwin, Louise A.; Bartonicek, Nenad; Yang, Jessica; Wu, Sunny Z.; Deng, Niantao; Roden, Daniel L.; Chan, Chia-Ling; Al-Eryani, Ghamdan; Swarbrick, Alexander; Junankar, Simon
title: DNA barcoding reveals ongoing immunoediting of clonal cancer populations during metastatic progression and in response to immunotherapy
date: 2021-01-11
journal: bioRxiv
DOI: 10.1101/2021.01.11.426174
sha: 800a48f681d4c548ac475733b5bd75fe499b7e33
doc_id: 264758
cord_uid: ol7isvg8

Cancers evade the immune system in order to grow or metastasise through the process of cancer immunoediting. While checkpoint inhibitor therapy has been effective for reactivating tumour immunity in some cancers, many solid cancers, including breast cancer, remain largely non-responsive. Understanding the way non-responsive cancers evolve to evade immunity, what resistance pathways are activated and whether this occurs at the clonal level will improve immunotherapeutic design. We tracked cancer cell clones during the immunoediting process and determined clonal transcriptional profiles that allow immune evasion in murine mammary tumour growth in response to immunotherapy with anti-PD1 and anti-CTLA4. Clonal diversity was significantly restricted by immunotherapy treatment at both the primary and metastatic sites. These findings demonstrate that immunoediting selects for pre-existing breast cancer cell populations, that immunoediting is not a static process and is ongoing during metastasis and immunotherapy treatment. Isolation of immunotherapy resistant clones revealed unique and overlapping transcriptional signatures. The overlapping gene signature was predictive of poor survival in basal-like breast cancer patient cohorts. Some of these overlapping genes have existing small molecules which can be used to potentially improve immunotherapy response.

All cancers must find ways to evade the immune system so that they can continue 37 to grow (1). Previous studies have established that this occurs through a process 38 called immunoediting (2). During immunoediting, the more immunogenic cancer 39 cells are selectively eliminated by the immune system thus leaving behind less 40 immunogenic cancer cells that are then free to expand. Immunoediting can occur 41 through multiple mechanisms, these include the elimination of cells with strong 42 immunogenic mutations, leading to the loss of neo-antigens (3) or the selection of 43 cells with elevated expression of various immunosuppressive programs (4). 44

Immunotherapies look to overcome some of the immune evasion pathways 45 established by the cancer cells to avoid recognition and elimination by the 46 immune system during immunoediting. The prominent clinically approved 47

immunotherapies for solid tumours target T cell checkpoint molecules (eg. anti-48 2 CTLA-4 and anti-PD1) to overcome T cell exhaustion (5, 6). In select cancer types 49 such as melanoma, these drugs have dramatic effects in a large proportion of 50 patients (7). Unfortunately for metastatic breast cancer early clinical trials have 51 seen few patients experiencing durable responses even in the most sensitive 52

basal-like subtype of breast cancer (8). This indicates that in metastatic breast 53 cancer, resistance can rapidly develop to anti-PD1/PDL1 therapy and suggests 54

that alternate immune drug targets are needed for breast cancer. 55 While the immune system is known to play a role in breast cancer outcome (9) 56 and recent evidence has indicated that immunoediting can occur in a transgenic 57 mouse model of breast cancer (10), very little is known about the specifics of 58 immune evasion by breast cancer cells. The majority of studies examining this 59 phenomenon were performed in the highly mutated MCA carcinogen driven 60 sarcoma model, and could not track the response at a clonal level (11). Of interest, 61 a recent study suggested that immunoediting by T cells can occur at the clonal 62 level, by demonstrating the selection of clones that contain less-immunogenic 63 fluorophores (12). This leaves an important gap in our collective knowledge as to 64 the mechanisms employed in less immunogenic tumours such as breast. 65 NK cells and T cells have both been demonstrated to play a role in immunoediting 66 (13, 14) . However, the majority of recent research has focussed on pathways 67 relevant to T cell recognition (15) (16) (17) (18) (19) . Downregulation of MHC is one mechanism 68 by which cancer clones become impervious to T cells (20), but this inherently 69 makes them targets of NK activity. In breast cancer NK dysfunction is noted and 70 this is regulated by microenvironmental factors (21). Data on resistance pathways 71 that allow for immune evasion from both T cells and NK cells are currently more 72 limited. 73

