Synergy of BCL2 and histone deacetylase inhibition against leukemic cells from cutaneous T-cell lymphoma patients

Cutaneous T-cell lymphoma (CTCL) is a form of non-Hodgkin lymphoma (NHL) with a variety of clinical manifestations ranging from mycosis fungoides (MF; characterized by localized skin patches, plaques, and tumors) to leukemic CTCL, where malignant T cells may predominate the peripheral lymphocyte compartment.¹  In advanced stages, CTCL is a fatal disease²  that is incurable with conventional therapies, with blood involvement portending poorer survival outcomes.³  With rare exceptions in cases of hematopoietic cell transplantation,⁴  the overall response rates for novel agents including retinoids, histone deacetylase (HDAC) inhibitors, and pralatrexate range from 30% to 50% and are generally not durable.⁵  There remains an unmet medical need for new and more effective treatments.

Recent studies^6-10  have made significant strides in understanding the molecular pathogenesis of CTCL, most notably via exome sequencing and expression analysis. These analyses have shown a predominance of gene copy-number alterations (GCNAs) over single-nucleotide variant (SNV) mutations. The categories of genetic alterations include changes in the behavior of the malignant T-cell population and their imprint on the immune system, and suggest clustering under 3 major pathways: constitutive T-cell activation, resistance to apoptosis/cell-cycle dysregulation, and DNA structural/gene expression dysregulation. With this wellspring of new information, recently discovered and repurposed agents targeting pathways or specific gene mutations may be screened as a patient-specific treatment algorithm is developed.

With 30% of drugs in clinical trials failing due to lack of efficacy,¹¹  a focus on expanding indications of new molecular therapies allows us to leverage established safety profiles to fasttrack new treatment options for patients. One such opportunity for the repurposing of existing therapies involves the dysregulation of B-cell lymphoma 2 (BCL2)-driven apoptotic pathways in CTCL. Four common gene alterations identified in CTCL are STAT3 amplifications, STAT5B amplifications, P53 deletions, and CTLA4 deletions, the frequency of which was previously validated by our group in the development of a new diagnostic tool, an 11-gene fluorescence in situ hybridization (FISH) panel.¹²  Each of these mutations has been linked to the inhibition of apoptosis through the upregulation of BCL2 transcription, in turn leading to increased BCL2 activity and dependence.^13-20 

Venetoclax (ABT-199) is a BCL2 homology 3 (BH3)-mimetic, BCL2-selective inhibitor without additional cross-reactivity with BCL-XL, BCL-W, or myeloid cell leukemia 1 (MCL1).²¹  BCL2 family proteins are regulators of the intrinsic apoptosis pathway, in which cell death is caused by the permeabilization of the outer mitochondrial membrane, release of cytochrome c, and the activation of caspases.²²  These proteins additionally regulate autophagy via the binding of Bclin-1.²³  BCL2 itself is an antiapoptotic protein that promotes cell survival by sequestering proapoptotic factors. Venetoclax was first approved by the US Food and Drug Administration in 2016 and received accelerated approval for the treatment of relapsed or refractory chronic lymphocytic leukemia (CLL) with 17p deletion and is the only BCL2 inhibitor that has received approval by the US Food and Drug Administration for clinical use.²⁴  Venetoclax is also currently undergoing trials for follicular lymphoma, diffuse large B-cell lymphoma, acute myeloid leukemia, multiple myeloma, Waldenström macroglobulinemia, and NHL (excluding CTCL).

The potential for BCL2 inhibition in the treatment of CTCL may extend beyond its use as a monotherapy. Combined inhibition of BCL2 members and HDACs has shown synergistic effects in CTCL²⁵  and other malignancies, including mantle cell lymphoma²⁶  and glioblastoma.²⁷  Romidepsin and vorinostat are HDAC inhibitors²⁸  that have widespread effects including modulation of gene expression, induction of cellular differentiation, and modulation of apoptosis effector genes.²⁹  One mechanism of action of apoptosis is via alterations in the expression of the intrinsic apoptosis pathway, namely the BCL2 family genes,^30-33  including upregulation of proapoptotic factors such as BH3 interacting-domain death agonist (BID) and Bcl-2-like protein 11 (BIM), and downregulation of the antiapoptotic factors such as BCL-XL and MCL1.^34-36 

Through in vitro testing of CTCL patient-derived blood samples, we demonstrate the potential of BCL2 inhibition by venetoclax as a novel treatment of CTCL. Furthermore, we reveal that this effect is synergistically potentiated via HDAC inhibition afforded by romidepsin or vorinostat. These results suggest that BCL2 inhibition, alone or in combination with HDAC inhibition, represents a viable novel strategy in the treatment of advanced CTCL with blood involvement.

