Background: Preclinical studies suggest synergy between PARP inhibitors and immune checkpoint inhibitors.  In the I-SPY 2 TRIAL, the anti-PDL1 therapeutic antibody durvalumab combined with the PARP inhibitor olaparib showed increased efficacy relative to control in both the HR+/HER2-and TN subtypes. Response to immunotherapy has been associated with intra-tumoral immune infiltrate/activation, whereas PARP inhibitors seem most effective for DNA repair deficient (DRD) cancers. Pre-specified biomarker analysis was performed to test 7 immune genes/signatures previously associated with response to pembrolizumab [Pembro] and/or durvalumab and a DRD signature previously associated with response to veliparib/carboplatin, as specific predictors of response to durvalumab/olaparib [Durva]. We also assessed MammaPrint High1/(ultra)High2 risk class (MP1/2), a prognostic signature used in the trial’s adaptive randomization engine, and performed exploratory analysis on additional signatures.

Methods: 105 patients (Durva: 71, controls: 34) had Agilent 44K gene expression from FFPE pre-treatment biopsies and pCR data; and 370 (Durva: 71, controls: 299) had MP1/2 and pCR data. We evaluated 13 genes/signatures (10 immune, 1 DRD, 1 ER, 1 proliferation) and MP1/2 as biomarkers of Durva response, using logistic modelling to assess performance. A biomarker is considered a specific predictor of Durva response if it associates with response in the Durva arm, andif the biomarker x treatment interaction is significant (likelihood ratio test, p<0.05). pCR rates within MP1/2 classes are estimated using Bayesian logistic modelling. Analysis is also performed adjusting for HR status as a covariate, and numbers permitting, within receptor subsets. Our statistics are descriptive rather than inferential and do not adjust for multiplicities.

Results: 8/10 immune biomarkers, including the genes PD1 and PDL1, and B-cell, dendritic cell and mast cell (but not T-cell or CD68) signatures associate with response to Durva in the population as a whole and in a model adjusting for HR status.  As seen in previous immunotherapy trials, higher levels generally associate with pCR, with the exception of the mast cell signature, where high levels associate with non-response as was also shown for Pembro (I-SPY 2). In addition, high levels of the DRD (PARPi7) and proliferation signatures, as do low levels of ER signalling (ESR1/PGR average).  Many of these biomarkers also associate with response in the control arm, and for no immune biomarker is the treatment interaction significant, suggesting a lack of predictive specificity.  In subset analysis, 13/14 biomarkers (all but CD68) predict Durva response in the HR+HER2-subset, with the strongest association to pCR being a low level of ESR1/PGR (p=2E-08).  In our Bayesian analysis, the difference in estimated pCR rates between arms are primarily observed in the MP2 subtype, particularly in the HR+HER2-MP2 patients (estimated pCR rate of 64% in Durv vs 22% in Ctr). In the TN subset, only 3/14 biomarkers associate with response: the STAT1 and TAM/TcCassII-ratio signatures that also associate with durvalumab response in a prior study (NCT02489448) and, interestingly, the proliferation signature.  Notably, the dendritic, T-cell and tumor inflammatory signatures (TIS) predicting TN response to Pembro (I-SPY2, GeparSixto) do not associate with Durva response in TNBC, suggesting differences in the biology underlying response to PD1 and PDL1 inhibitors.  

Conclusion: Multiple immune, DRD, proliferation, and ER signatures associate with response to durvalumab/olaparib therapy, but many lack predictive specificity.  MP2 class and/or low ESR1/PGR are the strongest predictors of pCR in the HR+HER2-subset; whereas for TNs cytokine-and monocyte-dominated immune signatures like STAT1 [PMID:19272155] and TAM/TcClassII ratio [PMID:24205370] are most predictive. These results require validation.

‍