Adaptive Responses to PARP Inhibition Predict Response to Olaparib and Durvalumab: Multi-omic Analysis of Serial Biopsies in the AMTEC Trial

Introduction

Triple negative breast cancer (TNBC), comprising 15-20% of all breast cancers, predominantly affects younger women and certain ethnic groups. Compared to other breast cancer subtypes, TNBC is characterized by aggressive features, a higher propensity for metastasis, and an overall worse prognosis¹. Despite advancements in understanding TNBC biology, outcomes for metastatic TNBC (mTNBC) remain poor, with an overall survival of 1-2 years with standard therapies^2–4.

The DNA damage response (DDR) pathway is frequently dysregulated in TNBC, with nearly half of all cases exhibiting aberrations. BRCA1/2 mutations are observed in about 15% of TNBC, while another 45% of TNBC with genomic instability have wild type BRCA1/2, suggesting that other aberrations can lead to defects in this pathway^1,5–7. DDR dysregulation renders TNBC sensitive to DNA damaging therapies such as chemotherapy, radiation therapy, and more recently, PARP inhibitors (PARPi). PARPi are of particular interest in mTNBC because of the high frequency of defects in homologous recombination repair, the high-fidelity repair pathway for double-stranded DNA breaks⁸.

In mTNBC, the PARP inhibitors olaparib and talazoparib, FDA-approved for patients with germline BRCA1/2 mutations, improve clinical outcomes primarily by exploiting synthetic lethality with HRD. However, even in this population, responses to PARPi monotherapy are often short-lived, with most patients eventually experiencing disease progression^9,10. Moreover, PARPi monotherapy has limited activity in BRCA1/2 wild-type (BRCA1/2^wt) tumors, which constitute the majority of mTNBC¹¹, with the VIOLETTE trial showing minimal benefit in these patients¹². Preclinical and early clinical data suggest that PARPi combination therapies are safe and more effective in BRCA1/2^wt TNBC, highlighting potential opportunities for these treatments for this population^13–15.

Immune checkpoint blockade (ICB) has improved outcomes for various tumor types, especially those with high tumor mutational burden (TMB) and neoantigenic propensity, which can bolster anti-tumor immunity^16,17. In mTNBC, the combination of the programmed death-1 (PD-1) inhibitor pembrolizumab and chemotherapy is FDA-approved for patients with tumors expressing programmed death ligand 1 (PD-L1). However, this indication excludes the majority of mTNBCs, which are frequently PD-L1 negative, and clinical benefit is frequently brief even in PD-L1 positive patients¹⁸. Additionally, TMB in TNBC is usually below the generally accepted 10 mutations/Mb that predicts ICB efficacy in other cancer types¹⁹.

Extensive preclinical data demonstrate that PARP inhibitors induce adaptive responses in malignant cells and the tumor ecosystem, including immune cells, in models of HRR-deficient and competent tumors^20–24. Similar adaptive changes have been observed in human tumors, such as breast cancer, under therapeutic stress induced by PARPi^14,15,25,26. While specific adaptive responses vary across patients, the number of stress responses engaged following PARPi appears to be limited, facilitating the identification of patient-specific adaptive responses. Targeting these adaptive responses – for example, by increasing PARPi-induced stress, blocking protective responses, or capitalizing on the rewiring of malignant cells and the immune microenvironment²⁷ – has the potential to improve responses to PARP therapies, including in BRCA1/2^wt TNBC. The identification of the specific adaptive response to target may require collecting tumor biopsies taken before and during PARPi monotherapy, then cataloguing drug-induced changes to exploit in a combined treatment regimen.

Combining PARPi with ICB is a promising therapeutic approach to leverage PARPi-induced adaptive changes in immune signaling. PARP inhibition enhances antigen presentation, recruits cytotoxic T lymphocytes, and activates the stimulator of interferon genes (STING) pathway^21,22,28, all of which could increase the efficacy of ICB. Notably, PARPi can activate the STING pathway in BRCA1/2^wt syngeneic cancer models that are DNA repair competent²³, potentially leading to an infiltration of adaptive immune cells that would recognize neoantigens generated through low fidelity DNA repair. Indeed, clinical trials combining ICB and PARPi showed activity in both BRCA1/2^mut and BRCA1/2^wt mTNBC, although response rates are lower in the wild-type group^14,29–31.

Building on these findings, we initiated the Adaptive Molecular Therapy of Evolving Cancers (AMTEC) trial to evaluate the efficacy of combining the PARP inhibitor olaparib with the PD-L1 inhibitor durvalumab in BRCA1/2^wt mTNBC³². This trial includes an olaparib monotherapy phase followed by a combination phase with olaparib and durvalumab, with biopsies collected at baseline and during olaparib monotherapy. Both biopsies are analyzed with a suite of multi-omic assays in an in-depth examination of changes in malignant cells and the tumor ecosystem in response to PARPi-mediated therapeutic stress. Using this approach, the AMTEC trial aims to identify adaptive changes induced by PARPi monotherapy that could serve as biomarkers for selecting patients likely to benefit from PARPi/ICB combination therapy. An additional aim is to discover resistance mechanisms to combined PARPi/ICB, which could be targeted with alternative PARPi combinations. Here we report the clinical and translational results of stage 1 of the AMTEC trial.

Results

AMTEC is a single-arm, open-label phase II trial assessing the efficacy of olaparib combined with durvalumab in participants with BRCA1/2 wild type mTNBC (Fig. 1a). Eligible participants undergo a pre-treatment biopsy upon enrollment, followed by a 4-week olaparib treatment. At the end of this olaparib lead-in, all participants undergo a second biopsy on olaparib, and then durvalumab is added to the treatment regimen.

• Download figure
• Open in new tab

Figure 1. Description of the AMTEC trial.

a) Schematic of AMTEC trial design, key eligibility criteria, timing of treatment and biopsy collections, treatment discontinuation, and post-treatment follow-up.

b) Swimmer plot of participant time on treatment (x-axis) in months for 15 trial participants A through O (see Extended Data Fig. 1a). Best response based on Response Evaluation Criteria in Solid Tumors (RECIST), version 1.1³³, is labeled adjacent to treatment duration bars. Treatment durations for participants with best response of SD or PR is in orange and PD is in purple. Participant B * = deceased from trial-unrelated accident; Participant M ** = voluntarily withdrew from study by personal choice. SD = Stable Disease; PR = Partial Response; PD = Progressive Disease.

c) Best percentage change from baseline in the sum of all target lesions for 14 trial participants. Lesions from Participant N were not evaluable. The dashed lines at 20% and −30% indicate progressive disease and partial response, respectively, based on RECIST. * = new lesions.

d) Kaplan-Meier curves of Overall Survival (left) and Progression Free Survival (right) in months for 14 trial participants, excluding Participant M (see Extended Data Fig. 1a). Median survival is indicated as dashed line. Upper and lower 95% confidence intervals are based on plain survival probability.

e) Kaplan-Meier curve of Overall Survival (left) and Progression Free Survival (right) for 14 trial participants, stratified by RECIST best response PD (purple) and PR or SD (PR/SD, orange). Median survival per strata is indicated as dashed line. P-values based on log-rank test. Upper and lower 95% confidence intervals are based on plain survival probability.

The primary endpoint of the study is the overall response rate (ORR) for olaparib plus durvalumab. Secondary endpoints include clinical benefit rate (CBR), duration of response, progression-free survival (PFS), and overall survival (OS), as well as safety and tolerability of the treatment.

