Analysis of genetic signatures of tumor microenvironment yields insight into mechanisms of resistance to immunotherapy

Background Therapeutic intervention targeting immune cells have led to remarkable improvements in clinical outcomes of tumor patients. However, responses are not universal. The inflamed tumor microenvironment has been reported to correlate with response in tumor patients. However, due to the lack of appropriate experimental methods, the reason why the immunotherapeutic resistance still existed on the inflamed tumor microenvironment remains unclear. Materials and methods Here, based on integrated single-cell RNA sequencing technology, we classified tumor microenvironment into inflamed immunotherapeutic responsive and inflamed non-responsive. Then, phenotype-specific genes were identified to show mechanistic differences between distant TME phenotypes. Finally, we screened for some potential favorable TME phenotypes transformation drugs to aid current immunotherapy. Results Multiple signaling pathways were phenotypes-specific dysregulated. For example, Interleukin signaling pathways including IL-4 and IL-13 were activated in inflamed TME across multiple tumor types. PPAR signaling pathways and multiple epigenetic pathways were respectively inhibited and activated in inflamed immunotherapeutic non-responsive TME, suggesting a potential mechanism of immunotherapeutic resistance and target for therapy. We also identified some genetic markers of inflamed non-responsive or responsive TME, some of which have shown its potentials to enhance the efficacy of current immunotherapy. Conclusion These results may contribute to the mechanistic understanding of immunotherapeutic resistance and guide rational therapeutic combinations of distant targeted chemotherapy agents with immunotherapy.

Although immunotherapy has revolutionized tumor treatment, it still has some limitations [1]. 74 For example, the success of adoptive cell therapy (ACT) on hematological malignancies 75 can't be reproduced on solid tumors [2]. The responsive rate of Immune Checkpoint 76 inhibitors (CPI) variates by tumor type, from 45% for melanoma [3,4] to only 12.2% for 77 head neck squamous cancer (HNSC) [5,6]. 78 To better understand the reasons for these limitations, a number of studies try to investigate 79 the phenotype of tumor microenvironment (TME) and suggesting TME phenotype (broadly 80 Figure.1 Overall design of this study. 156 As mentioned above, the responsive rate of immunotherapy variates by tumor type. To 157 understand the factors that contribute to the differences in susceptibility to immunotherapy, 158 we integrated two single-cell RNA sequencing datasets respectively from Head and Neck 159 squamous carcinoma (HNSC) and melanoma, which were characterized by different 160 immunotherapeutic sensitivity (approximately 45% response rate for melanoma [3,4], 161 significantly higher than the 12.2% of HNSC[36]). 162 The integration result was shown in Figure.

Pan-cancer prognostic significance of TME subtypes 182
To classify TME phenotypes across distant tumor types, immune cell gene signatures (GS) 183 were identified in the above integrated single-cell data. Then, we classified TCGA pan-184 cancer samples into three TME subtypes based on the unsupervised clustering pattern of 185 GS, each assigned as high-immune score (Inflamed), intermediate immune score, low-186 immune score(non-inflamed). ( Figure.3 A) As shown in Figure. To further investigate mechanistic differences between inflamed and non-inflamed TME, 212 we compared gene expression profiles between inflamed and non-inflamed TME. As 213 shown in figure.4A, up-regulated genes in non-inflamed TME related to processes such as 214 activated GPCR signaling pathway. 215 In inflamed TME (figure.4B), upregulated genes mainly related to multiple cytokine 216 pathways, such as IFN and the interleukin-4, interleukin-13, and interleukin-10. In addition, 217 the CD28 costimulatory molecule family associated signaling pathways, which includes 218 PD-1, CTLA-4, was also significantly upregulated in the inflamed TME.

TME phenotypes correlated with the immunotherapeutic sensitivity 225 226
To better understand the association between TME phenotypes and the response to To further offer mechanistic insights into CPI resistance in inflamed TME, we identified 232 several differentially expressed genes (DEGs) in inflamed non-responders versus inflamed 233 responders ( Figure.5 D). These genetic signatures of TME phenotype may sever as 234 potential targets for improving current immunotherapy. 235 For instance, CLDN18 was the signature of inflamed responsive TME. Therapy that directly 236 targets on CLDN18 has shown its potential to improve the efficacy of ACT in treating solid 237 tumors [41]. On the other side, inhibiting the signature of inflamed non-responsive TME 238 may be another promising way. Here, SIGLEC5 was significantly overexpressed in 239 inflamed non-responders, and its family member SIGLEC15 has been proven as an 240 efficient target to enhance anti-tumor immunity [15].

Mechanistic differences between inflamed responsive TME and inflamed non-251
responsive TME.

