Breast and ovarian cancers in BRCA1/2 germline mutation carriers show relatively low overall immune activity at diagnosis, compared to very immune active non-carriers
In our clinical trial samples, we observed a striking difference in the gene expression profiles between germline mutation carriers and non-carriers. There were 1308 genes differentially expressed between carriers and non-carriers (Posterior Probability of equal expression < 0.05). Of these, 813 showed significantly higher expression in non-carriers (log fold change >1.5) The biological processes most highly enriched in non-carriers identified with Gene ontology tool (Panther Classification System: http://www.pantherdb.org ) were all related to immune functions (Figure 1). Other biological processes which were also enriched in non-carriers include calcium ion transport and signaling, regulation of cell adhesion, motility and chemotaxis, protein secretion, cell signaling (MAPK, ERK1/2 and JNK), cell proliferation, differentiation and cell death (Supplementary Table S2). Many of these processes are related to biology of immune cells. Genes overexpressed in carriers, on the other hand, were not enriched for any particular biological process (data not shown).
We have focused on the 500 biological processes, highly enriched in non-carriers, which were related to immune functions such as T cells differentiation and selection, B cells activation and regulation, production of various Interleukins and signaling via TNF alpha and interferon gamma. This data was highly significant suggesting that the immune environment of sporadic breast and ovarian cancers in our cohort was much more active relative to that of carriers of germline mutations in BRCA1/2 genes. This was independent from the type of germline mutation, BRCA1 or BRCA2 (Supplementary Tables S3 and S4) and was true for both types of cancers when analyzed independently (Supplementary Tables S5 and S6). Many genes overexpressed in breast non-carriers overlapped with those overexpressed in ovarian non-carriers (60 genes). The commonly upregulated genes in breast and ovarian non-carriers were all involved in immune functions (Figure 2 and Supplementary Table S7).
Recently there have been attempts to characterize the immune components of the tumor microenvironment from high-throughput expression data [17–21]. The most complete analysis of immune infiltrates in tumor microenvironment was performed by the group of Trojanoski [21]. They developed a comprehensive and interactive database for immunogenomic studies: The Cancer Immunome Atlas (TCIA) (https://tcia.at/home), which allows exploration of specific immune related gene sets and assessment of cellular composition of infiltrates from 20 solid cancers. We have used their list of 782 genes, which characterize 28 different cell types present in tumor infiltrates [22] to analyze the global immune landscapes of individual carriers and non-carriers in our cohort (Figure 3). The gene list is shown in Supplementary Table S8. All four breast carriers of germline BRCA1/2 mutation showed overall low expression of genes associated with various immune cell types, while three non-carriers showed relatively high expression of most of those genes. The picture was different for ovarian cancers, where some carriers and some non-carriers showed various expression of immune genes consistent with less robust differential expression results. Thus, the expression of 28 meta-gene sets validated our results obtained from differential expression analysis. Expression of these meta-gene set can be a convenient way of representing global immune activity of tumors.
BRCA1/2 germline mutation related breast and ovarian cancers show a range of phenotypes similar to that of sporadic cancers.
There is still controversy if hereditary BRCA1/2 mutation related tumors represent a separate phenotypic identity. Both TNBC and HGSOC represent heterogenous groups of cancers and recently both tumor types were subdivided into several subtypes [12][16][23,24][14][25–27]. Six subtypes of TNBC (IM, BL1, BL2, LAR, M and MSL) were identified from clustering of gene expression data [12]. The Immunomodulatory (IM) subtype is enriched in immune cell signaling. Two other subtypes (basal-like 1 and basal-like 2 (BL1 and BL2)) express high levels of the genes involved in cell proliferation and DNA damage response (DDR), however BL2 is of basal myoepithelial origin and can be distinguished by activated signaling pathways (EGF, NGF, MET, Wnt/b catenin and IGFR1) and glycolysis. Luminal androgen receptor (LAR) subtype is the most distinct of all subtypes, characterized by luminal features and expression of androgen receptor. Mesenchymal (M) and mesenchymal-stem like (MSL) subtypes are characterized by expression of genes involved in epithelial/mesenchymal transition. Patients with BL1 tumors show relatively good prognosis, while patients with BL2 tumors have very poor outcome [28].