MyLa 2059,³⁷  Sez4,³⁸  HH,³⁹  and Hut78⁴⁰  are well established CTCL cell lines. MyLa 2059 and HH were cultured in RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS), glutamine (2 mM), penicillin (100 U/mL), and streptomycin (100 μg/mL) at 37°C, 5% CO₂, and 95% humidity. Sez4 was cultured in RPMI 1640 medium supplemented with 10% heat-inactivated FBS, glutamine (2 mM), penicillin (100 U/mL), and streptomycin (100 μg/mL), as well as interleukin 2 (IL2) (20 ng/mL; R&D Systems) at 37°C, 5% CO₂, and 95% humidity. Hut78 was cultured in Iscove modified Dulbecco medium (IMDM) medium supplemented with 20% heat-inactivated FBS, glutamine (2 mM), penicillin 100 U/mL, and streptomycin (100 μg/mL) at 37°C, 5% CO₂, and 95% humidity. All media and supplements were obtained from Invitrogen (Carlsbad, CA).

Blood samples were obtained from CTCL patients at the Yale Cancer Center. Written informed consent was obtained from all subjects in accordance with the Yale Human Investigational Review Board. Peripheral blood samples from consenting patients and healthy donors were collected in lithium heparin tubes. Peripheral blood mononuclear cells (PBMCs) were separated from whole blood by Ficoll density gradient. CD4⁺ T cells were selected with the MACS CD4⁺ negative selection kit (Miltenyi Biotec, Bergisch Gladbach, Germany), per the manufacturer’s protocol, and supplemented with antibodies to select for CD26⁻ and/or CD7⁻ cells, depending on the known aberrant phenotype of the patient. Aliquots of unfractionated PBMCs and purified samples were obtained to assess purity and verify expression of phenotypic markers. Following isolation, cells were cultured in RPMI 1640 medium supplemented with 10% heat-inactivated FBS, glutamine (2 mM), penicillin (100 U/mL), and streptomycin (100 μg/mL), as well as IL2 (10 ng/mL), IL7 (5 ng/mL), IL15 (10 ng/mL), and IL13 (10 ng/mL; all cytokines from R&D Systems) at 37°C, 5% CO₂, and 95% humidity.

Unfractionated PBMCs and isolated malignant T-cell populations were analyzed using flow cytometry. All samples were matched with an isotype control. The following monoclonal antibodies were included: CD3 from BD Biosciences (San Jose, CA); CD4, CD7, and CD26 from eBioscience (San Diego, CA); and TCR-Vβ from Beckman Coulter (Brea, CA). Cells were blocked with human TruStain FcX from BioLegend (San Diego, CA) for 10 minutes then incubated with selected antibodies for 20 minutes at 4°C. Cells were then washed 3 times and fixed in 1% paraformaldehyde. Flow cytometry was conducted on the Stratedigm-13 from Stratedigm Inc (San Jose, CA). All flow cytometric data were analyzed using FlowJo software (v10; FlowJo, LLC).

Following isolation of the desired cell population, cells were incubated at a density of 100 cells per microliter in 100 μL of media, for a total of 10⁵ cells per well, in a black optical 96-well plate. Cells were incubated for 72 hours with venetoclax (3 nM to 100 µM), romidepsin (0.1-10 nM), and/or vorinostat (0.1-10 µM). The CellTiter-Glo Luminescent Cell Viability Assay (Promega, WI) was used to measure the number of viable cells in culture based on quantitation of the adenosine triphosphate (ATP) present. Plates were read using the Perkin Elmer Victor Light Luminescence Counter (Waltham, MA). Drug concentrations were applied in approximate half-log10 increments to patient samples, and in twofold increments for cell lines. Cell luminescence was normalized to a vehicle control containing 0.2% dimethyl sulfoxide (DMSO) and corrected for media.

Patient samples were incubated for 24 hours as described in “Cell viability assay.” Following incubation, the Promega Caspase-Glo 3/7 assay (Madison, WI) was used to quantitate caspase activity, as per the manufacturer’s protocol. Plates were read using a PerkinElmer Victor X Light Luminescence Counter.