Tumor-Intrinsic Analytes

The following tumor-intrinsic measurements did not show statistically significant correlations with participant outcomes:1) mutations in TP53, the HRD pathway, or the MAPK pathway; 2) TMB (no sample exceeded 10 mutations/Mb); 3) percentage of malignant cells expressing PD-L1 or Ki-67 by IHC; 4) AR positivity by IHC or DSP; 5) biopsy site; or 6) tumor content, although this tended to be higher in participants with PD.

PD-L1 tumor cell positivity was assessed by IHC using two different cutoffs due to discrepancies in the literature³⁴. PD-L1 tumor proportion scores ≥10% were observed in at least one biopsy from four of the fifteen participants, including three of nine (33%) demonstrating clinical benefit (PR/SD) and one of six (17%) participants who did not (PD; Extended Data Fig. 1b). PD-L1 scores ≥1% were observed in at least one biopsy from ten of the fifteen participants, including seven of nine (78%) demonstrating benefit (PR/SD) and three of six (50%) who did not (PD). Although not statistically significant, participants with PD-L1 ≥1% were more likely to achieve PR/SD, aligning with other studies indicating an association between PD-L1 positivity and response to ICB in TNBC^35,36.

PD-L1 protein levels were also measured by NanoString DSP. Unlike IHC, high DSP protein levels in the pre-treatment biopsy, but not the on-olaparib biopsy, showed a statistically significant association with OS (pre-treatment p=0.03; see Extended Data Fig. 4o). The differences in statistical significance between IHC and DSP may be attributed to the greater quantitative sensitivity of DSP, variations in cutoff values for positive sample calls, or slight differences in sample sets. Thus, while PD-L1 may have influenced the response to olaparib and durvalumab, PD-L1 protein measurements by IHC are insufficient to accurately predict response.

Interestingly, olaparib-induced increases in the number of PD-1⁺ cells, but not in PD-L1⁺ cells, were significantly associated with achieving clinical benefit (PR/SD) from combined olaparib/durvalumab, as measured by mIHC (see Fig. 6a; Extended Data Fig. 6b). High PD-1⁺ cells in the on-olaparib biopsy also showed a trend toward significance for OS (p=0.1, data not shown).

The most frequently mutated gene was TP53, observed in 14 out of 15 participants (93%). No tumor had a deleterious aberration in BRCA1, BRCA2, BARD1, PALB2, RAD51C, or RAD51D, which are known to confer sensitivity to PARPi^37,38 (Extended Data Fig. 1b). However, there were alterations in two other genes involved in homologous recombination (HR): a potentially pathogenic germline ATM mutation in Participant A and a potentially clinically significant somatic CDK12 mutation in Participant G. Regarding mutations possibly related to durvalumab response, Participants E and F had deleterious mutations in beta-2-microglobulin (Β2M), and Participant K had a low frequency JAK1 mutation detected only in the on-olaparib biopsy.

PI3K/AKT pathway alterations were present in biopsies from two of nine PR/SD participants and in five of six PD participants. Note that Participant N had only one of the two biopsies with clinically reported PTEN loss, possibly due to tumor content variability constraining copy number loss detection. Receptor tyrosine kinase or MAPK pathway aberrations were present in only two participants: one with SD and one with PD as best response.

RPPA proteomics data were generated from a subset of samples: 9 pre-treatment biopsies, 8 on-treatment biopsies, and 7 matched pairs. Protein PARylation (PAR), which is catalyzed by PARP upon binding single-strand DNA breaks, was elevated in all evaluable pre-treatment biopsies from PD participants, but in only one PR/SD sample (Participant F) (Fig. 2a; Extended Data Fig. 1c). Although PAR was initially high in the PD samples, it significantly decreased to levels similar to those in the PR/SD samples upon olaparib treatment (Fig. 2a). This indicates that the poor response to the olaparib/durvalumab combination in these participants was not due to insufficient olaparib activity. No other proteins or protein pathways assessed by RPPA were associated with treatment outcomes, likely due to the limited number of samples with sufficient material for analysis.

• Download figure
• Open in new tab

Figure 2. Tumor-intrinsic signatures of response to olaparib and durvalumab.

a) Boxplot of PAR protein levels (RPPA; y-axis) in pre-treatment (Pre-) and on-olaparib (On-) biopsies (x-axis), separated by best response PD (purple, left) and PR/SD (orange, right). P-values are based on paired Student’s t-test comparison of pre-treatment vs. on-olaparib. * = p-value less than 0.05; ns = not significant.

b) Survival curve using Kaplan-Meier of Overall Survival in months, stratified by High (red) and Low (blue), for the Mutation Signature 3 (MutSig3) based on WES from the pre-treatment biopsy. Hazard Ratios (HR) and p-values for the Cox proportional hazard (CoxPH) model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability.|

c) Survival curve using Kaplan-Meier of Overall Survival in months, stratified by presence (Positive, blue) or absence (Negative, red) of mutated or copy-number altered genes in the PI3K-AKT pathway or PI3K-AKT pathway activation by DSP or both. The pre-treatment biopsyis shown on the left and the on-olaparib biopsy is shown on the right. See Methods for gene and phosphoprotein inclusions and sequencing panel details. Hazard Ratios (HR) and p-values for the CoxPH model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability. Insets: Stacked bar plot of the percent of samples (y-axis) with (Positive) or without (Negative) PI3K-AKT pathway mutations (x-axis). The percent of samples from participants with best response PD are indicated in purple and PR/SD in orange. P-values are from the Fisher’s exact test.

d) Survival curve using Kaplan-Meier of Overall Survival in months, stratified by presence (Positive, blue) or absence (Negative, red) of RTK-MAPK pathway activation as measured by NanoString DSP. See Methods for phosphoprotein inclusions and pathway activation criteria. Hazard Ratios (HR) and p-values for the CoxPH model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability.

e) Survival curve using Kaplan-Meier of Overall Survival, stratified by presence (Positive, blue) or absence (Negative, red) of either RTK-MAPK or PI3K-AKT pathway activation as measured by NanoString DSP. Hazard Ratios (HR) and p-values for the CoxPH model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability. Insets: Stacked bar plot of the percent of samples (y-axis) with (Positive) or without (Negative) RTK-MAPK or PI3K-AKT pathway activation (x-axis). The percent of samples from participants with best response PD are shown in purple and PR/SD in orange. P-values are from the Fisher’s exact test.

Defects in HR are well-established predictive biomarkers for PARPi response, even in the absence of BRCA1/2 mutations^39–42. To assess HR deficiency in our BRCA1/2^wt samples, we developed an immunofluorescence-based assay that distinguishes HR-competent (HRC) and HR-deficient (HRD) cells. Cells in G2 phase were evaluated for the presence of functional HR repair complexes and double-strand DNA breaks within the same nuclei, then samples were categorized as predominantly HRC, HRD, or Intermediate (Extended Data Fig. 1b-e). Although HRC or HRD status alone was insufficient to predict clinical outcomes, the two participants with tumors classified as HRD (Participants M and E) achieved a best response of either PR or SD. In contrast, no analyzable samples from participants with PD were classified as HRD.

Genomic scarring is a common indicator of HRD due to its strong association with BRCA1/2 mutations and other DNA damage mechanisms⁴³. Given its association with HRD, we assessed the predictive value of the MutSig3 signature in the WES from the pre-therapy biopsy. Interestingly, MutSig3 was significantly associated with OS (p-val = 0.03; Fig. 2b) but not with best response or PFS (data not shown).

PI3K-AKT and RTK-MAPK pathway activity is associated with therapeutic resistance, including to PARPi and ICB^15,44,45. To evaluate whether activation of these pathways influenced the response to olaparib/durvalumab in AMTEC, we catalogued PI3K-AKT and RTK-MAPK gene alterations and pathway activation via NanoString DSP phosphoprotein profiling.