Four subtypes of HGSOC (IMR, DIF, MES and PRO) were identified by gene expression profiling. The immunoreactive subtype (IMR) is enriched in immune cell signature, the differentiated subtype (DIF) expresses differentiation markers, the mesenchymal subtype (MES) is characterized by stromal expression signature indicating activated stroma, while the proliferative (PRO) subtype is characterized by low expression of ovarian cancer markers, but overexpression of proliferation and extracellular matrix (ECM) related genes. Importantly, the expression clusters distinguishing the subtypes strongly correlate with histological types of HGSOC [25]. Among all subtypes, the IMR shows the best prognosis and MES subtype has relatively poor outcome [14].
Only one of six subtypes of TNBC (IM) and one of four subtypes of HGSOC (IMR) are characterized by a highly immune active microenvironment. We used a publicly available tool for TNBC classification developed by Lehmann to classify breast tumors from the clinical trial samples [13], http://cbc.mc.vanderbilt.edu/tnbc ). The classification of HGSOC was obtained using the CLOVAR signature (see Methods section for details). Indeed, one of the three sporadic TNBC in this cohort was immunomodulatory, while two others belonged to different categories (MSL and BL2) (Figure 4A). Breast tumors from BRCA1/2 germline mutation carriers expressed M and LAR subtypes and none were immunomodulatory. Interestingly, two BRCA2 germline mutation related breast tumors were classified as not TNBC. Most of the HGSOC from carriers and non-carriers of germline mutations belonged to MES subtype and none were immunomodulatory. Thus, none of the patients in this cohort, who carried germline mutation in BRCA1/2, developed highly immune-active tumors at diagnosis (Figure 4A). In addition, none of the TNBC were classified as BL1, which is associated with good prognosis and the majority of HGSOC (70%) expressed MES subtype associated with the poor prognosis. This is consistent with the history of the patients in this cohort (lack of response to conventional therapies and progression to metastasis).
To put this data into perspective we examined the classification of all BRCA1/2 germline mutation related breast and ovarian tumors from The Cancer Genome Atlas (TCGA) datasets (Figure 4B and Supplementary Table S9). Consistent with the subtyping in our clinical trial samples, few BRCA1 germline mutation related breast tumors in TCGA database are immunomodulatory (7% versus 21% of TNBC from non-carriers) and most BRCA2 germline mutation related breast cancers do not classify as TNBC (12 out of 15, 80%) (Figure 4B and 4C). The results for ovarian cancers show a similar pattern. However, it is important to emphasize that HGSOC often express multiple signatures. Therefore, classification into mutually exclusive subtypes may be less specific than in other cancers [14]. Nevertheless, BRCA1 /2 germline mutation related HGSOC are not enriched in immunoreactive phenotype (Figure 4D).
Thus, indeed BRCA1/2 germline mutation related tumors do not belong to the most immune active category of breast and ovarian cancers. The data also suggest that BRCA1/2 germline mutation related breast and ovarian cancers express range of phenotypes similar to sporadic cancers and therefore it is unlikely that they represent unique phenotypic identity within TNBC or HGSOC.
However, BRCA1/2 hereditary tumors have unique mutational signature [29] and BRCA 1 tumors have characteristic genomic copy number alterations [30]. Thus, it seems that mostly genotypes, but not phenotypes, make tumors related to BRCA1/2 germline mutation carriers unique.
BRCA1/2 germline mutation related breast and ovarian cancers show relatively low overall immune activity in their microenvironment despite having elevated mutation burden.
The relatively low immune activity in cancers (breast and ovarian) from BRCA1/2 germline mutation carriers is counterintuitive. Tumors with compromised DNA repair usually have a high mutational load and would be expected to generate a high number of neo-antigens [31]. In addition, hypermutated cancers such as melanoma or lung cancer, as well as colon cancer deficient in mismatch repair show positive response to immunotherapy [32–34].