Stocks of 50 nM venetoclax, 20 mM vorinostat, and 20 nM romidepsin from ApexBio (Houston, TX) were prepared in DMSO and stored at −20°C. Camptothecin (10 μM) from Sigma-Aldrich (St. Louis, MO) was used as a positive control for both the viability assay and the Caspase-3/7 assay.

For the cell viability assay and Caspase-3/7 assays, each drug concentration was performed in quadruplicate, and the mean values were plotted with their respective standard deviation. The mean inhibitor concentration was determined using GraphPad Prism (version 7.01). Patient samples were grouped using hierarchical clustering with the Ward method⁴¹  in R (version 3.0.1). Combination index (CI) values and standard errors were calculated using the Chou-Talalay method⁴²  in Microsoft Excel.

RNA was extracted using the All-Prep DNA/RNA Micro kit for patient samples, or with the All-Prep DNA/RNA Mini kit for cell lines from Qiagen (Hilden, Germany) as per the manufacturer’s specifications. RNA was quantified using the BioTek Epoch spectrophotometer. RNA (2 μg) was converted to complementary DNA (cDNA) using the High Capacity Reverse Transcription kit from Applied Biosystems (Foster City, CA) per the manufacturer’s instructions. Patient cDNA samples were then preamplified using the Applied Biosystems TaqMan Pre-Amplification Master Mix as per the manufacturer’s instructions. Gene expression was quantified using the TaqMan primers and Gene Expression Master Mix and the Applied Biosystems 7500 Real Time Polymerase Chain Reaction (PCR) System. Samples were characterized in duplicates. Obtained cycle threshold (Ct) values were normalized to hypoxanthine-guanine phosphoribosyltransferase 1 (HPRT1) expression and expression differences relative to healthy controls were calculated using relative quantification, RQ = 2^−ΔΔCt.

To begin to assess the potential of the BCL2 inhibitor venetoclax in the treatment of CTCL with blood involvement (B1 or B2), peripheral blood samples were obtained from 25 patients diagnosed with CTCL who have a malignant cell population measured by flow cytometry. Patients were classified by their initial diagnosis as erythrodermic/Sézary syndrome (13 of 25), MF (10 of 25), or folliculotropic MF (2 of 25) (Table 1). Patient treatment statuses ranged from extracorporeal photopheresis (ECP) with or without adjunct therapies (18 of 25), to multimodal therapies (6 of 25), to cytoreductive therapy (1 of 25). Each patient sampled underwent clinical flow cytometry, from which blood parameters were calculated (Table 2). Patients were additionally tested for clonality by either PCR-based T-cell receptor (TCR) gene rearrangement studies or the presence of a clonal Vβ antigen detectable by flow cytometry, with 25 of 25 patient samples positive for clonality based on 1 or both tests. Eight patients were classified as B1, and the remainder were classified as B2 based on the 2007 International Society for Cutaneous Lymphomas (ISCL) classification⁴³  and the 2016 Gibson et al criteria.⁴⁴ 

We observed the cytotoxic/cytostatic effects of venetoclax in all 25 patient samples after 72 hours of incubation as measured by the change in cell viability via media ATP quantitation. Patient samples demonstrated a wide range of sensitivities to venetoclax, with mean 50% inhibitory concentrations (IC₅₀) ranging from picomolar (85 pM) to micromolar (6.6 µM) levels. The observed heterogeneity in sensitivity to venetoclax is not dissimilar to the variation observed in T-cell acute lymphoblastic leukemia (T-ALL)⁴⁵  or CLL.²¹  The most sensitive samples compare favorably to preclinical studies in B-cell lymphomas, the results of which have since been validated via clinical trial.²⁴  To better study the significance and origins of the observed variation, we used hierarchical clustering based on their responsiveness to venetoclax to divide the tested samples into 2 groups, low and high responders. The majority of tested samples (20 of 25) were defined as high responders with a group mean IC₅₀ of 79.0 nM compared with the low responders (5 of 25) with a group mean IC₅₀ of 3.2 µM. Shown in Figure 1Bi are dose-response curves for 3 representative samples: the most sensitive, the median, and the most resistant sample.