To understand the process of immunoediting in breast cancer we conducted DNA 74 barcoding (22, 23) of murine EMT6 and 4T1 mammary carcinoma cells, which 75

were introduced into both immune competent and immunocompromised mice 76 (Fig 1) . DNA barcoding allows tracking of individual clones and clonal expansion 77 while avoiding introduction of potentially immunogenic proteins (24). This 78 system allowed us to analyse immunoediting in vivo at primary and metastatic 79 sites and to study whether resistance to checkpoint immunotherapy developed 80 from pre-existing or de novo generated cell populations. We show that 81 immunoediting is initiated by the endogenous immune system in primary 82 tumours and that cancer cells undergo a second round of immunoediting during 83 metastasis. We further observed that immunoediting of specific clones is 84 enhanced at both sites by checkpoint immunotherapy. We also identified cancer 85 cell clones highly resistant to immunotherapy. RNA-sequencing (RNA-Seq) 86 analysis of these resistant clones demonstrated that each clone had activated 87 unique immune evasion pathways, with one downregulating MHC-I expression 88 and another upregulating PD-L1. However, these clones also contained a common 89 gene expression signature that is highly predictive of poor survival in both the 90 METABRIC and TCGA basal-like breast cancer cohorts. This study has thus 91 determined the patterns of immunoediting at a clonal level, that metastatic cells 92 undergo a second round of immunoediting, and that immunotherapy significantly 93 restricts clonal diversity. We go on to determine unique pathways activated in 94

immune evasive cancer cells that are targetable and could improve 95 immunotherapy response in breast cancer patients. 96

To understand the role of the immune system and immunotherapy in shaping the 99 clonal dynamics of cancer cells within primary tumours we used the 100 immunotherapy-sensitive syngeneic mammary carcinoma model EMT6 (25). The 101

ClonTracer DNA barcode library (22) was introduced into the EMT6 cells resulting 102 in ~41 000 unique barcodes identified by DNA sequencing. Following inoculation 103 of 250 000 cells (~6 fold over representation of each barcode) into the mammary 104 fat pad we compared the number of clones able to engraft and grow in immune-105 competent, syngeneic wild-type (WT) Balb/c mice or severely 106

immunocompromised NOD SCID Gamma (NSG) mice (Fig 2A) . 107

Tumour growth was much more rapid in the NSG mice with tumours reaching 108 ethical endpoint on day 14 post-transplant, whereas wild-type mice all reached 109 ethical endpoint by day 23 (Fig 2B) . This led to NSG mice having significantly 110

shorter overall survival (median survival 14 days) when compared to wild-type 111 mice (median survival 22 days, Mantel-Cox p=0.009), demonstrating that the 112 immune system plays an important role in controlling primary tumour growth in 113 the EMT6 model ( Fig 2C) . 114

To examine the influence of immunotherapy on tumour growth & clonal dynamics, 115

we compared wild-type mice treated with combination immunotherapy (anti-PD1 116 + anti-CTLA4) or control antibodies starting from day 10 when tumours were 117 approximately 200mm 3 (Fig 2A) . All control mice reached ethical endpoint by day 118 23 (Fig 2D) . In contrast, all treated tumours regressed following treatment, with 119 50% relapsing and reaching ethical endpoint between days 46 and 54 ( Fig 2D) . 120

The remaining immunotherapy treated mice remained tumour free when the 121 experiment was terminated on day 60. Kaplan-Meier analysis demonstrated that 122 immunotherapy significantly increased survival with median survival increasing 123 from 22 days to 57 days (Mantel-Cox p=0.0006) (Fig 2E) . 124