Samples with elevated phosphoproteins in the PI3K-AKT pathway (indicative of PI3K-AKT pathway activation) tended to also have PI3K-AKT gene aberrations: six of the seven pathway-activated samples had a concurrent pathway gene alteration (Extended Data Fig. 1b). Consistent with the notion that PI3K-AKT gene alterations may be leading to pathway activation, in either biopsy, gene alterations or pathway activation were associated with poor outcomes: gene alterations with progressive disease and reduced OS and PFS, and pathway activation with reduced PFS (Extended Data Fig. 2a-c). Additionally, a combined analysis of samples with a mutation and/or elevated pathway activity showed an association with progressive disease and reduced OS and PFS (Fig. 2c; Extended Data Fig. 2d).

In contrast, RTK-MAPK DSP pathway activation was independent of gene alteration: none of the four pathway-activated samples had concurrent pathway mutations, and none of the three samples with genomic alterations also had pathway activation (Extended Data Fig. 1b). Consistent with our previous studies^15,26, MAPK pathway activation appeared to be induced by PARPi, as all four pathway-activated samples were from on-olaparib biopsies. MAPK pathway activation in the on-olaparib biopsy but not the pre-treatment biopsy was associated with worse OS and PFS (on-olaparib p-value = 0.03 for OS and 0.004 for PFS; Fig. 2d; Extended Data Fig. 2e).

Elevated signaling activity in the on-olaparib biopsy in either the PI3K-AKT or RTK-MAPK pathway strongly indicated poor clinical outcomes, including OS, PFS, and RECIST best response (on-olaparib p-values of 0.005, 0.003, and 0.015, respectively; Fig. 2e; Extended Data Fig. 2f). One or both pathways were activated in four of six PD participants but not in any PR/SD participants (Extended Data Fig. 1b). In contrast, gene alterations in either pathway in the on-olaparib biopsy did not significantly associate with response (on-olaparib p-value = 0.1 for OS, 0.12 for PFS, and 0.103 for best response; Extended Data Fig. 2g and data not shown). This suggests that protein-based analyses from on-treatment biopsies may be more informative for anticipating resistance to olaparib/durvalumab driven by PI3K-AKT or RTK-MAPK signaling in mTNBC.

Tumor Microenvironment Analyses

We also evaluated the tumor microenvironment in pre-treatment and olaparib-treated samples to identify features predicting benefit from the olaparib/durvalumab combination. Using GSVA analysis of RNAseq data, high levels of the Cancer Hallmark “Angiogenesis” signature in the on-olaparib biopsy were associated with worse response to therapy (PD vs. PR/SD p-value = 0.021) and survival (OS p-value = 0.003 and PFS p-value = 0.0009) but had no predictive value in the pre-treatment biopsy (Fig. 3a; Extended Data Fig. 3a). Interestingly, olaparib treatment changed the expression levels of this signature for six of the fifteen participants, with a decrease from high to low in three good responders and an increase from low to high in two poor responders (Extended Data Fig. 1b).

• Download figure
• Open in new tab

Figure 3. Microenvironment signatures of response to olaparib and durvalumab: RNA signatures.

Survival curve using Kaplan-Meier of Overall Survival in months, stratified by High (red) or Low (blue) gene set enrichment of the indicated RNA signatures, in the pre-treatment (left) and on-olaparib biopsies (right). Enrichment scores (as a continuous variable) were discretized as High for scores ≥ 0 and Low for < 0. Hazard Ratios (HR) and p-values for the CoxPH model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability. Insets show boxplots of the gene signature enrichment score (y-axis) for samples from participants with best response (x-axis) PD are indicated in purple and PR/SD in orange for each RNA signature for the pre-treatment (left inset) and on-olaparib (right inset) biopsies. P-values from the Student’s t-test are labeled: * = p ≤ 0.05; ** = p ≤ 0.01; ns = not significant (p > 0.05).

a) High vs. Low Hallmark Angiogenesis RNA signature.

b) High vs. Low Hallmark IL6/JAK/STAT3 Signaling RNA signature.

c) High vs. Low Activated CD8 T Cells RNA signature.

d) High vs. Low T Helper Cells RNA signature.

Olaparib alters the immune landscape in model systems and in patient tumors, and immune contexture is a key determinant of durvalumab activity^46,47. While clinical IHC measurements of infiltrating CD4 and CD8 T cells showed no correlation with participant outcomes (data not shown), multiple immune features in the AMTEC on-olaparib samples from our multi-omic analyses were associated with PFS, OS, and RECIST best response, including gene expression signatures, protein markers, and immune cell density (summarized in Extended Data Fig. 1b and Extended Data Table 1d).

In the bulk RNAseq data, the immune-related GSVA gene sets “IL6/JAK/STAT3 Signaling”, “Activated CD8 T Cells”, and “T Helper Cells” were associated with favorable OS and RECIST best response (PR/SD) in the on-olaparib biopsy only (Fig. 3b-d). “Interferon Gamma Response” and “NK/CD56dim Cells” were both associated with best response (PR/SD) in the on-olaparib biopsy, but neither signature significantly associated with survival (Extended Data Fig. 3b,c). In contrast, enriched “Cytotoxic Cells” was associated with favorable OS and RECIST best response in both the pre-treatment and the on-olaparib biopsies (Extended Data Fig. 3d).

To spatially resolve changes in immune contexture in response to olaparib, we used the protein-based mIHC and DSP assays. Elevated densities of CD20⁺ B cells and multiple CD8⁺ T cell states were associated with improved outcomes in mIHC from on-olaparib biopsies (Fig. 4a-b, Extended Data Fig. 4a-c,q-s). Interestingly, both naïve (TBET/EOMES/PD-1^-; Extended Data Fig. 4q) and exhausted (PD-1/EOMES⁺) CD8⁺ T cells, particularly Ki-67⁺ (Extended Data Fig. 4a-c), were associated with improved outcomes. This indicates that olaparib may facilitate T cell recruitment and/or expansion.

• Download figure
• Open in new tab

Figure 4. Microenvironment signatures of response to olaparib and durvalumab: protein signatures.

Survival curves using Kaplan-Meier of Overall Survival (OS) or Progression Free Survival (PFS) for the indicated immune cell lineages/state densities (mIHC) or protein levels (DSP) in the pre-treatment (left) and on-olaparib biopsies (right), stratified by High (red) and Low (blue). Hazard Ratios (HR) and p-values for the CoxPH model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability. Insets show boxplots of sample scores from participants with best response of PD (purple) and PR/SD (orange) for pre-treatment (left inset) and on-olaparib (right inset). Boxplot P-values are from the Student’s t-test.

a) Progression Free Survival, High vs. Low CD20⁺ B Cells as measured by mIHC. Scaled density of mIHC-derived immune cell lineages as continuous variables were discretized as High for samples ≥ 0 and Low for < 0.

b) Overall Survival, High vs. Low Ki-67⁺ Exhausted T Cells (CD45⁺/CD3⁺/EOMES⁺/PD-1⁺) as measured by mIHC and scaled as in a).

c) Overall Survival, High vs. Low CD45 protein in the Tumor (PanCK⁺) compartment as measured by NanoString DSP. Z-scored, scaled percentiles of protein levels as quantitative variables were discretized into High for samples with scaled expression ≥ 25th percentile or Low for < 25th percentile.

d) Overall Survival, High vs. Low CD3 protein in the Tumor (PanCK⁺) compartment as measured by NanoString DSP and scaled as in c).