Four CTCL cell lines, MyLa 2059, Sez4, HH, and Hut78, were analyzed for sensitivity to venetoclax (Figure 1Bii). All 4 cell lines demonstrated relatively low sensitivity to venetoclax with IC₅₀s of 2.47 µM, 7.4 µM 10.2 µM, and 33.6 µM, respectively. In addition to the differences in IC₅₀, cells lines and patients were observed to have different patterns in their dose-response curves. Cell lines demonstrated a significantly higher Hill slope relative to patient samples (P < .0001), suggesting distinct mechanisms of action between cell lines and patients (Figure 1C). Among patient samples, there was a statistical trend noted between response status and Hill slope (P = .07; data not shown). Hypothesizing an association between tumor burden and increased sensitivity to induction of apoptosis via inhibition of BCL2, we compared the mean inhibitory concentration of venetoclax between patients classified as B1 vs B2 (ISCL classification),⁴³  as well as between patients classified as MF (nonerythrodermic) vs Sézary syndrome (erythrodermic), and found a higher sensitivity in patients with more advanced disease, and in patients with Sézary syndrome (Figure 1D-E).

To study the differences in response to venetoclax among both primary patient samples and cell lines, the baseline genetic expression of 4 BCL2 family members was measured using quantitative reverse-transcriptase PCR (RT-PCR) and expressed relative to the mean expression of the healthy controls (Figure 2A). Significant differences were noted in BCL2 expression between high and low responders as well as high responders and cell lines, whereas none were observed between low responders and cell lines. Patient samples expressed lower amounts of BCL2 than healthy controls (on average 29% less), but this difference was not statistically significant (P = .19). Also notable was the similar expression of BCL2L1 (the gene encoding BCL-XL) among patient samples and cell lines, with both sets demonstrating no difference relative to the healthy controls. Significant differences were noted in the expression of BCL2L11 (the gene encoding BIM) among cell lines, with high expression noted in Hut78, and low or undetectable expression of BCL2L11 messenger RNA (mRNA) in the MyLa, Sez4, and HH lines (supplemental Figure 2A, available on the Blood Web site).

For the 20 patient samples examined, IC₅₀ values were plotted as a function of relative expression of BCL2 member mRNA. Although a negative correlation was found between baseline expression of BCL2 and IC₅₀ (Spearman r = −0.52; P = .018) (Figure 2B), mRNA expression of other BCL2 family members (BCL2L1, MCL1, BCL2L11, BAX) did not correlate with the IC₅₀ values (supplemental Figure 2B). Furthermore, we examined ratios of expression of BCL2 to the other tested BCL2 family members and found that higher ratios of BCL2 to BCL2L1 correlate with greater sensitivity to venetoclax in vitro, whereas ratios of BCL2 to MCL1 and BCL2 to BCL2L11 do not (Figure 2C). These results point to the importance of both BCL2 and BCL2L1 expression, and indeed relative expression of one another, to the survival of malignant T cells. Additionally, the limited correlations of both BCL2 and the BCL2-to-BCL2L1 ratio to sensitivity suggest that a more complex relationship between BCL2 expression and inhibition of apoptosis may exist.

Fifteen of the CTCL patient samples were tested for synergy between venetoclax and 2 HDAC inhibitors, vorinostat and romidepsin. Each of the 3 drugs was tested individually and in combination, and IC₅₀ values and Hill slopes were calculated. From these dose-response curves, the CI was calculated using the Chou-Talalay method in Microsoft Excel, with a CI <1 indicating a synergistic interaction.⁴⁶  For combinations of venetoclax with either HDAC inhibitor, synergy was observed at the higher dose range (Figure 3A-B). Ninety-three percent of tested samples demonstrated synergy of response with combinations of venetoclax and vorinostat at 10% viability (0.9 fraction affected), whereas 73% demonstrated synergy of response between venetoclax and romidepsin (Figure 3C-D; supplemental Figure 1A; supplemental Table 1). No correlation was found between either expression of BCL2 or sensitivity to venetoclax and the degree of synergy (data not shown). Additionally, no differences were found between MF patients or Sézary syndrome patients and degree of synergy, for combinations of venetoclax with either HDAC inhibitor (supplemental Figure 1Bi-ii).

