The serine hydroxymethyltransferase-2 (SHMT2) initiates lymphoma development through epigenetic tumor suppressor silencing

Abstract

Cancer cells adapt their metabolic activities to support growth and proliferation. However, increased activity of metabolic enzymes is not usually considered an initiating event in the malignant process. Here, we investigate the possible role of the enzyme serine hydroxymethyltransferase-2 (SHMT2) in lymphoma initiation. SHMT2 localizes to the most frequent region of copy number gains at chromosome 12q14.1 in lymphoma. Elevated expression of SHMT2 cooperates with BCL2 in lymphoma development; loss or inhibition of SHMT2 impairs lymphoma cell survival. SHMT2 catalyzes the conversion of serine to glycine and produces an activated one-carbon unit that can be used to support S-adenosyl methionine synthesis. SHMT2 induces changes in DNA and histone methylation patterns leading to promoter silencing of previously uncharacterized mutational genes, such as SASH1 and PTPRM. Together, our findings reveal that amplification of SHMT2 in cooperation with BCL2 is sufficient in the initiation of lymphomagenesis through epigenetic tumor suppressor silencing.

Follicular lymphoma (FL) and diffuse large B-cell lymphoma (DLBCL) are the most common forms of B-cell lymphoma that result from the expansion of germinal center B cells¹. Fifty percent of FLs eventually transform into an aggressive disease resembling DLBCL and this has been linked to MYC activation and loss of tumor suppressor genes (p53, B2M and CDKN2A/B)^2–4. The genetic hallmark of FL is the t(14;18) translocation that leads to increased BCL2 expression^5,6. However, this translocation is detectable in 30–50% of healthy adults and not sufficient for FL development⁷. Loss of epigenetic control of gene expression programs is considered a key initiating event in human lymphomagenesis. In particular, the histone modifiers KMT2D (MLL2), CREBBP and EP300 are among the most prevalent and early genetic lesions in human lymphomas; mutations in these genes permit pro-oncogenic gene expression programs in B cells^8–11. By contrast, the role for aberrant DNA methylation is less established in B-cell lymphoma. Despite the fact that systematic changes in DNA methylation overlap with somatic mutation patterns and are linked to outcomes, DNA methyltransferases are rarely direct targets of somatic mutations^12,13.

Lymphomas show a myriad of changes in cell metabolism that are thought to support malignant growth. These include activation of oxidative phosphorylation, serine catabolism and the folate cycle^14,15. Serine hydroxymethyltransferase-2 (SHMT2) is a key mitochondrial enzyme in serine catabolism that converts serine to glycine and a one-carbon unit that yields S-adenosyl methionine (SAM) and NADPH through the folate and methionine cycles^16,17. The enzyme is highly expressed in many fast-growing cancer cells^15,16 and increased serine catabolism supports malignant growth through diverse mechanisms, including enhanced nucleotide synthesis^17,18, redox balance¹⁵, mitochondrial translation^19,20, DNA and histone methylation^17,21, and suppression of retrotransposon activation²². In the present study, we report that SHMT2 is an oncogenic driver of BCL2-expressing lymphomas and acts, at least in part, through silencing of previously uncharacterized tumor suppressor genes.

Results

The SHTM2 gene is amplified in human B-cell lymphomas.

To identify potential drivers of lymphoma development, we examined chromosomal changes in large collections of FL (n = 176) and DLBCL (n = 568) samples. Briefly, we analyzed DNA copy number data based on array-based comparative genomic hybridization (rCGH) (FL) and single-nucleotide polymorphism (SNP) arrays (DLBCL) using the GISTIC 2.0 algorithm²³. We identified 11 and 16 statistically significant amplified regions in FL and DLBCL, respectively (Supplementary Tables 1 and 2). The most prevalent amplicon was centered on chromosome 12q14.1 (Fig. 1a,b). The minimal common region encompassed 308 genes, including known lymphoma oncogenes (CDK4, MDM2) and SHMT2 (gains in 27.84% of FLs and 26.7% of DLBCLs), whose expression has been previously linked to fast-growing cancer cells^14,18,24 (Supplementary Table 2 and Extended Data Fig. 1a). The cytoplasmic SHMT1 gene was neither amplified nor a target of somatic mutations. Other serine metabolism genes also showed gains, for example, PHGDH was amplified in 9% of FLs and 18.8% of DLBCLs, and PSPH had copy number gains (chromosome 7p11.2) in 21.6% of FLs and 24.3% of DLBCLs; gains for PSPH showed a trend (37.5%) toward coincidence with SHMT2 gains (Supplementary Tables 3 and 4, and Extended Data Fig. 1b,c). We confirmed these data in a second paired SNP array and gene expression dataset (n = 249 DLBCL samples). This further revealed that SHMT2 gains are more common in the GCB subtype (42%) compared to the ABC subtype (12%) or unclassified samples (14%) (Supplementary Table 5 and Extended Data Fig. 1d). Across subtypes, samples with copy number gains also showed significantly higher SHMT2 messenger RNA expression (two-tailed P < 0.0001; Fig. 1c). Hence, SHMT2 is a target of the chromosome 12q14.1 amplicon in a large fraction of human B-cell lymphomas.

Fig. 1 |. Genomic amplification of SHMT2 in human B-cell lymphomas.

a, Gain (red) across the genome in the analysis of SNP arrays of human DLBCL (n = 568 tumors). b, rCGH array analysis of human FL (n = 176 tumors). c, Gene expression of SHMT2 in the subtypes of 249 human DLBCL samples with wild-type (diploid) or amplified (gain) SHMT2. A two-tailed Student’s t-test was used to determine statistical significance. GC: P(diploid versus gain) = 0.000016, n = 52 diploid tumors, n = 39 tumors with SHMT2 gain. ABC: P(diploid versus gain) = 0.0001, n = 88 diploid tumors, n = 13 tumors with SHMT2 gain. Unclassified: P(diploid versus gain) = 0.1206, n = 35 diploid tumors, n = 6 tumors with SHMT2 gain. The numerical data for this figure are presented in Source Data File 1.

Increased SHMT2 expression triggers FL development in vivo.