To determine if immune control of tumour growth was driven at a clonal level, we 125 examined the number and distribution of barcodes present in primary tumours. 126 We found that at ethical endpoint the tumours grown in NSG mice had over 50 127 times the number of unique barcodes as tumours grown in control WT mice 128 (p=0.0002, unpaired t-test), which in turn had more than 20 times the number of 129 unique barcodes found in immunotherapy treated WT mice (p=0.0019, unpaired 130 t-test) (Fig 2F) . We applied Shannon diversity analysis to these samples to 131 understand how the immune system influenced the diversity of barcodes. 132

Shannon diversity index is determined by how evenly distributed the barcodes are 133 within a population and is only moderately influenced by barcode number. 134

Analysis of barcode diversity revealed a trend to a reduction in barcode diversity 135 in the tumours from the control Balb/c mice compared to the NSG mice, whereas 136

there was a dramatic reduction in barcode diversity following immunotherapy 137 treatment (p<0.001 unpaired t-test) when compared to the control treated mice 138 ( Fig 2G) . This data indicates that a subset of EMT6 cells are more resistant to the 139 endogenous immune system but that this selection does not skew the evenness of 140 the barcode distribution dramatically. This suggests that the clones that are 141 resistant to the immune system all have a similar level of resistance. In contrast 142

immunotherapy applies a much more stringent bottleneck that only a limited 143 4 number of clones can readily overcome and with a high variability in the levels of 144 resistance. Further analysis identified specific EMT6 clones that were 145

reproducibly enriched across multiple replicate mice following immunotherapy 146 treatment indicating that they had a pre-existing resistance phenotype that was 147 being positively selected for ( Fig 2H) . 148

To determine whether immunoediting continued during metastatic dissemination 150 and whether specific metastatic clones were enriched or depleted, we turned to 151 the highly metastatic 4T1 mammary carcinoma model, as the EMT6 cell line is 152 poorly metastatic (26 resection was equivalent between the groups (Sup Fig 1A) . Adjuvant 162 immunotherapy with combination anti-PD1 + anti-CTLA4 led to a modest but 163 significant increase in survival (37.5 days) versus control treated mice (33 days; 164 p=0.0121, Mantel-Cox Log-rank test) ( Fig 3C) . 165

We then examined whether the endogenous immune system shaped metastatic 166 clonal dynamics. While primary tumours contained similar numbers of clones and 167 barcode diversity in NSG and WT hosts ( Fig. 3D and Sup Fig 1B, C) , metastatic 168 lungs of NSG mice contained ~3 times as many barcode clones as WT controls (Fig  169  3D ). We next sought to determine if the increase in survival following 170 immunotherapy ( Fig. 3C ) was associated with alterations in clonal dynamics. As 171 the treatment was only given after excision of the primary tumour, the 172 immunotherapy would only affect the outgrowth of cancer cells that had already 173 metastasised to the lung. Despite only observing a modest increase in survival 174

following combination immunotherapy (Fig 3C) , we observed a 70% reduction in 175 the number of clones able to form metastases ( Fig 3E) . 176

The increase in barcode number in the lungs of NSG mice was associated with a 177 significant increase in diversity as measured using the Shannon diversity index 178 ( Fig 3F) . This shows that the endogenous immune system restricts the number 179 and skews the diversity of metastatic clones that can reach and outgrow in the 180 lungs. In addition to the reduction in barcode number following immunotherapy 181 treatment, we also saw a significant reduction in barcode diversity as measured 182

using the Shannon diversity index (Fig 3F) . This indicates that immunotherapy is 183 leading to the immunoediting of specific clonal cell populations over others. 184