To investigate the mechanism of action, the induction of apoptosis at 24 hours was measured by caspase 3/7 activation. Dose-response curves were again produced for venetoclax, vorinostat, and romidepsin individually and in combination. The effect of combination therapy upon caspase 3/7 activity displayed a dose-dependent action (Figure 3E). Our measurements of caspase activation support the observations made with the cell-viability assay regarding the form of interaction between venetoclax and the HDAC inhibitors, in which there are disparate effects at low and high drug concentrations. At low doses of venetoclax and either HDAC inhibitor, antagonism was observed. In contrast, we observed a significant increase in caspase 3/7 activity with combination therapy compared with either drug alone at high doses (Figure 3F).

Synergy studies were similarly performed on the cell lines (Figure 4A-B). The only cell line to display significant synergy was MyLa 2059 when venetoclax was combined with vorinostat. In both the HH and Hut78 cells, venetoclax and vorinostat had antagonistic responses. In combinations of venetoclax and romidepsin, only the HH cell line demonstrated synergistic sensitivity. Unlike the response of patient cells, which are well modeled by the Hill and median-effect equations, the cell lines display significant deviations from the expected responses, especially at high doses. This may be attributed to the heterogeneity of the cell lines, which are not necessarily clonal, or to the highly cytotoxic effects of drugs such as romidepsin at higher doses.^5,47  Thus, our ability to detect synergy at these dose ranges is limited.

To provide insight into the mechanism of action of synergy and antagonism between venetoclax and the HDAC inhibitors romidepsin and vorinostat, we incubated the 4 CTCL cell lines for 24 hours with 5 μM vorinostat or 5 nM of romidepsin before isolation of RNA and measured the relative expression of 7 BCL2 family members compared with a DMSO vehicle control via quantitative RT-PCR (qRT-PCR) (Figure 5A). Following exposure to vorinostat, there was a 40-fold increase in BCL2L11 expression and a 12-fold increase in BMF expression. The dramatic increase in BCL2L11 mRNA is critical to the function of BCL2, which acts to inhibit apoptosis via direct binding to BIM. This shift in balance toward proapoptotic factors is not only 1 of the methods of action of vorinostat, but sheds light on interactions between venetoclax and HDAC inhibition. Furthermore, we were able to detect a statistically significant trend between the change in expression of BCL2 in response to vorinostat and the calculated CI at a fraction affected of 0.9 (Figure 5B). MyLa, the only cell line in which BCL2 expression was increased, was the only cell line to demonstrate significant synergy. Romidepsin caused a 13-fold increase in BMF and fourfold increase in BCL2L11 expression. No significant changes in expression of BCL2 were noted following exposure to romidepsin in any of the 3 cell lines.

Guided by recent studies^6,7,48-51  that suggest that increased resistance to apoptosis in CTCL is at least partly due to gene alterations that may increase the transcription of BCL2, and encouraged by the success of venetoclax in the treatment of CLL,⁵²  we sought in this study to explore the potential of venetoclax for the treatment of CTCL. We identified a group of patients, the high responders, with exquisite sensitivity to venetoclax at a mean IC₅₀ of 79.0 nM. This range is comparable to the range of sensitivities, <300 nM previously observed in BCL2-dependent immature CD4⁻CD8⁻ T-ALL subtypes. Three clinical trials^53-55  investigating the treatment of CLL and NHL found the maximum plasma concentrations for 400 mg, 800 mg, and 1200 mg to be 2.2 µM, 3.5 µM, and 4.0 µM. These values far exceeded the in vitro mean inhibitory concentrations observed in our study. We are further encouraged as the maximum tolerated dose of venetoclax has not yet been well established, necessitating dose escalation beyond 1200 mg.^56,57  Indeed, the most significant adverse events in the trials for CLL were deaths due to tumor lysis syndrome.⁵⁸ 

Having established the response of primary patient samples to venetoclax in vitro, we next searched for a biological assay that would predict sensitivity and distinguish high responders from low responders. BCL2 protein and mRNA expression have been found to have an inverse relationship with sensitivity to venetoclax in vitro CLL,⁵⁹  NHL,²¹  and T-ALL.⁴⁵  Our data indicate that a similar relationship exists in CTCL, additionally supported by the significant difference in expression of BCL2 among high and low responders. Other work has shown that expression of BCL2L1 and MCL1 contributes to resistance to venetoclax,^24,34,36,60  but our work suggests that higher expression of either at baseline does not correlate with sensitivity in the patient samples tested.