Next, we tested the ability of wild-type SHMT2 to promote lymphoma development in cooperation with BCL2 in vivo. The VavP-Bcl2 mouse model (that is, transgenic mice where Bcl2 expression is controlled by Vav gene regulatory sequences) recapitulates the genetics, morphology and germinal center origin of human, BCL2-translocated FL^24,25. We transduced hematopoietic precursor cells (HPCs) collected from fetal livers of VavP-Bcl2 transgenic embryos (embryonic day 14.5) with MSCV-GFP (green flourescent protein) plasmids carrying either empty vector (n = 43) or the SHMT2 complementary DNA (cDNA) (n = 34), injected the HPCs into lethally irradiated syngeneic mice, and monitored for lymphoma development (Extended Data Fig. 2a). VavP-Bcl2 vector mice rarely develop FLs (15% at day 200). By contrast, enforced SHMT2 expression led to the rapid onset of lymphomas with much higher penetrance (75% by day 100) (log-rank test; P = 0.000009) (Fig. 2a,b). We tested the metabolic consequences of increased SHMT2 levels in lymphoma cells by isotope tracing with ¹³C,¹⁵N-serine on isolated B220⁺ B cells labeled for 3 h ex vivo. Liquid chromatography-mass spectrometry (LC-MS) readily showed increased glycine labeling with ¹³C and ¹⁵N resulting in increased fractions of C+2 (¹³C₂-glycine) and N+1 (¹⁵N-glycine) (Fig. 2c). We also detected a modest increase in SAM production (SAM/methionine ratio) in SHMT2-expressing FLs (Fig. 2d). SHMT2/GFP but not control vector/GFP-expressing cells were enriched during lymphomagenesis, confirming SHMT2 and not viral integration as the cause (Extended Data Fig. 2b,c). All lymphomas showed follicular expansion of B220⁺ B cells with low apoptosis by terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL), low Ki-67 and positivity peanut agglutinin (PNA) indicating germinal center origin (Fig. 2e and Extended Data Fig. 2d). The germinal center origin was further supported by VDJH4 sequencing revealing evidence of somatic hypermutation (Supplementary Table 6 and Extended Data Fig. 2e). Detailed analysis of viable spleen cells by flow cytometry showed the expansion of CD19⁺GL7⁺ lymphoma cells in both VavP-Bcl2; vector (that is, VavP-Bcl2 mice with empty MSCV vector; approximately 50%) and VavP-Bcl2; SHMT2 (that is, mice with viral vector encoding SHMT2 cDNA; approximately 70%) tumors with an equal fraction (approximately 20%) of immunoglobulin D (IgD)⁺ cells (Fig. 2f). Therefore, the wild-type SHMT2 enzyme acts as an oncogenic trigger in a Bcl2 transgenic mouse model of FL.

Fig. 2 |. SHMT2 acts as an oncogenic driver in a mouse model of FL.

a, Kaplan–Meier survival curve of female C57BL/6 mice bearing the VavP-Bcl2; vector (black; n = 43 mice) or VavP-Bcl2; SHMT2 (red; n = 34 mice) tumors. A log-rank test determined the statistical significance of survival between two different groups; VavP-Bcl2; SHMT2 versus VavP-Bcl2; vector, P(log-rank test) = 0.000009. b, Representative SHMT2 immunoblotting of sorted B cells from three VavP-Bcl2; vector and three VavP-Bcl2; SHMT2 tumors. The immunoblotting was performed independently three times with the same results. The uncropped images of the original blots are presented in Source Data File 2. c, Fraction of labeled and unlabeled glycine to total glycine measured by LC–MS in extracts of isolated B cells from VavP-Bcl2; vector (n = 3 mice) or VavP-Bcl2; SHMT2 (n = 2 mice) tumors labeled with ¹³C₃,¹⁵N-serine. C: glycine; C+1: ¹³C₁-glycine; C+2: ¹³C₂-glycine; N+1: ¹⁵N-glycine. d, Ratio of SAM to methionine measured by LC–MS in extracts of isolated B cells collected from VavP-Bcl2; vector (n = 4 mice) or VavP-Bcl2; SHMT2 (n = 5 mice) tumors. A two-tailed Student’s t-test was used to determine statistical significance. P(VavP-Bcl2; vector versus VavP-Bcl2; SHMT2) = 0.1332. e, Representative histological micrographs of spleens collected from VavP-Bcl2; vector and VavP-Bcl2; SHMT2 tumors. The sections were stained with H&E, anti-B220, anti-Ki-67, TUNEL and anti-PNA. This experiment was independently repeated on samples from three different mice in each genotype with similar results. Scale bar, 500 nm. The histological micrographs of replicates are presented in Source Data File 4. f, Representative images of flow cytometry analysis of cellular composition of VavP-Bcl2; vector versus VavP-Bcl2; SHMT2 splenic tumor cells. This experiment was independently repeated on samples from three different mice in each genotype with similar results. The flow cytometry analysis of replicates are presented in Source Data File 4. The numerical data for this figure are presented in Source Data File 3.

SHMT2 is overexpressed in human aggressive B-cell lymphomas.

We also examined the role of SHMT2 in aggressive B-cell lymphomas. Gene expression analysis of human transformed FLs (tFLs) compared to indolent FLs (n = 6 paired FL and tFL samples) showed the highest SHMT2 levels in the aggressive tumors (Fig. 3a). The same elevated SHMT2 was seen in aggressive murine MYC-induced lymphomas (n = 8 VavP-Bcl2; vector, n = 8 VavP-Bcl2; MYC) (Fig. 3a and Supplementary Table 7). Also in human DLBCL, SHMT2 RNA levels were significantly correlated with MYC (Pearson’s r = 0.3839; two-tailed P = 0.000001; Fig. 3b). In human P493-6 lymphoma cells with doxycycline (DOX)-regulated MYC expression²⁶, MYC activation induced the expression of SHMT2 along with other enzymes in serine-glycine biosynthetic pathway (Fig. 3c). Consistently, murine MYC/Bcl2-driven lymphomas showed high expression of all enzymes of the serine-glycine biosynthetic pathway compared to control lymphomas or normal spleens (Fig. 3d). To test the contribution of SHMT2, we knocked the gene down and observed a loss of proliferation and viability across a panel of human DLBCL cell lines (Fig. 3e and Extended Data Fig. 2f). Metabolic labeling (6 h) with ¹³C,¹⁵N-serine in SU-DHL-4 cells readily confirmed reduced SHMT2 activity and glycine labeling, although we could not detect ¹⁵N labeling in these samples, with a corresponding decrease in the SAM/methionine ratio (unpaired t-test; P = 0.03; Fig. 3f,g). Consistent with reports from Maddocks et al.¹⁷, we also detected effects on nucleotide (ATP) levels, such that SHMT2-expressing P493-6 cells showed a substantial increase and SHMT2-deficient SU-DHL8 cells had decreased ATP levels (Fig. 3h).