Within the myeloid compartment, high macrophage density was associated with better OS in the on-olaparib biopsy but poor PFS in the pre-treatment biopsy (Extended Data Fig. 4f,g). This contradiction may be due to differences in CD163⁺ myeloid populations associated with response: high infiltrating CD163⁺ macrophages in the pre-treatment biopsy associated with shorter PFS, whereas high CD163^- macrophages and CD68⁺ CD163^- monocytes in the on-olaparib biopsy associated with better clinical outcomes (Extended Data Fig. 4h-j). These results indicate that elevated CD163⁺ myeloid subsets at baseline may be an indicator of poor outcomes, and that olaparib increases the number of CD163^- myeloid cells that contribute to improved responses.

Using the DSP platform, we focused on immune markers within the tumor (PanCK⁺) compartment, which represents immune cells proximal to malignant epithelial cells. Consistent with mIHC, elevated CD45 (immune cells) and CD3 (T cells) in the tumor compartment were associated with improved outcomes in the on-olaparib biopsy, including OS and best response (Fig. 4c,d). In addition, CD68 (myelomonocytic cells), CD56 (natural killer cells), and CD11c (dendritic cells). were all associated with improved PFS and/or OS in the on-olaparib biopsy (Extended Data Fig. 4k-n). As noted above, elevated PD-L1 in the tumor compartment was associated with better OS and RECIST best response in the pre-treatment biopsy, even though clinical IHC was not (Extended Data Fig. 4o).

Our data show that while each analysis platform revealed multiple immune cell populations or markers that predicted clinical outcomes on the AMTEC trial, these predictors varied from platform-to-platform (Extended Data Fig. 1b, Extended Data Table 1d). There are a variety of possible technical explanations for this, some of which we present in the “Discussion” below. To identify cross-platform response predictors that minimize the technical nuances of each platform, we combined the findings from the three assay platforms into a unified consensus signature, using the most predictive individual findings from each.

We first combined the response predictors from each of the three individual assay platforms into single-assay consensus immune signatures (see Methods and Extended Data Fig. 1b for response predictor inclusions). A “Positive” RNAseq consensus signature was defined as high enrichment scores in four or more immune RNA signatures (GSVA gene sets) and was significantly associated with favorable OS and RECIST best response but not PFS (“IMMUNE RNA SUMMARY”; Fig. 5a, Extended Data Fig. 5a). Similarly, a “Positive” mIHC consensus signature was defined as having six or more high mIHC cell types and was significantly associated with favorable PFS and best response as well as a trend toward OS (“IMMUNE mIHC SUMMARY”; Fig. 5b, Extended Data Fig. 5b). A “Positive” DSP consensus immune signature was defined as having four or more high Immune DSP proteins and was significantly associated with favorable PFS, OS and best response (“IMMUNE DSP SUMMARY”; Fig. 5c, Extended Data Fig 5c). Importantly, the consensus immune signatures for each of the three assay platforms was associated with improved clinical outcomes when observed in the on-olaparib biopsy but not the pre-treatment biopsy.

• Download figure
• Open in new tab

Figure 5. Microenvironment signatures of response to olaparib and durvalumab: Immune Consensus signatures.

Survival curves using Kaplan-Meier of Overall Survival (OS) and Progression Free Survival (PFS) in months for the indicated Immune Consensus signatures, stratified by signature Negative (red) and Positive (blue) in the pre-treatment (left) and on-olaparib biopsies (right). Hazard Ratios (HR) and p-values for the CoxPH model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability. Insets show stacked bar plots of the percent of samples that are Negative or Positive for each Immune Consensus signature, with samples from participants with best response PD indicated in purple and PR/SD in orange. P-values are labeled from the Fisher’s exact test. See Methods for Immune consensus design and assignment. Individual sample assignment is summarized in Extended Data Fig. 1b.

a) Overall Survival and Fisher’s exact test for the RNAseq-based Immune Consensus.

b) Progression Free Survival and Fisher’s exact test for the mIHC-based Immune Consensus.

c) Overall Survival and Fisher’s exact test for the DSP-based Immune Consensus.

d) Overall Survival and Fisher’s exact test for the Multi-Omic Immune Consensus, which combines the Immune RNA, mIHC, and DSP protein predictors.

The consensus calls showed strong concordance from platform-to-platform: RNA vs. DSP was 76% concordant, RNA vs. mIHC 74%, and DSP vs. mIHC 83% (Extended Data Fig. 1b: compare “IMMUNE RNA SUMMARY”, “IMMUNE mIHC SUMMARY”, and “IMMUNE DSP SUMMARY”). This finding indicates that combining individual assay predictors within each assay platform can effectively compensate for the significant technical differences across the assays as detailed in Discussion.

Importantly, a multi-omic consensus immune signature combining all three platforms strongly predicted PFS (p=0.003), OS (p=0.01), and PR/SD (p=0.005) in the on-olaparib biopsy but not the pre-treatment biopsy (Fig. 5d, Extended Data Fig. 5d). Taken together, evaluation of immune contexture, particularly within the on-olaparib biopsy, provides a compelling ability to predict clinical response to combined olaparib/durvalumab.

In the analyses above, biopsies from participants with better responses and longer survival tended to have higher levels of immune-related RNA signatures, mIHC cell types, and DSP proteins than non-responders, suggesting that participant outcomes are influenced by adaptive changes in tumor-infiltrating immune cells in response to olaparib. To investigate this further, we examined changes in immune signatures in matched pairs of pre- and on-olaparib biopsies to identify statistically significant changes correlated with RECIST best response. As shown in Fig. 6 and Extended Data Fig. 6, multiple markers increased on olaparib in matched biopsy pairs from participants with favorable responses but not in those with poor responses, such as CD4⁺ T cells (pre-treatment to on-olaparib change, p=0.0017 for PD vs. PR/SD), CD8⁺ T cells (p=0.00055 for PD vs. PR/SD), and PD-1+ cells (p=0.00024 for PD vs. PR/SD). Moreover, several changes induced by olaparib in responding participants were also associated with longer survival, including exhausted CD8⁺ T cells and CD68⁺ CD163^- monocytes from mIHC (compare Fig. 6a and Extended Data Fig. 6a to Extended Data Fig 4b,c,j) and CD3 and CD68 from DSP (compare Fig. 6b to Fig. 4d and Extended Data Fig. 4k).

• Download figure
• Open in new tab

Figure 6. Olaparib-induced response signatures and markers.

Boxplots of signature and marker changes between pre-treatment and on-olaparib biopsies for participants with a best response of PD in purple dots and PR/SD in orange. Each dot represents the difference in assay score between the on-olaparib and pre-treatment values for paired samples from the same participant, where a positive score indicates a feature was higher on the on-olaparib biopsy and a negative score indicates it was lower. Only features with statistically significant difference for paired change between paired PD and PR/SD (p < 0.05) based on Student’s t-test are shown: * = p <= 0.05; ** = p <= 0.01; *** = p <= 0.001.

a) Signature changes from pre-treatment to on-olaparib for mIHC cell lineages or functional states.

b) Protein marker changes from pre-treatment to on-olaparib for DSP immune proteins in the tumor (PanCK+) compartment.

c) Signature changes from pre-treatment to on-olaparib for the Activated CD8 T Cells RNA Signature.

TNBC classification systems that use bulk RNAseq data to define luminal, basal, mesenchymal, and immune states have been refined to encompass basal-like immune-activated (BLIA), basal-like immune-suppressed (BLIS), luminal androgen receptor (LAR), and mesenchymal (MES) subtypes^48–51. We have since implemented and validated the latter classification algorithm within the CLIA/CAP certified OHSU Knight Diagnostic Laboratories, adding an indeterminant (IND) category for samples with insufficient confidence to discriminate between the BLIA and BLIS subtypes.