As expected, germline mutation carriers from our clinical trial samples show a higher tumor mutational burden (TMB) compared to non-carriers (Figure 5A) and this is in contrast to global immune activity, which is lower in mutation carriers (Figure 5B). Thus, we asked if there is a correlation between TMB and global immune activity in TCGA.
Within breast cancers, TMB was higher for BRCA1 and BRCA2 germline mutation carriers relative to non-carriers and was also elevated in BL1 subtype. Within HGSOC, TMB was higher only for germline mutation carriers and did not vary among other subtypes (Figure 5C and 5F). Remarkably however, the global immune activity of tumor microenvironments, calculated as averaged expression of genes from 28 meta-gene sets, varied widely between subtypes (Figure 5D and 5G).
Another measure of global immune activity is the leukocyte fraction of tumors. The leukocyte fraction for samples from TCGA is available on the web-based interactive platform: the Cancer Research Institute iAtlas https://www.cri-iatlas.org/about/. iAtlas was designed from extensive immunogenomic analysis and integration of the data for 33 cancer types [35]. The leukocyte fractions in subtypes of hereditary and sporadic TNBC and HGSOC from the TCGA data base showed a very similar pattern to the expression of 28 meta-gene sets (Figure 5E and 5H) and also did not correlate with TMB. Thus, BRCA 1/2 germline mutation related hereditary breast and ovarian tumors, have low overall immune activity within their tumor microenvironments despite their elevated TMB. The data suggest that diversity of immune responses in the microenvironments of hereditary and sporadic TNBC and HGSOC is likely determined by factors other than TMB.
Pattern of genomic instability is different in BRCA1 versus BRCA2 germline related tumors
TNBC and HGSOC are characterized by frequent mutations in TP53 gene and a high degree of genomic instability. Considering that elevated TMB in hereditary breast and ovarian cancers was not associated with high immune activity in the tumor microenvironment, we looked at other measures of instability that potentially could influence immune response in breast and ovarian cancers. Recently, the extensive Pan-Cancer analysis of DNA damage repair (DDR) deficiencies in cancer was published [36] and the results were made available in iAtlas (https://www.cri-iatlas.org/about/). Using this resource, we explored several measures of genomic instability including: mutation load (expressed as non-silent mutation rate and SNV neoantigen count), CNV load (expressed as number of segments and fraction genome altered), aneuploidy and HR deficiency. Genomic instability varies widely between the subtypes of breast and ovarian cancers. As expected, all tumors from germline mutation carriers display high HR deficiency but also relatively low aneuploidy. Consistent with the results shown in Figure 5, breast and ovarian cancers from germline mutation carriers have a relatively high mutation load compared to non-carriers. (Figure 6A and B). BRCA2 related tumors reveal a very different pattern of instability compared to BRCA1 germline related tumors with a low CNV load. This confirms that the characteristic copy number pattern published earlier for hereditary breast cancers applies only to BRCA1-related tumors [37][30]. The relationship between measures of genomic instability and the immune activity in tumors may be complex and require further investigation.
High HR deficiency score characterize all BRCA1/2 germline mutation carriers and is predictive of response to platinum in HGSOC
HR deficiency is particularly relevant for hereditary TNBC and HGSOC. Ovarian cancer has the highest HR deficiency score of all 33 cancers included in TCGA (average value >40) while breast cancers show much lower HR deficiency score (average value >20) (Figure 7A) [36]. However, TNBCs show a HR deficiency score as high as ovarian cancers (average value >40), with the only exception of the LAR subtype (Figure 7B). As expected, breast and ovarian tumors from BRCA1/2 germline mutation carriers have even higher HR deficiency scores (average value >50 for BRCA2 and >60 for BRCA1 mutation carriers) (Figure7 B and C). Similar to TMB, HR deficiency did not correlate with immune activity. However, HR deficiency in ovarian cancers did correlate with platinum sensitivity (Figure 7D). The sensitive and resistant ovarian cancers were selected from TCGA database. Tumors were defined as sensitive if there was no evidence of progression or recurrence at least six months from the date of primary platinum treatment. Tumors that recurred within six months of primary treatment were considered resistant [38]. The ovarian cancers sensitive to platinum had average HR deficiency score of 46.5 and resistant tumors had the score of 36.4. The difference was statistically significant.