Fig. 3 |. MYC regulates the expression of SHMT2 in both human and mice tFL.

a, Scatter plot of genes differentially expressed in mouse and human tFLs compared to mouse and human FL. The red dots label genes differentially expressed in both mouse and human samples. The red box outlines the genes overexpressed in both mouse and human tFLs. n = 6 human FL; n = 6 human tFL; n = 8 VavP-Bcl2; vector; n = 8 VavP-Bcl2; MYC. b, Scatter graph representing the correlation of SHMT2 expression with MYC expression in 249 human DLBCLs. Pearson’s r = 0.3839, two-tailed P(SHMT2 versus MYC) = 0.000001. c, Relative mRNA expression of serine-glycine biosynthetic pathway genes in the P493-6 cell line. Cells (n = 2 independent experimental replicates) were treated for 72 h with DOX (MYC^off); subsequently, the DOX was removed (MYC^on) and mRNA expression was measured at several different time points after DOX treatment. d, Heatmap showing the expression of serine-glycine biosynthetic pathway genes in spleen (n = 4 mice), VavP-Bcl2; vector (n = 8 mice) and VavP-Bcl2; MYC (n = 8 mice) B cells. e, Bar graphs representing the viability of DLBCL cell lines following SHMT2 knockdown by shRNA (shSHMT2_1 or shSHMT2_2). A two-tailed Student’s t-test was used to determine statistical significance. SU-DHL-4: P(vector versus shSHMT2_1) = 0.0246; P(vector versus shSHMT2_2) = 0.0343; SU-DHL-8: P(vector versus shSHMT2_1) = 0.0034; P(vector versus shSHMT2_2) = 0.00007; Toledo: P(vector versus shSHMT2_1) = 0.0216; P(vector versus shSHMT2_2) = 0.000039. *P < 0.05, **P < 0.01, ***P < 0.001. n = 8 cell culture replicates for all experimental groups presented in this panel. Eight replicates are the combination of three independent experiments. f, Fraction of labeled and unlabeled glycine to total glycine measured by LC-MS in extracts of ¹³C₃,¹⁵N-serine labeled (0.8 mM) SU-DHL-4 cells carrying control vector or shSHMT2. C: glycine; C+1: glycine ¹³C₁; C+2: glycine ¹³C₂. Vector: n = 2 experimental replicates; shSHMT2_1: n = 2 experimental replicates; shSHMT2_2: n = 2 experimental replicates. g, SAM/methionine measured by LC-MS in extracts of SU-DHL-4 carrying vector or short hairpin against SHMT2 (shSHMT2_1 or shSHMT2_2). A two-tailed Student’s t-test was used to determine statistical significance; n = 2 independent experimental replicates. h, Bar graphs showing ATP content in human B-cell lymphoma cell lines after overexpression and knockdown of SHMT2. n = 2 vector and n = 5 SHMT2 technical cell culture replicates of the P493-6 cell line; n = 3 technical cell culture replicates of SU-DHL-8. This experiment was repeated independently three times with similar results. i, Survival curve of C57BL/6 mice carrying VavP-Bcl2; MYC (n = 27 mice) and VavP-Bcl2; MYC; shShmf2 (n = 28 mice) tumors. A log-rank test was used to determine the statistical significance of survival between two different groups; P(VavP-Bcl2; MYC versus VavP-Bcl2; MYC; shShmt2) = 0.0033. j, Immunoblot against SHMT2 and MYC in B cells collected from VavP-Bcl2; MYC and VavP-Bcl2; MYC; shShmt2 tumors. The immunoblot was performed twice on three samples from three different mice with the same genotype with similar results. The uncropped images of the original blots are presented in Source Data File 5. The numerical data for this figure are presented in Source Data File 6.

Next, we probed the requirement for SHMT2 in aggressive lymphomas in vivo. Shmt2 knockdown reduced lymphomagenesis driven by MYC in the context of VavP-Bcl2 transgenic hematopoietic stem cells (HSCs) leading to significantly increased survival (coexpression of MYC and shShmt2: n = 28; MYC and control vector: n = 27; log-rank test; P = 0.0033) (Fig. 3i,j). This suggests that tested experimental compounds that target SHMT2 activity or expression could have therapeutic activity against lymphoma. For example, SHIN1 is a small molecule inhibitor of SHMT1 and SHMT2 (ref. ²⁷); across four DLBCL lines, we observed apoptotic cell kill at low micromolar concentrations (half maximal inhibitory concentration (IC₅₀): 5 μM) as evidenced by decreased viability (Fig.4a) and increased annexin V staining (Fig. 4b). Expression of SHMT2 partially protected DoHH2 cells from SHIN1 treatment, whereas a catalytically dead mutant of SHMT2 (SHMT2 K280A) did not protect cells (Fig. 4c). Alternatively, the histone H3-K9 methyltransferase 3 (G9a) is required for the expression of serine-glycine biosynthetic pathway genes; G9a inhibitory compound BIX-01294 reduces their expression²⁸ (Fig. 4d). Across DLBCL cell lines, BIX-01294 showed only modest cell kill (IC₅₀ approximately 1.5 μM with 3 d of exposure; Fig. 4e). However, we noticed synergistic cell killing when we combined the G9a inhibitor (BIX-01294: 2.5 μM) with a clinical BCL2 inhibitor (venetoclax (ABT-199): 250 nM) as indicated by annexin V flow cytometry and increased caspase-3 cleavage (Fig. 4f,g and Supplementary Table 8). Thus, SHMT2 contributes to the biology of aggressive human and murine lymphomas; targeting its expression or activity shows antilymphoma activity that is enhanced by concurrent BCL2 inhibition.

Fig. 4 |. Targeting SHMT2 activity or expression for lymphoma therapy.

a, Bar graphs presenting the viability of DLBCL cell lines in response to SHIN1 on day 4 compared to day 0 at different concentrations of the drug (n = 3 technical cell culture replicates). This experiment was repeated independently three times with similar results. b, Representative bivariate density plot (upper panel) and scatter plot (lower panel) of cell death analysis (annexin V and 7-ADD) of the SU-DHL-4 cell line treated with SHIN1 (10 μM) or dimethyl sulfoxide (DMSO) for 48 h. This experiment was repeated independently three times with similar results. The flow cytometry analysis of the replicates is presented in Source Data File 9. c, Bar graphs showing the percentile of viable cells in SHIN1-treated versus DMSO-treated DoHH2 cells carrying vector, catalytic dead SHMT2 or SHMT2 construct; n = 3 technical cell culture replicates for all the treatment groups presented. This experiment was repeated independently twice with similar results. d, Relative mRNA expression of serine biosynthesis pathway enzymes in Karpas-422 cells after treatment with 5 μM BIX-01294 for 24 h. n = 3 technical cell culture replicates for all the treatment groups presented. This experiment was repeated independently with three different cell lines with similar results. e, Bar graphs demonstrating the viability of DLBCL cells after 3 d of treatment with different concentrations of BIX-01294. n = 4 technical cell culture replicates for all the treatment groups presented. This experiment was repeated independently three times with similar results. f, Representative scatter plot of cell death analysis (annexin V and 7-ADD) of the SU-DHL-4 cell line treated with BIX-01294 (2.5 μM) or ABT-199 (250 nM) or both for 72 h. This experiment was repeated independently three times with similar results. The flow cytometry analysis of the replicates is presented in Source Data File 9. g, Cell death analysis by immunoblotting against total and cleaved caspase-3 in DLBCL cell lines after treatment with BIX-01294 (2.5 μM), ABT-199 (250 nM) or their combination. This experiment was repeated independently twice with similar results. The uncropped images of the original blots are presented in Source Data File 7. The numerical data for this figure are presented in Source Data File 8.