In the AMTEC study, participants with BLIS or LAR subtypes in either the pre-treatment or the on-olaparib biopsy had significantly poorer outcomes than those with either BLIA or IND (Fig. 7a,b), with a decreased OS (on-olaparib biopsy p-val=0.001) and PFS (p-val=0.001) and a significantly likelihood of RECIST best response of PD (p-val=0.035). Tumors classified as BLIA were associated with a high likelihood of benefit. Indeed, all four BLIA participants achieved a best response of SD/PR and all three BLIS/LAR participants had PD (Fig. 7c). No MES tumors were present in the sample set.

• Download figure
• Open in new tab

Figure 7. TNBC subtype correlates of response to olaparib and durvalumab.

a) and b) Survival curves using Kaplan-Meier of a) Overall Survival (OS) and b) Progression Free Survival (PFS) in months for RNAseq-computed TNBC subtypes, stratified by BLIS or LAR (BLIS/LAR, red) and BLIA or Indeterminate (Non-BLIS/LAR, blue) in the pre-treatment (left) and on-olaparib biopsies (right). Hazard Ratios (HR) and p-values for the CoxPH model and Log-rank tests are labeled. Upper and lower 95% confidence intervals are based on log survival probability. Insets show stacked bar plots of the percent of samples that are classified as BLIS/LAR or Non-BLIS/LAR, with the percent of samples from participants with best response of PD indicated in purple and PR/SD in orange. P-values from the Fisher’s exact test are labeled.

c) Swimmer plot for 15 AMTEC trial participants showing participant time on treatment in months (x-axis) and best response RECIST (adjacent to treatment duration bars). TNBC Subtype is annotated to the left of the participant IDs, with BLIA in pink, BLIS in green, indeterminate in blue, and LAR in yellow. TNBC subtyping was performed on the on-olaparib biopsy for all participants, except for Participant A where the pre-treatment biopsy was used and Participant I where no RNA was available. * = deceased from trial-unrelated accident; ** = voluntarily withdrew from study by personal choice.

Olaparib treatment changed the TNBC subtype for several participants, from either IND to BLIA or BLIS to IND (Extended Data Table 1a). These changes were clinically relevant, as the subtype of the on-olaparib biopsy had a higher predictive value than the pre-treatment biopsy for both survival and best response (Fig. 7a,b).

The predictive value of the TNBC subtype was tested on an independent sample set treated under the same AMTEC protocol using samples not included in the stage 1 dataset described here. In this set, the best response of 7 of 8 participants (88%) was accurately predicted by the BLIA/BLIS algorithm (data not shown). This training and test set approach demonstrates that despite the limited number of samples analyzed, TNBC subtype is a strong signature predicting benefit from olaparib/durvalumab that warrants further investigation, particularly in the on-olaparib biopsy.

A striking observation was that a subset of the highly ranked BLIA genes (centroid >0.5) was interferon-induced, with a median fold change of >2 (based on data from interferome.org; Extended Data Fig. 7a). However, while some genes in the BLIA signature were interferon inducible, classical interferon-induced genes such as IRFs and IFITs did not score among the highly ranked genes, indicating that the BLIA predictor potentially captures a novel interferon-induced gene set.

We analyzed an internal single-cell RNAseq data set from 13 breast cancer specimens to determine which kinds of cells express the genes defining the TNBC subtypes. As shown in Extended Data Fig. 7b, BLIA-associated genes were predominantly expressed in immune cells, including various subtypes of T cells, NK cells, myelomonocytic subsets, and dendritic cells. Consist with this, these genes were strongly associated with antigen presentation, T cell localization, and T cell killing. The BLIA-associated genes showed minimal expression in other cell types in the dataset, including cancer-associated fibroblasts, B cells, malignant cells, or normal epithelial cells. However, like the interferon-induced genes, classic T cell, macrophage, monocyte or dendritic cell genes were not part of the BLIA signature.

Discussion

The AMTEC trial, which included one month of olaparib monotherapy followed by combined olaparib and durvalumab, showed significant benefit for BRCA1/2 wild type participants, with a median PFS of 5.52 months and OS of 12.88 months. Of the 15 evaluable participants, 4 (27%) achieved PR and 8 (53%) demonstrated clinical benefit. The overall PFS did not differ between participants with PR and those with SD, indicating that for this combination, the clinical benefit rate is the most important determinant of efficacy. The combination therapy in this trial was reasonably well-tolerated, with toxicity resolving through dose holds or reductions. Many, but not all, AMTEC participants demonstrated significant clinical benefit from the combination of olaparib and durvalumab. We therefore conducted a deep analysis of our multi-omic data sets to identify predictive markers that differ between responding participants and non-responders, particularly those that emerge as an adaptation mechanism in response to olaparib monotherapy.

Despite initial expectations, several markers were non-predictive of response. These include TMB; percent PD-L1 or Ki-67 positive cells by IHC; mutations in TP53, the HRD pathway, or the MAPK pathway; numbers of CD4⁺ or CD8⁺ T cells by IHC; AR levels by IHC or DSP; biopsy location; and percent tumor content. Interestingly, two participants had mutations in B2M that are expected to decrease class I HLA expression on the cell surface and impair antigen presentation. Loss of B2M and HLA expression has been linked to resistance to immune checkpoint blockade⁵². Despite this, both participants with B2M mutations benefited from treatment, with one showing evidence of constitutive NK cell activation and the other exhibiting an olaparib-induced immune response.

Using a RAD51 foci assay, we found that neither HRC nor HRD status was sufficient to predict clinical outcomes. This may be due to the limited number of participant samples definitively classified as HRC or HRD. Importantly, our method of counting RAD51 foci exclusively in G2 cells with DNA damage revealed tumors with mixed populations of HRC and HRD cells. This suggests that clinical outcomes in PARPi-treated mTNBC patients may be adversely affected by intratumoral heterogeneity in HRR competence.

PD-L1 protein levels were predictive of outcomes in our quantitative DSP assay. Although PD-L1 clinical IHC was not a statistically significant outcome predictor, higher percentages of PD-L1⁺ tumor cells were frequently seen in participants with favorable clinical outcomes and less common in those with unfavorable outcomes, particularly in the on-olaparib biopsy. Notably, PD-1⁺ T cells were more abundant in on-olaparib biopsies from participants with PR/SD, indicating that olaparib either remodeled immune contexture or enhanced antigenic priming in responding participants.

We found that AKT and MAPK pathway mutations and activity identified a subset of participants with limited response to the combination of olaparib and durvalumab. The predictive value was particularly strong in on-olaparib biopsies. This supports the concept that the AKT and MAPK pathways are activated as an adaptive response to PARP inhibition, which subsequently limits the efficacy of olaparib plus durvalumab. These pathways have been implicated in the response to olaparib combination therapies in other preclinical studies and clinical trials^{15,26,37,45,53}. Based on these findings, we initiated a follow-on multi-arm AMTEC trial (NCT03801369), which includes olaparib combined with either a MEK inhibitor (selumetinib) or an AKT inhibitor (capivasertib). Arm allocation in this study is largely based on analysis of MAPK and AKT pathway activation in the on-olaparib biopsy, assayed in the CLIA environment using the DSP platform.