To further understand the key immune cell types that control clonal outgrowth in 185 the metastatic lungs we depleted either CD8 T cells (anti-CD8) or NK cells (anti-186 asialo-GM1) in wildtype mice starting one day prior to tumour resection. Neither 187 treatment led to a significant change in overall survival (Sup Fig 2A) . However, 188 depletion of either cell type led to an increase in the number of clones detected 189 within the lungs, reaching statistical significance with T-cell depletion ( Fig 3G) . These results were reproducible, as demonstrated by a second pool of barcoded 193 4T1 cells containing a larger barcode library (300 000 barcodes). Following the 194 injection of 50 000 barcoded cells we recovered approximately 10 000 -12 000 195 barcode sequences from each primary tumour and this was relatively unchanged 196 in the NSG mice (Sup Fig 3A) . This suggests that roughly a fifth of the injected cells 197 are able to engraft and grow in the mammary gland. As clone diversity was 198 unchanged in NSG mice this confirms that the immune system does not play a 199 major role in restricting growth of 4T1 carcinoma cells in the primary tumour 200

setting. In contrast when we examined the number of clones that had spread to 201 the lungs of NSG mice we again found approximately 3 times as many when 202 compared to the wild-type mice (Sup Fig 3A) . In addition, we similarly saw a ~3-203

fold reduction in barcode diversity in response to immunotherapy (Sup Fig 3B) . 204

Due to the high complexity of the 300 000 barcode library each mouse received 205 only a partially overlapping complement of barcode clones. 206

To better understand how specific clonal cell populations responded to the 208 immune system and immunotherapy we combined the barcode frequencies from 209 the two datasets utilising the 5000 barcode library (WT vs NSG and Control vs 210 Immunotherapy). We performed unsupervised hierarchical clustering of these 211 samples and selected barcodes that were observed at greater than 5% frequency 212 in any one sample. We found that the primary tumours from the two experiments 213 cluster together irrespective of the immune status of the mouse (Balb/c or NSG), 214

further suggesting that 4T1 cells do not undergo immunoediting at the primary 215 site ( Fig 4A) . In contrast, lung tumours formed in the NSG hosts did not cluster 216 with lung tumours formed in Balb/c lungs, with the immunotherapy treated 217 samples mostly clustering alone or with metastases formed in WT mice. A number 218 of specific barcodes were enriched in the metastatic lungs of all the NSG mice 219

indicating these clones were highly metastatic ( Fig 4A) . This agrees with the 220 findings of Wagenblast and colleagues (23). 221

We identified a number of barcodes that had striking patterns of enrichment or 222 depletion in response to the immune system and immunotherapy, we replotted 223 these using a dot plot ( Fig 4B) . As these barcodes were enriched or depleted in a 224

reproducible manner across replicate mice this suggests these are due to inherent 225 features of these clones. Firstly, there are three barcodes that were enriched 226 within the NSG lungs (NSG1-3), observed at lower abundance in the untreated 227 wild-type mice, and that were completely eliminated following immunotherapy 228 treatment. This suggest that while these clones are highly metastatic in the 229 absence of an immune system, they are immunogenic and are thus subjected to 230 immunoediting in WT mice particularly following immunotherapy. Another group 231 of metastatic clones that were present in the lungs from NSG and WT mice were 232 further enriched following immunotherapy (IE1-2). These immunotherapy 233 enriched clones were detected in the lungs of all six replicate mice. With the 234 dramatic reduction in the number of barcodes present following immunotherapy, 235 the odds ratio of this happening by chance is 0.0034 (95% confidence interval: 236 0.0010-0.0079; chi square p value: 3.67x10 -251 ). This suggests that these clones 237 6 have a pre-existing resistance phenotype and are positively selected for following 238 immunotherapy. 239

To further analyse how specific barcodes were enriched in lung metastases 240 following immunotherapy we visualised the top nine clonal populations (based on 241 average barcode proportions in the metastatic lungs) and generated fish-plots. 242

These showed that different clones were preferentially enriched in the lungs of 243 NSG mice when compared to WT mice ( Fig 4C) . Furthermore, we observed that a 244 small subset of clones were highly enriched in the lungs of immunotherapy treated 245 mice ( Fig 4D) . 246