SHMT2 changes the epigenetic control of gene expression.

Impaired epigenetic regulation is an early and initiating event in B-cell lymphoma; key genes (for example, KMT2D, CREBBP, EP300) are among the earliest and most frequent mutational targets^8–11. Increased serine catabolism has been linked to epigenetic silencing of retrotransposons²² and we reasoned that the availability of activated methyl groups may affect epigenetic mechanisms of tumor suppression and trigger lymphomagenesis. To explore this hypothesis, we first performed gene expression (RNA sequencing (RNA-seq)) analyses on sorted B220⁺ cells from VavP-Bcl2; vector (control) and VavP-Bcl2; SHMT2 lymphomas. We identified significantly (P < 0.05) decreased expression of 507 genes and increased expression of 1,065 genes upregulated in SHMT2 versus controls (Supplementary Table 9).

Next, we sought to associate these expression changes with epigenetic markers. A global survey of histone marks by mass spectrometry showed no overall changes in histone methylation at H3K4, H3K27 and H3K9 in SHMT2-expressing compared to control murine FL B220⁺ cells (n = 3 each; Fig. 5a). SAM availability has been specifically linked to dimethylation and trimethylation of H3K4 (H3K4me2, H3K4me3) (ref. ²⁹). Therefore, we performed chromatin immunoprecipitation sequencing (ChIP–seq) for H3K4me2 (VavP-Bcl2; vector: 35,302 peaks; VavP-Bcl2; SHMT2: 47,300 peaks; both n = 2) and for H3K4me3 (VavP-Bcl2; vector: 34,651 peaks; VavP-Bcl2; SHMT2: 39,163 peaks; both n = 2). H3K4me2 peaks increased in promoters (9.1% in VavP-Bcl2; vector versus 20.8% in VavP-Bcl2; SHMT2) and decreased in intergenic (9.1% in VavP-Bcl2; vector versus 4.2% in VavP-Bcl2; SHMT2) and distal (31.8% in VavP-Bcl2; vector versus 25% in VavP-Bcl2; SHMT2) region B cells (Fig. 5b and Supplementary Table 10). We observed a similar pattern for H3K4me3 peaks (promoter: VavP-Bcl2; vector 19.6% versus VavP-Bcl2; SHMT 245.3%; intergenic: VavP-Bcl2; vector 10.5% versus VavP-Bcl2; SHMT2 4.3%; distal: VavP-Bcl2; vector 34.6% versus VavP-Bcl2; SHMT2 15.4%; Fig. 5b). Differential binding analysis (Supplementary Table 11) identified only 24 gene loci with increased H3K4 dimethylation (8 in promoters, 16 in enhancers), 44 genes with loss of H3K4 dimethylation (4 in promoters, 40 in enhancers) in VavP-Bcl2; SHMT2 lymphoma versus controls. Similarly, differential binding showed 98 genes with increased H3K4me3 (53 promoters and 45 enhancers) and 148 gene loci with decreased H3K4me3 (33 promoters and 115 enhancers) in VavP-Bcl2; SHMT2 B cells versus controls (Supplementary Table 10). However, comparison of H3K4 methylation data with gene expression data from the same samples identified only 4 and 8 highly expressed genes in VavP-Bcl2; SHMT2 tumors with correspondingly increased H3K4me2 (enhancer) or H3K4me3 (3 promoters and 5 enhancers), respectively (Supplementary Table 11). These genes were enriched for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways including mismatch repair (P = 0.0103), fatty acid elongation (P = 0.012) and biosynthesis of unsaturated fatty acids (P = 0.012) (Supplementary Table 12). Hence, we detected no global change in histone marks; specific analysis of H3K4 methylation identified few changes linked to gene expression changes and lymphoma development.

Fig. 5 |. Epigenetic studies of VavP-Bcl2; SHMT2 B cells.

a, Bar graphs representing the relative abundance of mono-, di- and trimethylation of H3K4 in VavP-Bcl2; vector (n = 5 mice) and VavP-Bcl2; SHMT2 (n = 3 mice) B cells measured by LC-MS. A two-tailed Student’s t-test was used to determine statistical significance. Monomethylation: P(VavP-Bcl2; vector versus VavP-Bcl2; SHMT2) = 0.0175; dimethylation: P(VavP-Bcl2; vector versus VavP-Bcl2; SHMT2) = 0.5692; trimethylation: P(VavP-Bcl2; vector versus VavP-Bcl2; SHMT2) = 0.2318. b, Genomic distribution of H3K4me2 and H3K4me3 in VavP-Bcl2; SHMT2 (dimethylation: n = 2 mice; trimethylation: n = 3 mice) B cells compared to control (VavP-Bcl2; vector; dimethylation: n = 2 mice, trimethylation: n = 2 mice). c, Dot blot of DNA (100 nM, 50 nM and 25 nM) from SU-DHL-4 cells carrying vector or short hairpin against SHMT2 (shSHMT2). Methylene blue staining was used as the DNA loading control. This experiment was repeated independently three times with similar results. d, Dot blot of DNA (100 nM, 50 nM and 25 nM) from VavP-Bcl2; vector (n = 2) or VavP-Bcl2; SHMT2 HSCs (n = 2). Methylene blue staining was used as the DNA loading control. This experiment was repeated independently twice with three independent samples in each genotype with similar results. e, Dot blot of DNA (100 nM, 50 nM and 25 nM) from VavP-Bcl2; vector and VavP-Bcl2; SHMT2 lymphoma B cells. Methylene blue staining was used as the DNA loading control. This experiment was repeated independently twice with three different mice in each genotype with similar results. f, Pie charts showing the genomic distribution of hypo- and hyper-DMCs (upper diagrams) within CpG islands, shores and oceans (lower diagrams) in VavP-Bcl2; SHMT2 (n = 2 mice) versus VavP-Bcl2; vector (n = 2 mice) tumors. g, Scatter plot of genes downregulated in B cells from VavP-Bcl2; SHMT2 tumors versus control (VavP-Bcl2; vector) tumors with hyperDMCs on their promoters. The data from f were compared to the expression data (RNA-seq: n = 2 VavP-Bcl2; vector tumors versus n = 3 for VavP-Bcl2; SHMT2 tumors). The dotted line further indicates genes with >20 hyperDMCs. The uncropped images of the original dot blots in c,d,e are presented in Source Data File 11. The numerical data for this figure are presented in Source Data File 10.