Somewhat unexpectedly, most participants with a high “Angiogenesis” RNA signature in the on-olaparib biopsy had a poor clinical response to the olaparib/durvalumab combination. This observation is consistent with an ovarian cancer study that implicated VEGFR3 expression in resistance to olaparib and durvalumab⁵⁴, as well as the clinical benefit derived from angiogenesis-targeting PARPi combinations such as olaparib plus cediranib or bevacizumab^55–57. Our observation that this signature changed as an adaptive response to PARPi indicates that measuring angiogenesis in on-therapy biopsies may identify a population of patients more likely to benefit from the addition of an angiogenesis inhibitor, justifying the potential toxicity associated with these combinations.

Many predictive markers were associated with immune status in the on-olaparib biopsy, supporting the contention that olaparib-mediated alterations in the immune contexture contribute to the response to the olaparib and durvalumab combination. However, not all analyzed markers showed response-dependent changes with olaparib treatment (data not shown), indicating that the immune signature predictors reflect specific changes in immune contexture induced by olaparib rather than sample set differences or differential tumor heterogeneity between responders and non-responders Interestingly, we and others have demonstrated in preclinical studies that PARPi induces STING activation, leading to subsequent interferon production^22–24,28. In the current study, while a baseline interferon signature was associated with response, three PR/SD participants went from a low to a high “Interferon Gamma Response” RNA signature on the on-olaparib biopsy. In contrast, one PD participant went from low to high and a second went from high to low. Since interferon signals through the JAK-STAT pathway, the observed increase in the “IL6/JAK/STAT3 Signaling” RNA signature in two of the three PR/SD participants with the high interferon signature, along with the decrease in this signature in the PD participant with a low interferon signature, support the concept that activation of interferon signaling in response to olaparib could contribute to benefit from olaparib and durvalumab.

While immune indicators from the RNA, mIHC, and DSP platforms all predicted benefit from olaparib and durvalumab, each platform identified different predictive markers. This discrepancy is not surprising, as the mIHC and DSP platforms used different antibodies and the assays were sometimes run on different cores with varying degrees of tissue heterogeneity. Additionally, the mIHC platform measures antigens at the single-cell level, while DSP and RNA sequencing yield data from populations of cells within tumor-rich regions. RNA and protein levels are often poorly correlated, especially for proteins primarily regulated through post-translational modifications rather than RNA transcription⁵⁸. Furthermore, the limited number of samples assessed and the cut-offs for positive versus negative calls, compounded by a lack of community consensus on markers used for cell type assignments, likely influenced the statistical significance of specific markers. These factors highlight the need for further refinement of predictors within each platform using additional samples.

Despite these issues, our patient-by-patient analysis of immune activation state showed high concordance across the platforms. Interestingly, CD4, CD8, and CD20 levels measured by IHC were not predictive of response (data not shown), indicating that deeper analyses, such as those performed here, are necessary to uncover predictive signals. Further, while most of the immune markers were associated with an improved outcome, total macrophages and CD163⁺ macrophages (M2-like) were associated with a worse outcome. An association of M2-like macrophages with poor outcomes has been identified with multiple different therapies, including ICB⁵⁹.

The BLIA/BLIS algorithm for determining TNBC subtype is particularly interesting as we have moved it to the CLIA environment, allowing its use in patient selection. The first step in utilizing the algorithm is to eliminate LAR patients, and indeed, we have excluded participants who have >80% AR levels by IHC from the ongoing AMTEC trial. Importantly, BLIS or BLIA on the on-olaparib biopsy accurately predicted participant benefit. However, clinical use of this assay is confounded by the number of participants who are Indeterminant. While most participants who were Indeterminate on the on-olaparib biopsy benefited (4/6), some did not (2/6). Thus, it is necessary to develop approaches to refine the predictive value of the BLIA/BLIS algorithm.

It is important to note that the TNBC subtype (combining IND with BLIA and LAR with BLIS) is a strong predictor of benefit on the pre-therapy biopsy with p-values of 0.03 and 0.01 for PFS and OS, respectively. The predictive value of the on-olaparib biopsy was better than that of the pre-therapy biopsy, with the pre-therapy biopsy containing about 85% of the information content of the on-olaparib biopsy. Thus, a pre-therapy analysis of BLIA/BLIS, while not optimal, does provide a high predictive value for selection of patients who are likely to benefit from the olaparib and durvalumab combination, especially when an on-olaparib biopsy is unavailable.

This study has multiple limitations, the foremost being the limited number of participants. This concern will be alleviated somewhat by the ongoing AMTEC trial with a small test set to assess potential markers; however, prospective analysis of our potential predictors in future trials is necessary. Some of the assays were only completed on a subset of participants. Despite obtaining up to six cores as part of each biopsy, there was not always sufficient material and tumor content to complete the assays. Further, the DSP data were incomplete from some samples because the assay was upgraded during the trial and there was insufficient material to re-run the analysis; we only included data from the most recent version. Finally, several of the immune characterization assays were not concordant. As discussed above, there are likely multiple reasons for this that need to be resolved.

In summary, the AMTEC trial demonstrated marked activity with acceptable toxicity. Importantly, we identified a set of potential signatures that strongly predicted participant outcomes, with a subset of these markers already validated in the CLIA environment. The ongoing AMTEC trial will provide an independent test set to validate multiple potential biomarkers and to explore their utility in predicting response to additional combination therapies.

Code Availability

No new tools were developed for the analyses in this study.

Author Information

Ethics Declaration

Methods

EXTENDED DATA

Table 1. Clinical summary of AMTEC trial participants.

a) Summary of clinical metadata. Data are for each biopsy from the 15 participants included in the efficacy analysis (received both study drugs and had best response determined by imaging using RECISTv1.1 criteria³³; SD = stable disease, PR = partial response, PD = progressive disease). Progression free survival and overall survival are calculated from the beginning of the first cycle of olaparib. Percent tumor content was the fraction of tumor within pathologist-determined tumor-rich regions that were macro-dissected from FFPE tissue. TNBC subtypes were computed from RNAseq data using a list of 77 genes and classified as immune activated (BLIA), basal immune suppressed (BLIS), luminal androgen receptor (LAR), or indeterminate (IND).

b) Adverse Events. Safety analysis was performed on data from 17 participants who received at least one dose of either drug. Most treatment-related adverse events (91%) were Grades 1 or 2. Of the 38 Grade 3 or higher events, 14 (37%) were treatment-related.

c) Grades 3, 4, and 5 Adverse Events. Data are from 10 of 17 participants. The two immune-related events were resolved with a dose hold. The two Grade 5 events were unrelated to protocol therapy: one was from pain due to disease progression and one (Investigations – other*) was due to a motor vehicle accident.

d) Summary of statistically significant features correlating with response to olaparib and durvalumab, p < 0.05. The biopsy in which each feature was observed (pre-treatment or on-olaparib) is noted for Significant best response, Progression Free Survival (PFS), and Overall Survival (OS).

Figure 1. Description of the AMTEC trial.

a) CONSORT flow diagram for participant enrollment, exclusions, and analyses in the AMTEC trial.

b) Multi-omic clinical features and signatures from serial biopsies. Columns are sample identifiers (participant identifier appended to biopsy number), and rows are clinical or molecular signature data. White space indicates missing data. See Methods for detailed descriptions of each marker or signature. Signatures with parenthetical notations were found to be significant in one or more of the following assays: B = paired biopsy t-test; R = best response t-test; P = Kaplan-Meier analysis of Progression Free Survival (PFS); O = Kaplan-Meier analysis of Overall Survival (OS); * p-value <0.05; ** p-value < 0.005, ns = not significant.

“Best Response” is from RECIST read of tumor imaging.

“Anatomical Location” is the organ site containing the metastatic lesion from which the biopsy was obtained.