To understand more about the phenotype of these immunotherapy resistant 248

clones we established clonal cell populations from two of them (designated IE1  249 and IE2), and two independent control clones (NT1 and NT2) that were not 250 enriched following immunotherapy (3-4 independent clonal cell lines were 251 generated per barcode). These clonal cell lines were isolated from the parental 252 barcoded 4T1 cell population purely in vitro with no additional selective 253 manipulation. The barcode within each of these clonal cell lines was confirmed to 254 be correct using Sanger sequencing. All four clonal cell lines had similar growth 255 kinetics in vitro indicating no proliferative advantage of the immune evasive 256

clones in vitro (Sup Fig 4A) . 257

To identify the barcode integration sites and determine whether the clones 260 contained large scale genomic alterations we performed whole genome 261 sequencing (WGS) at around 30x coverage of the clones. The WGS analysis 262 determined the precise genomic location at which the barcodes integrated (Sup  263  Table 1 ). IE1 integrated in the intergenic region between Kpna2 and Smurf2 and 264

IE2 within an intron of Nrf1, neither integration site changed the coding sequence 265 of these genes. Copy number analysis determined that no clone demonstrated 266 dramatic copy number changes when compared to the other clones. Each clone 267

only contained a small number of single copy number gains and losses (IE1 only 6 268

CNVs and IE2 only 5 CNVs), with clone NT2 showing the greatest number of CNVs 269 at 41 (summarised in Sup Table 2 ). We found one locus with a single copy number 270 gain in both IE1 and IE2 that led to 3 copies of the genes Nc3r1 (the Glucocorticoid 271 receptor) and Arhgap26 a Rho GTPase that associates with focal adhesion kinase 272 (FAK), however, this gain was also present in the NT2 clone. These results 273 demonstrate that large scale genomic changes likely do not play a large role in the 274 various phenotypes of the different clones but suggest that copy number changes 275 may be selected against during immunoediting. 276

To investigate the mechanism of immune evasion by these clones we performed 278

RNAseq analysis and compared the two immunotherapy resistant clones to the 279 bulk 4T1 population. The IE1 clone had 1553 differentially expressed genes (Log 280 fold change >2 and FDR p<0.05) with 478 significantly upregulated and 1075 281 significantly downregulated ( Fig 5A) . The IE2 clone had 1099 differentially 282 expressed genes with 375 significantly upregulated and 724 significantly 283 downregulated ( Fig 5B) . The non-target clones had fewer gene expression 284 changes compared to the bulk 4T1 population with NT1 having 621 and NT2 285 having only 262 differentially expressed genes. We examined the top differentially 286 expressed genes between each of IE1 and IE2 with the parental 4T1 cells, 287 however, we did not find any with an obvious role in immune evasion (Sup Tables  288  3 & 4) . Gene set enrichment analysis revealed that two of the top ten gene sets 289

were upregulated in both IE1 and IE2 (CHEN_HOXA5_TARGETS_9HR_UP, 290 BLUM_RESPONSE_TO_SALIRASIB_UP), while most were unique to either cell line 291

(Sup Tables 5 & 6) . Hoxa5 is a known tumour suppressor gene in breast cancer 292

(27) and although we see an enrichment of its target genes, the expression of 293

Hoxa5 itself was significantly reduced in the IE1 clone and there was a trend to 294 reduced expression in the IE2 clone. There was no overlap in the top ten down 295 regulated gene-sets between IE1 and IE2. The top downregulated gene-set for IE1 296

was the REACTOME_UB_SPECIFIC_PROCESSING_PROTEASES gene-set, that 297

contained two genes involved in antigen processing for display by MHC-I (Psmb8 298

and Psmb9). As down regulation of the MHC-I pathway is a common mechanism 299 of immune evasion we investigated this in more detail. 300