We then turned to study SHMT2-driven DNA methylation. Interestingly, global DNA methylation was decreased in DLBCL cell lines with SHMT2 knockdown (Fig. 5c) and increased in VavP-Bcl2; SHMT2-expressing HSCs 3 d after infection (Fig. 5d). Furthermore, global analysis of DNA methylation in VavP-Bcl2; SHMT2 and VavP-Bcl2; vector lymphoma B cells revealed DNA hypomethylation consistent with a previous study¹³ (Fig. 5e). We also detected increased CpG island methylation in specific promoters in VavP-Bcl2; SHMT2 tumors. In detail, we performed enhanced reduced representation bisulfate sequencing (ERRBS) on VavP-Bcl2; SHMT2 (n = 2) and VavP-Bcl2; vector lymphoma B220⁺ cells (n = 2). We detected 2,878 differentially hypomethylated CpG sites (hypoDMCs) and 3,197 differentially hypermethylated CpG sites (hyperDMCs) in SHMT2 versus controls (Supplementary Table 13). In VavP-Bcl2; SHMT2 tumors, hyperDMCs were enriched in the promoters (33.9%), exons (21.79%) and introns (16.7%) of the genome within CpG islands (66.95%) (Fig. 5f). Promoter hyperDMCs corresponded to 624 genes; 80 (12.8%) of these had in excess of 10 hyperDMCs (Supplementary Table 13). By contrast, most hypoDMCs were found in downstream (5.6%), distal (33.2%) and intergenic (18.3%) regions of the genome or within CpG oceans (96.9%) (Fig. 5g). The promoters with hypoDMCs in VavP-Bcl2; SHMT2 tumors correlated with 250 genes; only three genes had in excess of 10 hypoDMCs and showed no significant change in their gene expression (Supplementary Table 13).

SHMT2 augments promoter silencing of tumor suppressor genes in lymphoma.

Next, we cross-referenced DNA methylation patterns and gene expression data in VavP-Bcl2; SHMT2 lymphomas. We identified 83 genes that showed simultaneously the most extensive reduction in their mRNA expression (we selected significantly affected genes (q < 0.05) with a >1.5-fold decrease in expression in VavP-Bcl2; SHMT2 tumors) and had differential promoter hypermethylation (Supplementary Tables 9 and 13). Among these, 30 genes had at least 3 hyperDMCs in VavP-Bcl2; SHMT2 B cells (Fig. 5h). Notably, this approach immediately pinpointed genes that were also targets of somatic mutations, for example, the tyrosine phosphatase PTPRM, which is a mutational target in approximately 9% of 58 DLBCL samples (Broad 2012), 4.2% of 48 DLBCL samples (The Cancer Genome Atlas (TCGA)) and silenced in 64% of acute lymphoblastic leukemia (ALL) samples³⁰ (Fig. 6a). Similarly, SASH1 is mutated in 4% of TCGA DLBCL samples (Fig. 6a) and a target of chromosome 6q24.1 deletion in approximately 31% of DLBCLs and 22% of FL (Fig. 6b,c). TUSC1 is a target of chromosome 9p deletions in lung cancer and melanoma^31,32. The pattern of mutations spread across the coding sequences in PTPRM and deletions of SASH1 is most consistent with loss-of-function lesions and marks these genes as candidate tumor suppressors. We confirmed epigenetic regulation of SASH1 and PTPRM by 5-azacytidine (5-Aza; 24 h) treatment in three DLBCL cell lines (SU-DHL-4, SU-DHL-6 and DoHH2) (Fig. 6d); SASH1 expression was inversely correlated with SHMT2 expression across 249 DLBCL samples (P = 0.000001) (Fig. 6e). We also confirmed repression of PTPRM and SASH1 in VavP-Bcl2; SHMT2-expressing lymphoma B cells (Fig. 6f). Therefore, differential DNA methylation pointed to previously uncharacterized candidate tumor suppressor genes.

Fig. 6 |. Epigenetic silencing of tumor suppressor genes contributes to the oncogenic action of SHMT2.

a, Mutation map of PTPRM and SASH1 (cBioPortal). b, Bar graph presenting the loss of SASH1 located on chromosome 6 by DNA copy number analysis in fractions of DLBCL (n = 568 patient samples) and FL (n = 176 patient samples) tumors. c, Analysis of SNP arrays of DLBCL (n = 568 patient samples) and rCGH array analysis of FL (n = 176 patient samples) tumors showing deletions between chromosomes 4 and 8. SASH1 is located on 6q24.1. d, Relative mRNA expression of selected target genes after 5-Aza treatment in DLBCL cell lines for 24 h. SASH1: SU-DHL-6: n = 2 untreated samples, n = 2 5-Aza-treated samples; SU-DHL-4: n = 3 untreated samples, n = 3 5-Aza-treated samples. PTPRM: SU-DHL-6: n = 2 untreated samples, n = 3 5-Aza-treated samples. DoHH2: n = 2 untreated samples, n = 2 5-Aza-treated samples. The number of samples presented are technical cell culture replicates. This experiment was repeated independently in at least three different DLBCL cell lines with similar results. e, Scatter graph representing the correlation of SHMT2 expression with SASH1 expression in human B-cell lymphoma (n = 249 human samples). Pearson’s r = −0.4295, two-tailed P(SHMT2 versus SASH1) = 0.000001. f, Relative mRNA expression of Ptprm and Sash1 in VavP-Bcl2; vector versus VavP-Bcl2; SHMT2 mouse B cells. A two-tailed Student’s t-test was used to determine statistical significance. Ptprm: P(VavP-Bcl2; vector versus VavP-Bcl2; SHMT2) = 0.0208, n = 2 VavP-Bcl2; vector mice, n = 3 VavP-Bcl2; SHMT2 mice. Sash1: n = 2 VavP-Bcl2; vector mice, n = 2 VavP-Bcl2; SHMT2 mice. The numerical data for this figure are presented in Source Data File 12.