“Tumor Content” is the pathologist-estimated percent tumor content after microdissection of tumor-rich regions from FFPE slides used for GeneTrails Comprehensive Solid Tumor Panel targeted DNA sequencing.

“Ki-67 IHC”, “PD-L1 IHC”, and “AR IHC” were assayed by clinical IHC and defined as % positive tumor cells (Ki-67), positive or negative for tumor proportion score >1% or >10% (PD-L1), or AR positive with a cutoff of >10% AR positive tumor cells.

“PD-1⁺ Cells mIHC”, “PD-1⁺ CD4⁺ T Cells mIHC”, “PD-L1⁺ Cells mIHC” are mIHC-derived cell phenotypes (see Fig. 6a and MULTIPLEX IMMUNOHISTOCHEMISTRY in Methods), with scaled and z-centered values ≥0 defined as “High” and <0 as “Low”.

“AR DSP” is AR protein measured within tumor cell ROIs by NanoString DSP, with z-score percentiles relative to a mixed breast cancer cohort of >25 defined as “High”.

“MutSig3” is the COSMIC Mutation Signature 3, discretized into “High” or “Low” using a threshold of 40 (see also Fig. 2b)

“RPPA PAR” depicts PAR protein levels for each sample measured by RPPA, with “High” defined as a Median Absolute Deviation > 0 relative to a Basal Classified subset of TCGA samples (see also Fig. 2a). “HRC/HRC ratio” is defined as the ratio of HRC to HRD nuclei, with ratios <15% defined as “HRD”, >65% as “HRC”, and in between as “Intermediate (see also Extended Data Fig. 1c-e).

“HR & Immune Genes” and “PI3K-AKT & RTK-MAPK Genes” are clinically reported by GeneTrails. Copy number gain is defined as at least 7 copies and loss is defined as less than or equal to 0.5 copies. Mutations include missense mutations, nonsense mutations, indels, and splice site substitutions. See “NGS: GENETRAILS® COMPREHENSIVE SOLID TUMOR PANEL” in Methods for pathway gene inclusions.

“Pathway Gene Alteration” for PI3K-AKT and RTK-MAPK Genes is defined as the presence of least one DNA alteration among the PI3K-AKT or the RTK-MAPK pathway genes (see also Fig. 2c and Extended Data Fig. 2a, d).

“DSP Phosphoproteins” is the level of each PI3K-AKT or RTK-MAPK pathway phosphoprotein measured within tumor cell ROIs by NanoString DSP. “High” is defined as a phosphoprotein level z-score percentile >75 relative to a mixed breast cancer cohort.

“DSP Pathway Activation” is defined as at least two of three high NanoString DSP phosphoproteins from p-AKT1, p-GSK3B, or p-PRAS40 for the PI3K-AKT DSP pathway and at least two of three high from p-MEK1, p-ERK1/2, or p-p90RSK for the RTK-MAPK DSP pathway (see also Fig. 2d-e and Extended Data Fig. 2b-c, e-k).

“Angiogenesis RNA” and the “Immune RNA Signature” gene set values were derived from GSVA scores calculated from RNA transcriptome sequencing analysis and discretized into “High” or “Low” using a threshold of zero (see also Fig. 3 and Extended Data Fig. 3).

“IMMUNE RNA SUMMARY” notes the number of Immune RNA Signatures as “High” and “Low” for each sample, with “Positive” defined as at least 4 “High” signatures (see also Fig. 5a and Extended Data Fig. 5a).

“mIHC Cell Type” is the immune cell phenotype assigned by hierarchical gating of protein co-expression and quantified to cell densities. “High” is defined as a logged and scaled z-score ≥0 and “Low” <0 (see also Fig. 4a,b and Extended Data Fig. 4a-j, p-u).

“IMMUNE mIHC SUMMARY” notes mIHC predictor score as described under “IMMUNE CONSENSUS” in Methods, with a predictor score of at least 6 (consisting of the sum of all positive predictors minus the negative predictors) defined as “Positive”. See also Fig. 5b and Extended Data Fig. 5b.

“Immune DSP” notes the levels of each immune protein measured within tumor ROIs by NanoString DSP, defining z-scores >25 percentile relative a mixed breast cohort as “High” and <25 as “Low”. See also Fig. 4c,d and Extended Data Fig. 4k-o.

“IMMUNE DSP SUMMARY” notes the overall DSP protein status for each sample, with “Positive” defined as at least 4 High “Immune DSP” proteins and “Low as less than 4 (see also Fig. 5c and Extended Data Fig. 5c).

“MULTI-OMIC IMMUNE CONSENSUS” notes the overall immune status in a combined three-platform signature as described in “IMMUNE CONSENSUS” in Methods. See also Fig. 5d and Extended Data Fig. 5d.

“TNBC Subtype” is described under “TNBC SUBTYPE CALLING” in Methods. See also Fig. 7a,b.

c) Representative images of Geminin, RAD51 foci, and γH2AX foci from homologous recombination deficient (HRD), intermediate, and competent (HRC) samples. RAD51 and γH2AX foci were counted in Geminin-positive nuclei. HRD nuclei were defined as having less than five RAD51 foci and more than three γH2AX foci, and HRC nuclei were defined as having more than five RAD51 foci. Scale bar is shown as indicated.

d) Custom machine learning segmentation results from the samples shown in (c) from participants M, G, and O. Sample HR calls were defined using the ratio of HRC to HRD nuclei. HRC/HRD ratios less than 15% are defined as “Deficient”, greater than 65% as “Competent”, and between 15% and 65% as “Intermediate”. RECIST best response is shown in orange (PR/SD) and purple (PD).

e) γH2AX positivity ratios calculated in the samples shown in (c) and represented as horizontal black bars. “Deficient” samples exhibited greater than 95% γH2AX⁺/Geminin⁺ co-positive nuclei as highlighted in green, while “Competent” samples exhibited less than 55% as highlighted in red. The middle “Intermediate” population (blue shading) tended toward higher γH2AX positivity while having a mixture of “Competent” and “Deficient” nuclei as highlighted in blue.

Figure 2. Tumor-intrinsic signatures of response to olaparib and durvalumab.

Survival curve using Kaplan-Meier of survival probabilities, stratified by presence (Positive, blue) and absence (Negative, red) of the indicated feature in the pre-treatment (left) and on-olaparib biopsies (right). See Methods for gene and phosphoprotein inclusions and pathway activation criteria. Confidence intervals are based on plain survival probability. Insets, where present, show Fisher’s exact test for absence (Negative) vs. presence (Positive) of the indicated feature for pre-treatment (left inset) and on-treatment biopsies (right inset), with percent of samples from participants with best response PD (purple) and PR/SD (orange) noted. Only statistically significant observations (p-value < 0.05) are shown.

a) Overall Survival and Fisher’s exact test for Negative vs. Positive PI3K-AKT pathway gene alteration (mutation or copy number gain/loss).

b) Progression Free Survival for Negative vs. Positive PI3K-AKT pathway gene alteration (mutation or copy number gain/loss).

c) Progression Free Survival for Negative vs. Positive PI3K-AKT pathway activation as measured by DSP.

d) Progression Free Survival for Negative vs. Positive PI3K-AKT gene alteration or pathway activation.

e) Progression Free Survival for Negative vs. Positive RTK-MAPK pathway activation.

f) Progression Free Survival for Negative vs. Positive PI3K-AKT or RTK-MAPK pathway activation. Note that some samples were activated for both pathways.

g) Progression Free Survival and Fisher’s exact test for Negative vs. Positive PI3K-AKT or RTK-MAPK gene alteration. No sample had gene alterations in both pathways.