Through this analysis we found that the IE1 clone had significantly reduced 301 expression of many genes related to antigen presentation including MHC-I (H2-302 k1), Tap2, Psmb8, Psmb9 and Psmb10 (Fig 5C) . H2-k1 is the main MHC molecule 303 expressed by the Balb/c strain of mouse that the 4T1 carcinoma cell line was 304 derived from. We validated the reduction in MHC-I expression levels seen in the 305 RNAseq data at the protein level using flow cytometry ( Fig 5D) . This analysis 306

showed that the IE1 clone had significantly reduced cell surface MHC-I protein 307

expression compared to the bulk 4T1 population. We thus examined the WGS data 308 and this showed that the loss of MHC-I expression in IE1 was not due to genomic 309 loss at the MHC locus on chromosome 17 (Sup Fig 5A. In contrast the IE2 clone 310

had elevated levels of a number of these MHC related genes (Fig 5C) , in addition 311 to a non-classical MHC molecule H2-t23 ( Fig 5E) that is known to negatively 312 regulate NK cells through their inhibitory receptor Nkg2a (28). Interestingly IE2 313 cells also demonstrated a significantly increased expression of the T cell inhibitory 314 molecule Cd274/PD-L1 ( Fig 5E) . This again was validated at the protein level using 315 flow cytometry (Fig 5F) . These results demonstrate that each of the two 316 immunotherapy resistant clones are phenotypically unique. 317

We next examined whether the copy number changes or barcode integration sites 318 identified above impacted gene expression. In IE1 and IE2 the copy number 319 changes in Nc3r1 (the Glucocorticoid receptor) and Arhgap26 were associated 320 with significantly increased expression of these genes but NT2 demonstrated no 321 change in expression (Sup Fig 5B) . Elevated Nc3r1 expression has been associated 322

with poor prognosis and metastasis in TNBC although whether it plays a role in 323 immune evasion is not known (29, 30). As stated above the barcode for IE1 324 integrated in the intergenic region between Kpna2 and Smurf2 and IE2 within an 325 intron of Nrf1. The expression of Smurf2 was the only of these genes that was 326 significantly altered with a modest log fold increase of 0.59 in IE1. 327

Demethylating agents such as 5-aza-2'-deoxycytidine (5-aza) are known to 329 upregulate MHC-I expression in cancer cells (31), thus we treated our clonal cell 330 8 lines utilising 5-aza for 72 hours to determine whether DNA methylation was a 331 mechanisms suppressing MHC expression in the IE1 clone. Using flow cytometry, 332

we observed that MHC-I expression was elevated in a dose dependent manner 333

following 5-aza treatment in all clones. However, MHC-I expression in the IE1 334 clone was consistently lower than the parental 4T1 cell line at all doses of 5-aza 335 (Sup Fig 6A) . This indicates that gene hyper-methylation is not the mechanism of 336 MHC-I suppression in the IE1 clone. 337 IFN-gamma stimulation is another mechanism by which MHC expression can be 338

increased on cancer cells. The IE1 clone responded to IFN-gamma treatment by 339 upregulating MHC-I expression but again it remained suppressed compared to the 340 parental 4T1 cells (Sup Fig 6B) . This suggests these cells broadly retain the 341 transcriptional regulatory machinery that is required to upregulate MHC-I in 342

response to IFN-gamma stimulation. These results indicate that MHC 343 downregulation is likely regulated by epigenetic factors other than DNA 344 methylation and that the majority of MHC expression in this clone can be restored 345 by IFN-gamma treatment. 346

We had noted that the GSEA analysis showed some overlap in enriched gene-sets 348 between the two immune evasive clones (IE1 and IE2), we thus reasoned that as 349 well as having unique immune evasion features these clones may have some 350 pathways in common. To identify the common immune evasion pathways being 351 initiated by both the immunotherapy resistant clones we generated Venn 352 diagrams to identify overlapping gene expression changes (Sup Table 7 ). This 353 analysis demonstrated that the immunotherapy resistant clones had more gene 354 expression changes in common with each other and less in common with the 355 control non-target clones (Fig 6A) . We generated a heatmap of the top 50 356 upregulated and downregulated genes across all the samples ( Fig 6B) and 357

performed GSEA analysis using C2 on the longer list (Sup Fig 7A, Sup Table 8 ). 358

