CD19 is a transmembrane protein present on the surface of all B lineage cells except plasma cells. Likely due to its role in B-cell receptor signaling, CD19 expression is rarely lost upon B-cell neoplastic transformation, and CD19 immunotherapeutic targeting has been proposed decades ago.1 Indeed, spectacular results have been achieved by anti-CD19 chimeric antigen receptor T cells (CART-19), which currently represent the most advanced personalized cell-based immunotherapeutic tool.2–6 Depending on the cancer type, the initial remission rate induced by CART-19 in patients with B-cell cancer can be as high as 90%.2 3 However, after an initial positive response, 30%–60% of CAR-treated patients will relapse. While this is mostly caused by suboptimal persistence of CART-19 cells, up to 20% of these relapses are due to antigen loss on the surface of neoplastic cells.7 8

Some of the possible means of CD19-negative tumor escape are already known. First, a dramatic upregulation of alternatively spliced CD19 transcripts lacking the epitope targeted by CART-19 was described.9 Later, direct CD19 mutations leading to a truncated protein with non-functional or absent transmembrane domain10 extended the landscape of genetic means of antigen loss. Occasionally, some of the genetic mechanisms might be at play already prior to CART-19 treatment, leading to a small CD19-negative population already detectable at baseline.6 11 Such CD19-negative clones are selected under the treatment pressure, expand and drive the subsequent relapse. The immune pressure, triggered by CART-19 treatment, combined with the inherent B-cell lineage’s plasticity can also induce antigen-negative relapse by lineage switch,12 13 for example, from lymphoid to myeloid. Finally, trogocytosis has recently emerged as a cause of invertible antigen loss provoked by CAR T cells.14 Herein, using one in vivo and various in vitro disease models of chronic lymphocytic leukemia (CLL), we show that CD19 epigenetic repression can also cause antigen escape, and this mechanism is partially reversible using demethylating agents.

Using HG3 cell line in NSG mice, we recently established an in vivo CLL disease model,15 which we treated with in house produced CART-19 or control T cells. Unlike the mice injected with control T cells, most of CART-19-treated mice overcame the progressive disease that developed after ~2 weeks of tumor cells’ injection and mostly affected spleen, liver and kidneys. However, up to 70% of the surviving mice recurred with subcutaneous tumors weeks after the initial positive response (figure 1A, online supplemental figure 1). Due to the striking difference in the neoplastic localization, we set out to characterize the malignant cells forming the recurring tumors. A prominent downregulation of CD19 expression on B cells isolated from these tumors was apparent both by flow cytometry and by IHC (figure 1A,B respectively).

Importantly, we were able to separate CD19-negative and CD19-positive cells from two such tumors by fluorescence-activated cell sorting (FACS). Of note, only <0.2% of the originally infused HG3 cells were detectable in the CD19-negative gate used for sorting (online supplemental figure 2). Data from one of these tumors are depicted in the main figures, while results from the other are summarized in online supplemental figure 3. We cultivated the sorted cells in vitro for several weeks and noticed that CD19 surface expression was gradually reconstituted over time in the cell culture originally containing only CD19-negative cells (figure 1C; repeatedly observed in 2 rounds of CD19-negative cell FACS sorting). Importantly, while freshly sorted CD19-negative cells did not express CD19 mRNA (figure 1D, online supplemental figure 3A), its loss was only temporary, and apparently, it was not due to alternative splicing described previously9 (figure 1E, online supplemental figure 3B) or aberrant expression of PAX5, transcription factor known to regulate CD19 expression18 (online supplemental figure 4).

Based on these results, we suspected the mechanism driving antigen-negative escape to be more flexible than those described previously, likely involving epigenetic orchestra. In this regard, CD19 promoter does not contain a CpG island per se, but there are several CpGs spanning the transcription initiation site. Indeed, when we studied the CD19 promoter methylation of CD19-negative and CD19-positive cells sorted from a relapsed tumor by direct bisulfite sequencing, we found that the entire assessed region was affected by hypermethylation in the CD19-negative cells (figure 2A, online supplemental figure 3C). To confirm the causal character of this methylation, we treated freshly sorted CD19-negative cells with demethylating agent 5-aza-2′-deoxycytidine (1 µM or 5 µM) in vitro. After 48 hours, increased CD19 expression on the surface of treated cells was apparent by flow cytometry (figure 2B, online supplemental figure 3D), and this was accompanied by a decrease in methylation of some of the CpGs assessed by the promoter methylation bisulfite sequencing (figure 2C). We next assessed whether 5-aza-2′-deoxycytidine treatment could prevent the loss of CD19 triggered by CART-19 in our NSG mouse model. Notably, dual treatment of CART-19 combined with 5-aza-2′-deoxycytidine in vivo indeed led to increased CD19 expression on leukemic cells compared with CART-19 monotherapy (figure 2D).