253
Then, gene functional annotation analysis was used to understand the role of TME 254 phenotype-specific genes. As shown in Figure.6 A, genes upregulated in inflamed and 255 responsive tumors enriched on complement cascade and bile metabolism. GSEA analysis 256 also confirmed multiple metabolism associated pathways were activated in this type of 257 TME ( Figure 6E). In terms of inflamed non-responsive tumors, signaling pathways, such 258 as IL-13, IL-4, IL-10 and IL-1 cytokines related signaling pathways and oxygen exchange 259 pathway were upregulated ( Figure 5E, 6 B). These results suggested tumor 260 hypermetabolism or activated multiple cytokines expression may confer resistance to 261

immunotherapy. 262
Finally, for a more systematic understanding of the resistant mechanism, we applied GSEA 263 analysis to investigate the alternation of the molecular pathways across four dimensions. 264 (Epigenetic modification, immune or others associated signaling pathway, metabolism). 265 As shown in Figure. is not completely clear how best to incorporate chemotherapy with immunotherapy. Here, 284 we calculated the genomic connectivity score of 1288 kinds of drugs to identify potential 285 phenotypes transformation drugs which could induce systemic favorable transcriptomic 286 alternation, including from non-inflamed TME to inflamed TME, or from inflamed non-287 responsive TME to inflamed responsive TME. 288 Mercaptopurine (6-MP) was identified as the most promising drug, which may promote the 289 transformation of inflamed responsive TME phenotype (Table.1

302
Molecular stratification of TME phenotypes is paving the way for a better understanding of 303 immunotherapeutic heterogeneity. Here, based on immune cell signatures developed from 304 integrated single-cell RNA sequencing data, we systematically analyzed the molecular 305 characteristics of inflamed TME across multiple cancer types and provided mechanistic 306 insights into immunotherapeutic resistance in inflamed TME. 307 Part of the identified mechanistic differences has been supported by recent reports. 308 Examples highlighted by these data including up-regulation of epigenetic-related signaling 309 pathways and inhibited PPAR-signaling pathways are observed in inflamed non-310 responsive tumors, suggesting potential targets for improving the sensitivity to 311 immunotherapy. Importantly, these results are in line with prior publications, which have 312 provided some evidence that inhibition of epigenetic modification [50]  These abovementioned studies also reveal the molecular characteristics of inflamed TME 316 shared by different tumor types. These results demonstrate that inflamed TME is related 317 to enhanced cytokines expression (interferon and IL-4, 13 and 10). Interestingly, these 318 cytokines except for interferon are also up-regulated in inflame non-responders, 319 suggesting a dual role of these interleukins and potential targets for therapy. These results 320 are in line with prior published reports[51, 52]. For example, IL-10 is widely recognized 321 as an immunosuppressive cytokine, but there is increasing evidence that it has a dual role 322 in anti-tumor immunity. Blocking or activation of IL-10 has been proven as an efficient way 323 to enhance anti-tumor immunity in different aspects [53,54]. According to our results, we 324 believe that TME phenotypes should be considered as a key factor in further study design 325 to illuminate the remaining mysteries of IL-10 function. 326 In addition, our results have far-reaching clinical implications including identification of 327 multiple potential molecular targets for developing novel immunotherapy and its combined 328 treatment strategies. For instance, the success of ACT therapy can't be reproduced on 329 solid tumors due to the obstacle of its microenvironment. Therefore, rather than directly 330 targeting on whole solid tumors, selectively targeting the inflamed and immunotherapy 331 responsive TME may be another easier therapeutic way. As expected, this hypothesis is 332 supported by a recent report. CLDN18, a signature of inflamed and responsive TME, has 333 been proven as an efficient target for improving efficacy of current ACT therapy on solid 334 tumors [41]. 335 Except for targeting on inflamed and responsive TME, examples highlighted by our data 336 also include inhibiting the signature of inflamed non-responsive TME to reverse therapeutic 337 resistance. For example, SIGLEC15, a signature of inflamed and non-responsive TME, 338 has shown its power in blocking immune escape. Interestingly, the enhancement effect of 339 anti-tumor immunity is independent of the PD-1/PD-L1 axis, suggesting that it may be an 340 ideal target to aid current anti-PD-1 therapy [15]. 341 Finally, based on a genomic-drugs perturbation database, we identify some drugs which 342 are promising for promoting the transformation from unfavorable TME phenotype to 343 favorable one. 344 In conclusion, our result provided an important view for understanding how inflamed TME 345 and inflamed resistant TME forms. This evidence has important clinical implications and 346 may help guide rational therapeutic combinations of distant chemotherapy agents with 347 immunotherapy depending on the desired treatment effect. 348 349 350

Conflict of Interest Statement 351
The authors declare no potential conflicts of interest. 352 353