Only two gene sets had significant p values when you consider multiple testing, 359 these were the HOXA5 gene set mentioned previously and a COVID19 related gene 360 set. Although not significant there were several additional COVID19 related gene 361 sets from the same recent publication identified in the overlapping upregulated 362 gene list suggesting an immune related role of these genes (32). 363

Therefore we wanted to understand the role of these genes in patients, so we 364 generated signatures from the top 25 upregulated and downregulated genes that 365 had human orthologs and were detectable in both the METABRIC (33) and TCGA 366 datasets (34). We then analysed the role of these signatures in survival using the 367 basal-like breast cancer patients from these cohorts. While the patients in these 368 cohorts have not been treated with immunotherapy it has previously been 369 demonstrated that immune features such as the number of tumour infiltrating 370 lymphocytes or regulatory T cells influence prognosis in basal-like breast cancer 371 patients (35). When we analysed overall survival in these cohorts we observed 372 that the upregulated signature associated with substantially poorer outcome in 373 both cohorts (METABRIC: p=0.0043, HR=2.0, Fig 6C; TCGA: p=0.042, HR=4.3, Fig  374  6D ). In contrast the downregulated gene signature did not show any significance 375 in either cohort (data not shown). We generated heatmaps with unsupervised 376 clustering to determine whether specific genes or groups of genes from the 377 signature were specifically driving the association with survival (Sup Fig 7B) . In 378 the TCGA data we observed a number of clusters that seemed to associate more 379 with survival, one of these included the genes FAM71F2, MASP2, HLF, PPP1R15A, 380

MMP23B and LIMS2. Interestingly one of these genes Ppp1r15a also known as 381 GADD34 has previously been demonstrated to be critical in blocking immunogenic 382 cell death following chemotherapy, and when it was inhibited the chemotherapy 383 response was improved in immunocompetent mice but not immunocompromised 384 mice (36). A second group of genes included SEZ6L2 and this gene has been 385 associated with survival in a number of cancers but not through an immune 386 related mechanism (37, 38). There was no enrichment of proliferation or invasion 387 gene sets in our GSEA analysis suggesting that these processes were not behind 388 the poor outcome of patients that highly express genes in the common 389 upregulated signature. 390

Previous studies have shown that CTL infiltration correlates with survival in 391

basal-like breast cancer so it is possible our signature is a surrogate measure of T 392 cell infiltration. To test this, we performed TIDE analysis (39) on the TCGA and 393 METABRIC cohorts followed by correlation analysis between the CTL signature 394 score and our upregulated immune evasion signature score. This showed no 395 correlation in the METABRIC cohort and only a weak negative correlation in the 396 TCGA cohort (Sup Fig 7C) demonstrated the utility of immunotherapies (46). A further limitation is that the 520 integration of the barcode and selection markers into the genome and the 521 potential immunogenicity of the RFP could affect the phenotype of the cancer cells. 522

In previous studies we and others have found some fluorophores and luciferase to 523 be immunogenic and negatively affect tumour growth and metastasis in the 4T1 524 model (24, 47) . However, in this study we found that tumour growth and 525 metastasis were unaffected by RFP expression. While the introduction of the DNA 526 barcodes could have influenced the phenotype of the specific clones we feel that 527 this is unlikely, with no dramatic impact on the expression of the genes closest to 528 the integration site. Furthermore, none of the genes associated with each 529 integration site came out as being significantly involved in cancer cell evasion of 530 CD8 T cell responses in a recent CRISPR screen (15). 531