Finally, to establish if this epigenetic escape mechanism was solely specific to HG3 cells, we cultured B-cell lines (HG3, RAMOS, MEC1) as well as primary cells from 2 CLL patients with CAR/control T cells for 24 hours in vitro and sorted out CD19-negative and -positive cells for the promoter methylation study. All three studied immortalized cell lines, as well as primary CLL cells, employed the same epigenetic downregulation of target antigen according to our results (figure 3A,B respectively). Since MEC1 cells did not fully abrogate CD19 expression in response to CART-19 cells, we exposed them solely to CART-19 and sorted out the CD19-negative and CD19-positive fraction from this single coculture. Furthermore, we mixed HG3 with CART-19 at 1:1 ratio for only 12 hours, to achieve a partial response, and sorted the most CD19-negative and the most CD19-positive cell fraction. In both these approaches, CD19-negative cells clearly showed a strong hypermethylation of CD19 promoter in contrast to their CD19-positive counterparts (figure 3C). Of note, CD19 downregulation in these short-term coculture experiments was highly transient, as within 48 hours CD19 levels were readily restored likely due to the short-term character of the immunotherapeutic pressure enforced by CART-19 as opposed to the in vivo situation (figure 3D).

Disease relapse represents the key limitation of CART-19 response durability. While optimized CAR designs, which would increase the specificity and persistence of effector cells, are undoubtedly critical to overcome this hurdle,19 it has become apparent that some tumors will still be able to escape through an antigen-loss mechanism.9 10 12–14 Herein, we have identified epigenetic repression as an additional means of CD19 downregulation. This mechanism was causing relapse in vivo in a recently established CLL disease model15 as well as in several in vitro models. According to our results, this could well be the first means of antigen-negative escape to CAR T-cell immunotherapy partially reversible with drugs to be discovered.

Until today, in vivo relapse models to CART-19 immunotherapy have been described only by a handful of studies.13 14 Of these, one uses limiting doses of CART-19 in acute lymphoblastic leukemia (ALL) mouse model to mimic relapse,14 while the other takes advantage of a syngeneic murine ALL model to induce lineage switch in CART-19 recipient mice.13 Herein, we had a unique opportunity to study CD19-negative recurrence in immunocompromised mice injected with a CLL cell line HG3. Despite the early potency and persistence of functional CAR T cells, up to 70% of CART-19-treated animals eventually developed a CD19-negative recurrence. To the best of our knowledge, this is the first in vivo antigen loss CLL recurrence model to CART-19 to be developed.

We were able to isolate and culture CD19-negative cells causing recurring tumors and perform some molecular analyses, which ruled out alternative splicing9 as the causal escape mechanism. On top of that, CD19 expression was steadily regained in in vitro culture. These were unexpected findings as they clearly contradicted all so far known mechanisms of target loss; neither genetic escape10 20 nor lineage switch12 13 were likely to be resolved so promptly during in vitro culture, and trogocytosis14 could not be at play since tumor-sorted B cells were cultured and re-sorted in absence of T cells.

Instead, we found an inverse correlation between CD19 promoter DNA methylation and its expression in the relapsed tumors, as well as in in vitro coculture experiments with B-cell immortalized cell lines and primary CLL cells. On top of that, when cells with methylated CD19 promoter were treated with a demethylating drug, an increase in the CD19 expression accompanied by a decrease in DNA methylation was observed, validating the repressive character of methylation. Interestingly, DNA methylation of the region studied herein had already been assessed once in CD19-negative relapse samples from patients with ALL, but no promoter methylation had then been found that would explain the loss of CD19 expression in the primary samples.9 However, ALL and CLL are two biologically distinct entities, and differences in gene regulation can be expected and have already been described.21 22

Finally, we cannot completely rule out the possibility of selecting a pre-existing CD19-dim or CD19-negative HG3 cell population at the time of in vivo recurrence as such event has already been described in the clinics.6 11 23 Nevertheless, it is highly unlikely that the entire starting CD19-positive target cell population would be replaced by CD19-negative cells within just 24 hours in in vitro coculture experiments, thereby undermining the selection theory. Notably, given how readily CD19 is reconstituted on surface of CD19-negative cells sorted after 24 hours of in vitro coculture, other epigenetic means (histone modifications, etc) could well be involved in stabilizing the effect of DNA methylation on CD19 repression in vivo.

These results indicate that at least a part of B-cell derived tumors could apply epigenetic silencing to escape CD19 immunotherapeutic targeting. Importantly, this mechanism could be at least partially reversed using demethylating agents. We propose that the use of agents like 5-aza-2′-deoxycytidine in combination with CAR T-cell therapy might prevent such CD19-negative relapses and thereby improve the response rate in these scenarios.