Previous work^45,61,62  with T-ALL has identified a wide variation in response to venetoclax among cell lines and primary patient samples, wherein T-ALL cells displaying the immature phenotypes CD4⁻CD8⁻, such as LOUCY cell line, exhibit BCL2 dependence, and correspondingly submicromolar sensitivity in vitro to venetoclax. In contrast, cell lines with mature phenotypes demonstrate dependency for survival on the protein BCL-XL, and decreased sensitivity to venetoclax with >1 μM IC₅₀ values. CTCL itself has been characterized by TCR sequencing⁶³  as a disease derivative of mature memory T cells, from which one would predict dependency on BCL-XL. Direct measurements of mitochondrial membrane depolarization^64,65  have highlighted the stark differences in responses to BH3 mimetics between BCL2 and BCL-XL survival-dependent cells. A recent study⁶⁶  looked at the BH3 profiles of T-cell lymphoma cell lines, including 2 examined in our study, MyLa and Hut78. The 2 cell lines were identified as BCL-XL dependent, which was confirmed in our study, as we measured similar expression of BCL2L1 in all 3 cells as well as significantly decreased expression of BCL2 and MCL1, relative to normal controls. These findings, along with our observation of poor sensitivity of the 3 cell lines to the BCL2-selective inhibitor venetoclax, suggest that some CTCL cases are BCL-XL dependent. This interplay of BCL2 and BCL-XL was additionally significant in primary patient samples where the majority of samples had higher expression of BCL2 than BCL2L1, and a larger ratio predicted of sensitivity to venetoclax.

Synergy has been described between BCL2 inhibitors with a number of different therapies and in numerous malignancies.^{26,27,45,61,67-70}  Here, we show evidence of synergy between venetoclax and 2 HDAC inhibitors, vorinostat and romidepsin. Synergy was consistently observed at the higher dose ranges of either HDAC inhibitor indicating a dose-dependent mechanism. Not only did we detect antagonism at low doses, but there was some evidence of direct inhibition of caspase 3/7 activity with administration of either HDAC inhibitor alone. This is consistent with reports on both romidepsin³⁰  and vorinostat,⁷¹  which showed the effects of either drug differ dramatically at low and high dose, including 1 study where romidepsin was found to inhibit the production of reactive oxygen species at low doses.³² 

The lack of synergy observed in the CTCL cell lines suggests, perhaps unsurprisingly, that expression of and dependence on BCL2 is required for synergistic interactions with other therapies. Although in the cell lines the observed synergy can be linked directly to changes in BCL2 expression, the preferential effect observed with vorinostat and venetoclax is likely due to the greater degree of change in the expression of BCL2L11. Dissociation of BCL2-BIM complexes has been implicated as the primary mechanism of induction of apoptosis by venetoclax and other BCL2 inhibitors.^21,46,68  The low baseline expression of BCL2L11 in the HH and MyLa cell lines further supports the proposed theory that change in expression of BCL2L11 is critical to apoptosis by romidepsin, whereas baseline expression is nonessential to therapeutic effects.^72,73 

In summary, this study supports the clinical implementation of venetoclax as a novel therapy for leukemic CTCL, and further demonstrates a synergistic effect when venetoclax is combined with the HDAC inhibitors romidepsin and vorinostat. The potential for an effective combination oral therapy with venetoclax and vorinostat is also intriguing. Further clinical trials are planned to explore this potential synergy in vivo.

This work was supported by the Drs Martin & Dorothy Spatz Charitable Foundation (M.G.); the Yale Specialized Program of Research Excellence in Skin Cancer (National Institutes of Health, National Cancer Institute 1P50CA121974) (M.G.); the Leon Rosenberg, MD Medical Student Research Fund in Genetics (B.M.C.); and the Taylor Opportunity Student Research Fellowship (B.M.C.).

Contribution: B.M.C., J.G.W., J.M.L., and M.G. designed experiments; B.M.C. and M.G. analyzed the data, prepared figures, and wrote the manuscript; B.M.C., J.G.W., and F.N.M. performed experiments; M.G., F.M.F., and K.R.C. provided patient samples; B.M.C., J.M.L., and M.G. participated in discussions of the data; and all authors reviewed and edited the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Michael Girardi, Yale University School of Medicine, PO Box 208059, 333 Cedar St, New Haven, CT 06520-8059; e-mail: michael.girardi@yale.edu.