Next, we tested the potential role of PTPRM and SASH1 in lymphoma initiation. Knockdown of Sash1 or Ptprm in HSCs from VavP-Bcl2 transgenic mice followed by adoptive transfer into syngeneic mice caused an increase in lymphoma development in vivo, thus providing evidence of in vivo tumor-suppressive activity (n = 17 VavP-Bcl2; vector mice; n = 38 VavP-Bcl2; shSash1 mice; n = 19 VavP-Bcl2; shPtprm mice; VavP-Bcl2; shSash1 versus VavP-Bcl2; vector: P(log-rank) = 0.0167; VavP-Bcl2; shPtprm versus VavP-Bcl2; vector: P(log-rank) = 0.07) (Fig. 7a). Lymphomas driven by Sash1 or Ptprm resembled VavP-Bcl2; SHMT2-driven tumors and showed the typical follicular structures, low Ki-67 levels, somewhat higher fraction of IgD⁺ cells and otherwise a marker profile resembling human and murine FLs (Fig. 7b,c). Furthermore, knockdown of Shmt2 in aggressive VavP-Bcl2/MYC tumors caused upregulation of Sash1 and Ptprm (Fig. 7d). Similarly, pharmacological inhibition of SHMT2 with the SHIN1 compound²⁷ led to increased SASH1 and PTPRM expression (Fig. 7e). Conversely, individual knockdown of SASH1 or PTPRM was partially protective from SHIN1-mediated cell killing in DLBCL cells (SU-DHL-4), although the incomplete protection implies that additional mechanisms are at play (Fig. 7f).

Fig. 7 |. Characterization of candidate tumor suppressor genes in vivo.

a, Kaplan–Meier survival curve of female C57BL/6 mice bearing VavP-Bcl2; vector (black; n = 17 mice) or VavP-Bcl2; shSash1 (purple; n = 38 mice) or VavP-Bcl2; shPtprm (blue; n = 19 mice) tumors. A log-rank test determined the statistical significance of survival between three different groups; P(VavP-Bcl2; shSash1 versus VavP-Bcl2; vector) = 0.0167; P(VavP-Bcl2; shPtprm versus VavP-Bcl2; vector) = 0.0703. b, Representative images of flow cytometry analysis of cellular composition of VavP-Bcl2; shSash1 mouse splenic tumor cells. This experiment was repeated independently three times with similar results. c, Representative histological micrographs of spleens collected from VavP-Bcl2; vector and VavP-Bcl2; shSash1 tumors. The sections were stained with H&E and antibodies for B220 to stain B cells, and TUNEL and Ki-67 to analyze apoptosis and proliferation. This experiment was repeated independently three times with similar results. Scale bar, 500 nm. d, Relative mRNA expression of Ptprm and Sash1 in VavP-Bcl2; MYC (n = 2 mice) and VavP-Bcl2; MYC; shShmt2 (n = 3 mice) mouse B cells. A two-tailed Student’s t-test was used to determine statistical significance. Ptprm: P(VavP-Bcl2; MYC versus VavP-Bcl2; MYC; shShmt2) = 0.04; Sash1: P(VavP-Bcl2; MYC versus VavP-Bcl2; MYC; shShmt2) = 0.008. e, Bar graphs presenting the relative mRNA expression of PTPRM and SASH1 measured by qRT-PCR in human DLBCL cell lines after treatment with 10 μM of SHIN1 for 48 h. PTPRM: SU-DHL-6: n = 2 DMSO, n = 2 SHIN1; SU-DHL-4: n = 2 DMSO, n = 3 SHIN1. SASH1: SU-DHL-4: n = 3 DMSO, n = 3 SHIN1. The number of samples presented are technical replicates. This experiment was repeated independently three times with similar results. f, Bar graph demonstrating the viability of SU-DHL-4 cells with and without short hairpin against PTPRM (shPTPRM) or SASH1 (shSASH1) after treatment with SHIN1 for 4 d. Vector: n = 3 DMSO; n = 3 SHIN1. shSASH1: n = 2 DMSO; n = 2 SHIN1. shPTPRM-1: n = 2 DMSO; n = 3 SHIN1. shPTPRM-2: n = 2 DMSO; n = 3 SHIN1. The number of samples presented are technical replicates. This experiment was repeated independently three times with similar results. g, Schematic diagram of the mechanism of SHMT2 oncogenesis in lymphoma. The flow cytometry analysis and histological micrographs of the replicates is presented in Source Data File 13. The numerical data for this figure are presented in Source Data File 14.

Discussion

Cancer cells, including B-cell lymphomas, show many adaptations in their metabolism that are thought to support malignant growth. These changes are not usually considered primary drivers of transformation and malignancy³³. We have shown that the wild-type (non-mutant) enzyme SHMT2 acts as an oncogenic driver in a murine model of BCL2-triggered FL. This is in contrast to the mutant isocitrate dehydrogenase enzymes that act as drivers of T-cell lymphoma and myeloid leukemia through the acquisition of a new metabolic function³⁴. The oncogenic function of SHMT2 is relevant to human disease because the gene encoding SHMT2 localizes to the most frequent amplicon in both FL and DLBCL at chromosome 12q14.1. SHMT2 copy number gains are also seen in sarcomas^35,36, and in pancreatic³⁷ and prostate adenocarcinoma³⁸. A first suggestion of potential oncogenic activity came from a genetic screen indicating that SHMT2 could partially rescue the proliferation defect of MYC-deficient cells³⁹. We also found that MYC-driven lymphomas depend on SHMT2. Indeed, SHMT2 has canonical promoter E boxes and is a direct transcriptional target of the MYC oncogene⁴⁰. Accordingly, SHMT2 is highly expressed in fast-growing cell lines and aggressive lymphomas where it has been linked to poor clinical outcomes^15,37,41. The requirement for SHMT2 is reminiscent of the requirement for phosphoglycerate dehydrogenase in breast cancer cells although it is not established whether phosphoglycerate dehydrogenase can act as an oncogenic driver^42,43. Together, our data indicate that the wild-type SHMT2 enzyme is a targetable driver of lymphomagenesis.

Our study further reveals a link between tumor metabolism and the epigenetic control of tumor suppressor genes. SHMT2 converts serine to glycine and simultaneously generates a methyl group that produces NADPH and SAM^14,16. In principle, serine catabolism can contribute to a myriad aspects of malignant cell growth including nucleotide synthesis and epigenetic regulation^17,20–22. Mutational evidence in lymphoma has highlighted epigenetic mechanisms (for example, mutations in the histone-modifying enzymes KMT2D, CREBBP and EP300) as key oncogenic mechanisms^8,9. We reasoned that the ability of SHMT2 to trigger lymphoma development may similarly be linked to B-cell epigenetics and gene expression. Indeed, SAM availability has been implicated in histone H3K4 methylation²⁹. However, we observed surprisingly few changes in H3K4 methylation that corresponded to altered gene expression or fingered genes and pathways linked to B-cell malignancy. DNA methylation and promoter CpG silencing are well-established oncogenic mechanisms in solid tumors and leukemia^44,45. By contrast, the role of DNA methylation in lymphomas is far less clear¹³. In VavP-Bcl2; SHMT2-driven tumors, we noticed hypermethylation of specific promoter elements and corresponding silencing of genes that were also mutational targets in human lymphomas. These include the signal adapters SASH1, PTPRM and TUSC1. The effect of SASH1 and PTPRM mutations and deletions has not been studied; we found that both act as tumor suppressors in vivo and contribute to the antiproliferative effects of SHMT2 inactivation. This indicates that SASH1 and PTPRM are among the effectors of SHMT2 in lymphoma initiation and maintenance, although other mechanisms clearly contribute. Together, our findings reveal that a wild-type metabolic enzyme can initiate lymphoma development in a manner that involves a surprising epigenetic mechanism and silencing of the previously unrecognized tumor suppressor genes SASH1 and PTPRM in lymphoma (Fig. 7g).