Figure 3. Microenvironment signatures of response to olaparib and durvalumab: RNA signatures. Survival curve using Kaplan-Meier of survival probabilities, stratified by High (red) and Low (blue) expression of the indicated RNA signature in the pre-treatment (left) and on-olaparib biopsies (right). Confidence intervals are based on plain survival probability. Insets, where present, show best response t-test for PD (purple) vs. PR/SD (orange) of each RNA signature for the pre-treatment (left inset) and on-treatment (right inset) biopsies.

a) Progression Free Survival for High vs. Low Angiogenesis RNA signature.

b) Overall Survival for High vs. Low Interferon Gamma Response RNA Signature. Insets show PD vs. PR/SD t-test.

c) Overall Survival for High vs. Low NK/CD56dim Cells RNA Signature. Insets show PD vs. PR/SD t-test.

d) Overall Survival for High vs. Low Cytotoxic Cells RNA Signature. Insets show PD vs. PR/SD t-test.

Figure 4. Microenvironment signatures of response to olaparib and durvalumab: protein signatures and markers.

Survival curve using Kaplan-Meier and best response t-tests for indicated proteins and protein signatures. Survival probabilities are stratified by High (red) and Low (blue) protein or protein signature expression, and confidence intervals are based on plain survival probability. T-tests compare protein expression in biopsies from participants with best response of PD (purple) vs. PR/SD (orange). Analyses of the pre-treatment biopsy are on the left and the on-olaparib biopsy are on the right. Only statistically significant observations (p-value < 0.05) are shown.

a) Progression Free Survival for High vs. Low Ki-67⁺ Exhausted CD8⁺ T Cells, as measured by mIHC.

b) Overall Survival for High vs. Low Exhausted CD8⁺ T Cells, as measured by mIHC. Insets show PD vs. PR/SD t-test.

c) Progression Free Survival for High vs. Low Exhausted CD8⁺ T Cells, as measured by mIHC.

d) Overall Survival for High vs. Low Ki-67⁺ Immune Cells, as measured by mIHC.

e) Overall Survival for High vs. Low T Regulatory CD4⁺ Cells, as measured by mIHC.

f) Overall Survival for High vs. Low Macrophages, as measured by mIHC.

g) Progression Free Survival for High vs. Low Macrophages, as measured by mIHC.

h) Progression Free Survival for High vs. Low CD163⁺ Macrophages, as measured by mIHC.

i) Progression Free Survival for High vs. Low CD163^- Macrophages, as measured by mIHC.

j) Progression Free Survival for High vs. Low CD68⁺ CD163^- Monocytes, as measured by mIHC. Insets show PD vs. PR/SD t-test.

k) Overall Survival for High vs. Low CD68 protein expression, as measured by NanoString DSP.

l) Progression Free Survival for High vs. Low CD68 protein expression, as measured by NanoString DSP.

m) Overall Survival for High vs. Low CD56 protein expression, as measured by NanoString DSP.

n) Progression Free Survival for High vs. Low CD11c protein expression, as measured by NanoString DSP.

o) Overall Survival for High vs. Low PD-L1 protein expression, as measured by NanoString DSP. Insets show PD vs. PR/SD t-test.

p) PD vs. PR/SD t-test for Immune Cells, as measured by mIHC.

q) PD vs. PR/SD t-test for Ki-67⁺ TBET/EOMES/PD-1^- CD8⁺ T Cells, as measured by mIHC.

r) PD vs. PR/SD t-test for Ki-67⁺ CD8⁺ T Cells, as measured by mIHC.

s) PD vs. PR/SD t-test for PD-1⁺ CD8⁺ T Cells, as measured by mIHC.

t) PD vs. PR/SD t-test for CCR2 Cells, as measured by mIHC.

u) PD vs. PR/SD t-test for Other Immune Cells, as measured by mIHC.

Figure 5. Microenvironment signatures of response to olaparib and durvalumab: Immune Consensus signatures.

Survival curve using Kaplan-Meier of survival probabilities of Immune Consensus signatures, stratified by Negative (red) and Positive (blue) in the pre-treatment (left) and on-olaparib biopsies (right). Confidence intervals are based on plain survival probability.

a) Progression Free Survival for Immune RNA. A “Positive” summary is defined as at least 4 of 6 High RNAseq-derived Immune RNA Signatures shown in Extended Data Fig. 1b.

b) Overall Survival for Immune mIHC. See Methods for details on the quantitation of Immune mIHC Cell Types for “Positive” and “Negative” consensus categorization.

c) Progression Free Survival for Immune DSP. A “Positive” summary is defined as having at least 4 of 6 High NanoString DSP Immune proteins shown in Figure Extended Data Fig. 1b.

d) Progression Free Survival for the Multi-Omic Immune Consensus, which combines the Immune RNA, mIHC, and DSP protein data set summaries above and as shown in Extended Data Fig. 1b. For samples with 3 complete data sets, “Positive” is defined as having either 2 or 3 positive data set summaries and “Negative” as either 0 or 1. For samples with 2 complete data sets, “Positive” is defined as having 2 positive data set summaries, “Indeterminate” as 1, and “Negative” as 0. Samples with only 1 complete data set are “Indeterminate”. Only “Positive” and “Negative” samples were included in these analyses.

Figure 6. Olaparib-induced response signatures and markers.

a) T-test box plots comparing mIHC Cell Type changes between pre- and on-olaparib biopsies for participants with a best response of PD (purple dots) versus PR/SD (orange dots). Each dot represents the difference in assay score between the on-olaparib and pre-treatment values for paired samples from the same participant, where a positive score indicates a feature was higher on the on-olaparib biopsy, and a negative score indicates a feature was higher on the pre-treatment biopsy. Only features with statistically significant differences for paired change between PD and PR/SD (p < 0.05) based on Student’s t-test are shown. All features showed an increase on average in the on-olaparib biopsy in participants with a favorable response.

b) Box plots and paired t-tests comparing mIHC Cell Type scores in the pre-treatment biopsy versus on-olaparib, separated by best response of PD (purple) and PR/SD (orange). Only features with statistically significant differences between the pre-treatment and on-olaparib values in paired samples are shown. PD-1⁺ Cells, PD-1⁺ CD4⁺ Cells, Ki-67⁺ Exhausted CD8⁺ T Cells, and CD68⁺ CD163⁺ Monocytes all increased on olaparib in samples from participants with a favorable response, while Ki-67⁺ TBET/EOMES/PD-1^- CD8⁺ T Cells decreased on olaparib in samples from participants with an unfavorable response.

c) Box plots and paired t-tests comparing DSP Immune Protein levels in pre-treatment versus on-treatment biopsies, separated by best response of PD (purple) and PR/SD (orange). Only features with statistically significant differences between the pre-treatment and on-treatment values in paired samples are shown.

Figure 7. TNBC subtype correlates of response to olaparib and durvalumab.

a) Interferon regulated genes associated with the BLIA TNBC subtype. The Interferome web tool, a database of genes regulated by interferons (http://www.interferome.org), was used to identify a subset of highly ranked, interferon-inducible genes (median Interferome gene change >2, dashed lines, Y-axis) associated with the BLIA subtype (>0.5, X-axis).

b) Dot plot visualizing expression of representative marker genes (x-axis, bottom) within TNBC subtypes (x-axis, top) and across distinct cell type clusters (y-axis). Data are from a single-cell RNA sequencing breast cancer dataset. Relative gene expression levels are represented by the color scale, and the fraction of cells within each cell type group expressing the gene is represented by the dot size. The plot highlights the differential expression patterns of key TNBC markers within each cell cluster, providing insights into the heterogeneity and distribution of TNBC-related gene expression profiles at the single-cell level.