Over the last two decades many new immunotherapy approaches to the treatment of cancer have entered clinical development due to the increased understanding of the mechanisms of antigen presentation, lymphocyte recognition, functions of the innate immune system, and the means of regulation of these responses and the means by which tumors can circumvent these responses. Many of these investigations have led to agents approved for standard clinical use, including infusional cytokine therapies for melanoma or renal cell cancer, intravesical BCG therapy for bladder cancer, and most recently an active cellular therapy targeting prostatic acid phosphatase (PAP, sipuleucel-T, Provenge, Dendreon) for patients with advanced metastatic prostate cancer. Many other agents have demonstrated benefit in large clinical trials, and approval is anticipated in the case of a monoclonal antibody targeting a T-cell checkpoint inhibitor targeting CTLA-4 (ipilimumab, Bristol-Myers Squibb) for advanced melanoma.
Ultimately, for these agents to be clinically approved there needs to be a demonstration that these treatments are relatively safe and patient care and outcome are positively affected. However, there is also an increasing recognition that some of these agents, while likely safe, may best be used in combination with other immune-activating or conventional therapies. This has presented challenges for evaluating these agents using traditional paradigms for clinical development. Consequently there is a need to identify markers of biological response, ideally associated with clinical outcome, but permitting an evaluation of biological effect of these agents used in combination. In the case of antigen-specific vaccines, it has been relatively straightforward to evaluate immune responses to the target antigen as a “biomarker” of immunological efficacy. Unfortunately, there are few instances in which target antigen immune response has been associated with clinical benefit. The situation is more difficult for broadly active immune modulating agents such as T-cell checkpoint inhibitors, including antibodies targeting CTLA-4 or PD-1, or TLR agonists, in which appropriate biomarkers of response have been more elusive. Studies with anti-CTLA-4 monoclonal antibodies, in particular, have sought to identify whether amplification of other T-cell costimulatory molecules [1
], or antibodies to defined antigenic tumor-associated proteins [2
], might be useful as biomarkers. For whole cell tumor vaccines where there is not a specific, defined antigen being targeted, surrogate antigens known to be expressed by the tumor vaccine have been used as a means of monitoring immune responses from the vaccine [4
]. The use of immunologically recognized surrogate antigens, including HER-2/neu, MUC1, and p53, has been possible in the case of breast cancer where T-cell and IgG responses to these antigens have been identified. However it is unknown whether responses to these antigens can be useful to study agents in combination or whether changes in responses to these antigens are associated with clinical outcome.
Over the last several years we have used SEREX- (serological analysis of recombinant cDNA expression libraries) based studies to identify immunologically recognized proteins expressed by normal and malignant prostate tissue that might serve as targets for anti-tumor vaccines [7
]. In particular, we have evaluated the targets of IgG responses in patients with chronic prostatitis or autoimmune disorders [8
], patients with prostate cancer treated with immune-modulating therapies [10
], and IgG responses to cancer-testis antigens in patients with prostate cancer [11
]. Over the course of these studies we have effectively identified hundreds of immunologically recognized proteins associated with prostate tissue and/or recognized by patients with prostate cancer. While the identification of hundreds of proteins presents challenges in prioritization for the development of antigen-specific vaccines, we previously questioned whether these antigens might also have diagnostic value with IgG responses being able to distinguish individuals with prostate cancer (or other inflammatory conditions of the prostate) from men without prostate disease. Other groups have similarly reported that IgG responses to tissue-associated antigens might have diagnostic value in identifying patients with prostate cancer [13
] or nonsmall cell lung cancer [14
]. We have previously reported that a subset of 23 of these antigens were recognized in patients with prostate cancer as well as individuals with symptomatic prostatitis, suggesting that such autoantibody signatures might be useful to identify inflammatory conditions of the prostate, and potentially in a premalignant setting [15
In the current report, we hypothesized that this same panel of previously identified prostate-associated antigens might be used as a monitoring tool to assess immune responses elicited following immune-modulating therapy. While B-cells or IgG production might not be an intended target of a particular therapeutic approach, IgG responses are often elicited with concurrent T-cell activation. We reasoned that IgG responses are easier to measure compared with antigen-specific T cells, and might be more stable over time in the peripheral blood compared with T-cell frequencies. Moreover, the identification of “off-target” IgG immune responses might further serve as an indication of “antigen spread” with secondary antigens recognized following immunological targeting and thus be more relevant to developing biomarkers associated with favorable clinical responses. To detect antibody responses to previously defined antigens, we applied a similar phage immunoblot approach evaluating IgG responses to multiple antigens simultaneously [15
]. These types of complex biomarker data sets are historically very difficult to work with for two reasons: first is the complexity associated with biological networks; second is the challenge of infrequent observation of immune biomarkers in a complex system. As such, the identification of useful biomarkers in data sets such as this study can be very challenging. In this paper, we sought to evaluate the use of machine-learned Bayesian Belief Networks (ML-BBNs) as a method for identifying potentially promising biomarkers and potential biomarkers networks [16
]. We sought to train several ML-BBNs to identify promising biomarkers and then use these networks to select a subset of features to train a network of immune biomarkers as they related to observed declines in serum prostate-specific antigen (PSA). Our objective was to demonstrate the feasibility of this method to identify promising early biomarkers of immune response to vaccine therapies in our data.
For the current studies, sera samples were collected prior to treatment and after three months of treatment from three separate trials, one in which patients (n
= 34) were treated with androgen deprivation (ADT) therapy only, a standard therapy known to elicit prostate-associated immune responses [18
], a trial in which patients with castrate-resistant prostate cancer (n
= 31) were treated with a viral vaccine encoding PSA (PSAV) [5
], and one in which patients with early recurrent prostate cancer (n
= 21) were treated with a plasmid DNA vaccine encoding PAP (PAPV) [6
]. Patients treated with vaccines were subclassified as immunologic or clinical “responders” based on previously reported criteria to distinguish these groups. We report here that IgG immune responses could be detected to individual antigens, and as long as one year after therapy the recognition of specific antigens was associated with individual treatments. The evaluation of IgG responses to groups of antigens at three months suggests that predictive models might be developed with diagnostic potential. These findings support the concept of using measures of “antigen spread” as biomarkers of immunological efficacy for immune-active therapies, and IgG responses to panels of tissue-associated antigens as measures of this antigen spread.