Overall this study has demonstrated that immunoediting occurs at the clonal level 532

in primary tumours and that a second round of immunoediting occurs during 533 metastasis. Immunotherapies dramatically enhanced immunoediting, however, 534

pre-existing resistant populations were still responsible for relapse. The large 535 reduction in clonal diversity following immunotherapy in the 4T1 model, a model 536 that is known to be poorly responsive to immunotherapy, suggests that slight 537 improvements through combination therapy could eliminate the remaining clones 538

and lead to dramatic improvements in survival. By isolating immunotherapy 539 resistant clones and phenotyping them, we identified common and distinct 540 immune evasion pathways. We anticipate through the targeting of the pathways 541 identified in this study, in particular the common pathways, it will be possible to 542

further reduce the numbers of resistant clones and improve the efficacy of 543

immunotherapies. Cellular DNA barcoding 566

The ClonTracer library was a gift from Dr Frank Stegmeier (Addgene #67267). 567

Lentiviral particles containing the high-complexity barcode library were 568 produced by transfecting 293T cells. 4T1 and EMT6 cancer cell lines were 569 barcoded by lentiviral infection using 0.8 µg/ml polybrene. Cells from each line, 570 13 were infected with a target MOI of 0.1, corresponding to 10% infectivity to ensure 571 single lentiviral integration. Cells that received a barcode were then sorted based 572 on the RFP reporter protein using a BD FACSAriaII, these cells were then expanded 573 and frozen into a number of aliquots for the subsequent experiments. 4T1 cells 574

were generated with two different barcode complexities, one with ~5000 575 barcodes (4T1 BC5000) and one with ~300 000 barcodes. 576 577

Mice 578

All animal experiments were approved by the Garvan Institute of Medical 579

Research/St. Vincent's Hospital Animal Experimentation Ethics Committee. 580

Immunocompetent BALB/c mice and immunocompromised NOD.Cg-Prkdc scid 581

Il2rg tm1Wjl /SzJ (NSG) mice aged 6-to-8 weeks were obtained from Australian 582

BioResources (Moss Vale, Australia) and housed at the Garvan Institute of Medical 583

Research. 584 585

In vivo tumour growth 586

For IE1_1  IE1_2  IE1_3  IE2_1  IE2_2  IE2_3  NT1_1  NT1_2  NT1_3  NT1_4  NT2_1  NT2_2  NT2_3 

Hallmarks of Cancer: The Next 759 Generation

Cancer Immunoediting: Integrating 761 Immunity's Roles in Cancer Suppression and Promotion

Evolution of Metastases in Space and Time under 764 Immune Selection

MYC regulates the antitumor immune response through 766 CD47 and PD-L1

The blockade of immune checkpoints in cancer 768 immunotherapy

A. Ribas, Primary, Adaptive, and 770 Acquired Resistance to Cancer Immunotherapy

The 772 hallmarks of successful anticancer immunotherapy

Atezolizumab and Nab-Paclitaxel in Advanced Triple-775 Negative Breast Cancer

Patterns of 778 Immune Infiltration in Breast Cancer and Their Clinical Implications: A 779 Gene-Expression-Based Retrospective Study

Immunosurveillance and immunoediting in MMTV-782

PyMT-induced mammary oncogenesis

Adaptive immunity maintains occult cancer in an 785 equilibrium state

The immune system profoundly restricts intratumor genetic 787 heterogeneity

Cancer exome analysis reveals a T-cell-dependent 789 mechanism of cancer immunoediting

Natural killer cell-mediated immunosurveillance of 791 human cancer

Functional genomic landscape of cancer-intrinsic 793 evasion of killing by T cells

In vivo CRISPR screening identifies Ptpn2 as a cancer 795 immunotherapy target

Genome-Wide CRISPR Screen Identifies Regulators of 797

Mitogen-Activated Protein Kinase as Suppressors of Liver Tumors in Mice

CRISPR-GEMM Pooled Mutagenic Screening Identifies 800 KMT2D as a Major Modulator of Immune Checkpoint Blockade

Identification of essential genes for cancer immunotherapy. 803

Human NK Cell Education by Inhibitory Receptors for MHC WANG_RESPONSE_TO_GSK3_INHIBITOR_SB216763_UP