Methods

DNA copy number data acquisition and processing.

Raw data from U133 Plus 2.0 gene expression (Asymetrix), SNP (250k Sty, 250k Nsp and SNP6.0 arrays; Affymetrix) and rCGH (244k; Agilent) microarrays were downloaded from the Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/) or shared by the authors of the originating publication (Supplementary Table 1). Gene expression microarray data were RMA-normalized and batch-corrected using ComBat v.3.0 (GenePattern). All SNP and rCGH data were uniformly processed using an approach described previously⁴⁶ to generate log₂ (copy number) changes with reference to HapMap controls, and then segmented using the circular binary segmentation tool in GenePattern⁴⁷. The circular binary segmentation files were analyzed using the GISTIC 2.0 algorithm²³.

Generation of mice.

All animal experiments were carried out according to the guidelines of the Memorial Sloan Kettering Cancer Center institutional animal care and use committee (protocol np. 07-01-002). The VavP-Bcl2 mouse model of FL²⁵ was adapted to the adoptive transfer approach using retroviral transduced HPCs. HPC isolation and transduction were performed as in Wendel at al.⁴⁸. SHMT2 was cloned directly from human cells. We used CCDS 8934.1 (https://www.ncbi.nlm.nih.gov/CCDS/CcdsBrowse.cgi?REQUEST=CCDS&GO=MainBrowse&DATA=CCDS8934.1) as a template and the following primers: SHMT2-XhoI-forward: TAGATCTCTCGAGGTTATGCTGTACTTCTCT, SHMT2-EcoRI-reverse: GGAATTCGTTAACTCAATGCTCATCAAAAC to clone SHMT2 into the MIG plasmid. To prepare the MLS-GFP-MYC plasmid, first GFP-IRES-MYC (2,500 base pairs (bp)) was amplified from the MYC-MIG plasmid and ligated into MLS (with no GFP; cut by AVRII and PacI). The short hairpin for Shmt2 was cut by AVRII and cloned into MLS-GFP-MLS. The short hairpins against mouse Sash1 (TTTTCTTCTTCCAATCTCCTTT) and mouse Ptprm (CCCGGTCAGTCACAGATCCAAA) were cloned into MLS plasmids. All mice bearing tumors in this study were C57BL/6 females (The Jackson Laboratory).

Mouse B220⁺ tumor sample preparation.

B220⁺ cells were purified from mouse lymphoma tumors by immunomagnetic enrichment with CD45R (B220) microbeads (Miltenyi Biotec).

Histology.

Mouse tissues were fixed overnight in 4% paraformaldehyde, embedded in paraffin blocks and sectioned. Tissue sections were stained with hematoxylin and eosin (H&E) or with Ki-67, TUNEL, B220 or PNA following standard procedures^49,50.

Flow cytometry analysis.

Lymphocytes were isolated from tumor cell suspensions of representative tumors for each genotype by Fico/Lite-LM (mouse) (catalog no. 140610; Atlanta Biologicals). The remaining red blood cells were removed from cell suspension by ACK lysing buffer (catalog no. A1049201; Thermo Fisher Scientific). Cells were blocked by purified rat antimouse CD16/CD32 (clone 2.4G2; catalog no. 553142; BD Biosciences) and then stained with the following antibodies (all BD Biosciences at 1:100 dilution): APC-B220 (CD45R; catalog no. 553092) or APC-IgG1 (catalog no. 560089); PE-B220 (CD45R; catalog no. 553090); PE-CD19 (catalog no. 557399); PE-IgM (catalog no 553409); PE-IgD (catalog no. 558597); and PE-GL7 (catalog no. 561530). Analysis was performed with a BD LSR Fortessa cell analyzer (BD Biosciences) and FlowJo software 10.4.1 (Tree Star).

Somatic hypermutation.

Somatic hypermutation analysis of B cells isolated from VavP-Bcl2; vector and VavP-Bcl2; SHMT2 tumors was performed as described by Ortega-Molina et al.⁸.

Tumor clonality.

PCR to evaluate immunoglobulin variable heavy chain rearrangements was performed on the complementary DNA of VavP-Bcl2 lymphoma cells with a set of forward primers that anneal to the framework region of the most abundantly used immunoglobulin light chain variable region gene families and a reverse primer located in the Jλ1,3 gene segment (IgL-Vλ1: GCCATTTCCCCAGGCTGTTGTGACTCAGG; IgL-Jλ1,3: ACTCACCTAGGACAGTCAGCTTGGTTCC).

Human cell culture and plasmids.

The human lymphoma cell lines (SU-DHL-4, SU-DHL-8, SU-DHL-6, DoHH2, Toledo, Karpas-422, P493-6) were maintained in Roswell Park Memorial Institute (RPMI) medium containing 10% FCS, 1% l-glutamine and 1% penicillin-streptomycin. Cell lines were authenticated by short tandem repeat DNA profiling by Bio-Synthesis (http://www.biosyn.com/cell-line-authentication.aspx). Mycoplasma contamination was routinely tested with the ATCC Universal Mycoplasma Detection Kit (http://www.atcc.org/products/all/30-1012K.aspx).

The P493-6 cell line was treated for 72 h with DOX (MYC^off, 0.1 μg ml⁻¹); subsequently, DOX was removed (MYC^on) and mRNA expression was measured 1, 6, 16, 24 and 48 h after DOX treatment⁵¹.

Empty lentiviral vector (PLKO.1) as control, short hairpin RNA (shRNA) against human SHMT2 (no. 1: TRCN0000234656; no. 2: TRCN0000234657) or SASH1 (no. 1: TRCN0000164684; no. 2: TRCN0000166643) or PTPRM (no. 1: TRCN0000002882; no. 2: TRCN0000002881) or TUSC1 (no. 1: TRCN0000078093; no. 2: TRCN0000078096) were purchased from Sigma-Aldrich. The catalytically dead mutant of SHMT2 (SHMT2 K280A), SHMT2res (resistant to shRNA) and empty vector were kindly shared by T. Oellerich (Goethe University).

Viability assay.

Lymphoma cells were seeded at 2.5 × 10⁵ cells per 1 ml concentration in one well of a 12-well plate in triplicate. The viability assay was performed mixing Guava ViaCount reagent (Millipore 4000-0040) with cells (3:1 dilution). The assay was read using a Guava Flow Cytometer.