Distribution of “BRCAness” in subtypes of breast and ovarian cancers
The term “BRCAness” phenotype was coined to describe sporadic breast and ovarian cancers that behave like hereditary BRCA1/2-related tumors [5,7].
The “BRCAness” characteristics of the subtypes of breast and HGSOC including BRCA1/2 germline mutation carriers from TCGA data base are presented in Table 1 and 2. The most important aspects of “BRCAness” phenotype chosen from literature were as follows: deficiency in HR, high genomic instability, frequent P53 mutations, but infrequent PI3K mutations in breast and ovarian cancers, in addition to basal like classification and high probability of pathological complete response (pCR) in breast cancers [6,7,39,40][28][41,42]. “BRCAness” is most often found in the BL1 and M subtypes of TNBC. Consistent with these results, most of the BRCA1 germline mutation carriers belong to BL1 or M subtype (Figure 4C) and the “BRCA1-like” tumors selected according to copy number criteria also belong mostly to the BL1 and M category [30]. The LAR subtype, on the other hand, has frequent PIK3CA mutations and a low HR deficiency score. The IM subtype does not meet genomic instability criteria, MSL is not basal type and BL2 subtype is characterized by very low pCR. Importantly, BRCA2 germline related tumors do not express any attributes of “BRCAness” except high genomic instability.
Similar analysis was performed for HGSOC subtypes (Table 2). According to our criteria, all subtypes of HGSOC score high on “BRCAness”.
PD-L1 expression reflects overall magnitude of the immune response in breast and ovarian cancers
PD-L1 is the target for anti-PD-L1 antibodies, which are currently being examined in a phase II clinical trial (NCT02849496). PD-L1 RNA expression was significantly higher in samples from non-carriers of germline mutations compared to the carriers in our clinical trial samples (#NCT01623349) (Figure 8A). Thus, higher overall immune activity corresponded with higher expression of this marker. We verified the expression of this marker in all subtypes of TNBC and HGSOC from TCGA database. All tumors expressed the protein, and the pattern of expression followed the pattern of overall immune activity in all samples including those from BRCA1/2 germline mutation carriers (see also Figure 5 C-H). The IMR subtype of HGSOC had the highest expression of PD-L1.
The immune response patterns in TNBC and HGSOC
The immune landscape of 33 cancer types was recently published and made available on the web-based interactive platform [35], Cancer Research Institute iAtlas https://www.cri-iatlas.org/about/). They identified six universal intratumor immune states or response patterns. Briefly, C1, wound healing subtype, have elevated expression of angiogenic genes and high proliferative rate, C2, INF-g subtype, have the highest M1/M2 macrophage polarization, C3 is an inflammatory subtype, C4 is lymphocytes depleted type displaying a more prominent macrophage signature, C5 is an immunologically quiet type and exhibit the lowest lymphocyte and highest macrophage response dominated by M2 and finally C6 is a TGF-b dominant type. When we applied the signatures for intratumor immune types (C1-C6) to our clinical trial samples, we found that the majority of non-carriers expressed C3 (inflammatory subtype), while majority of carriers expressed C1 (wound healing) subtype (Figure 4A and Figure 8D). The composition of the immune microenvironments within TNBC and HGSOC from TCGA varied widely, but almost universally the predominant subtypes were C2 (INF-g and macrophage-enriched) and C1(wound healing). Some HGSOC expressed also C4 (lymphocytes depleted) subtype. Interestingly, two the most “BRCAness” expressing TNBC showed very different immune environments. BL1 tumors with higher overall immune activity relative to M tumors are predominantly (82.8%) associated with macrophage-enriched (C2) immune signature, while M tumors, which have overall very low immunoactivity, are predominantly (77.8%) associated with wound healing (C1) signature (Figure 8D).