Total RNA and DNA extraction.

RNA extraction was performed using TRIzol (Ambion) according to the manufacturer’s protocol. DNA was extracted by Wizard Genomic DNA Purification Kit (Promega Corporation).

mRNA-seq library preparation and sequencing analysis.

RNA size, concentration and integrity were verified using a 2100 Bioanalyzer Instrument (Agilent Technologies). RNA-seq libraries were prepared using the Illumina TruSeq Stranded mRNA sample kits according to the manufacturer’s protocol. Libraries were validated using the 2100 Bioanalyzer Instrument and Quant-iT dsDNA HS Assay Kit (Thermo Fisher Scientific). The Casava v.1.8 software (Illumina) was used for base calling. RNA-seq reads were aligned to mm10 using STAR 2.5.3a with default parameters^52,53. Transcripts per million were calculated using Reference Sequence (RefSeq) gene annotation. The significance of differentially expressed genes was determined using the Wald significance test to compare different groups. For multiple hypothesis testing, the significance cutoff was 0.05. A log₂(fold change) greater than 1.5 or less than −1.5 was used for significance.

Human RNA-seq.

After RiboGreen quantification and quality control by Agilent BioAnalyzer, 160 ng of total RNA underwent poly(A) selection and TruSeq library preparation according to the instructions provided by Illumina (TruSeq RNA Library Prep Kit v2; catalog no. RS-122-2001), with 8 cycles of PCR. Samples were barcoded and run on a HiSeq 2500 System in rapid run mode using the TruSeq Rapid SBS Kit-HS and a HiSeq 2000 System with a TruSeq SBS Kit v3-HS (Illumina) in 50 bp/50 bp paired-end runs. Raw output BAM files were converted back to FASTQ format using Picard SamToFastq. Gene level counts were computed using htseq-count release_0.5.4 and the same gene models as used in the mapping step. HTSeq v.0.6.0 was used for feature counting. Counts were normalized and rescaled into log₂(counts) using the LIMMA R package ver-3.18. The genome used was HG19 with junctions from ENSEMBL (GRCh37.69_ENSEMBL) and a read overhang of 49. Human upregulated genes were upregulated genes (log fold change > 0, P < 0.05) in patients with FL versus patients with tFL (Fig. 3a).

Proton sequencing.

After RiboGreen quantification and quality control by Agilent BioAnalyzer, polyadenylated RNA was isolated using the Dynabeads mRNA DIRECT Micro Purification Kit (catalog no. 61021; Thermo Fisher Scientific) from 1–5 μg of total RNA. mRNA was then fragmented using ribonuclease III and purified. The quality and yield of the fragmented samples were evaluated using the Agilent BioAnalyzer before library preparation with the Ion Total RNA-Seq Kit v2 (catalog no. 4475936; Thermo Fisher Scientific) with 16 cycles of PCR. Samples were barcoded and template-positive ion sphere particles were prepared using the Ion OneTouch 2 System and Ion PI Template OT2 200 Kit v2 (Thermo Fisher Scientific). Enriched particles were sequenced on an Ion Proton System using 200 bp v2 chemistry. An average of 69 million reads per sample were generated. The output data (FASTQ files) were mapped to the target genome using the RNA Star aligner⁵², which maps reads genomically and resolves reads across splice junctions. We then computed the expression count matrix from the mapped reads using HTSeq (https://htseq.readthedocs.io/en/master/) and one of several possible gene model databases. The raw count matrix generated by HTSeq were then processed using the R/Bioconductor ver-2.13 package edgeR ver-3.4.2 to both normalize the data and compute genes with differential expression. Mouse upregulated genes were upregulated genes (log fold change > 0, P_adjusted < 0.1, Benjamini–Hochberg method) in B220⁺ cells isolated from mouse VavP-Bcl2; MYC versus VavP-Bcl2; vector (Fig. 3a,c).

Analysis of histone posttranslational modifications.

Cells (3 million) were lysed for 30 min on ice with nuclear isolation buffer (15 mM of Tris-HCl (pH 7.5), 60 mM of KCl, 15 mM of NaCl, 5 mM of MgCl_2, 1 mM of CaCl₂, 250 mM of sucrose, 0.3% NP-40; 1 mM of dithiothreitol (1:100) and 10 mM sodium butyrate were added immediately before use) to isolate intact nuclei. After pelleting at 600g for 5 min at 4 °C, nuclei were washed twice with nuclear isolation buffer without NP-40. Histones were extracted with 5 volumes of 0.2 M of H₂SO₄ for 1 h at room temperature and insoluble cellular debris was removed by centrifugation at 4,000g for 5 min. Soluble histones were precipitated from the supernatant by adding trichloroacetic acid at a final concentration of 20% (v/v) for 1 h at room temperature. Precipitated histones were pelleted at 10,000g for 5 min, washed once with 0.1% HCl in acetone then twice with 100% acetone with centrifugation at 15,000g for 5 min. After the final acetone wash, histones were dried briefly in a fume hood and stored at −20 °C until derivatization. Histones were propionylated and digested as outlined by Garcia et al.⁵⁴ with the modification of employing just one round of propionylation for 1 h before and following digestion. Targeted LC–tandem MS was performed on a TSQ Quantiva (Thermo Fisher Scientific) and raw data were analyzed in Skyline v4.2⁵⁵ according to published methods⁵⁶.

ChlP and ChIP–seq library preparation and sequencing analysis.

Histone ChIP, including H3K4me1 and H3K4me3 in mouse splenic tumor cells, were performed as described previously⁹. Briefly, 3 × 10⁶ mouse B220⁺ cells were fixed with 1% formaldehyde and fragmented by sonication of isolated nuclei (E220 Focused-ultrasonicator; Covaris) to achieve the enrichment of short chromatin fragments (100–500 bp). One microgram of antibody (H3K4me1, catalog no. ab8895; Abcam; H3K4me3, catalog no. ab8580; Abcam) was added to the chromatin lysate and incubated overnight at 4 °C. Enriched DNA was isolated by Dynabeads Protein A (Thermo Fisher Scientific) and subsequent reverse cross-linking and purification. ChIP–seq libraries were prepared from 1–5 ng ChIP DNA according to the instructions of the TruSeq ChIP Library Preparation Kit (catalog no. IP-202-1012; Illumina). Libraries were then sequenced on a HiSeq 2500 System as 50-bp single-end runs. ChIP–seq reads were aligned to the mm10 genome using the Burrows–Wheeler Aligner. Enriched loci were determined using model-based analysis of ChIP–seq 2 with narrow peaks for H3K4me3 and broad peaks for H3K4me1. Differentially enriched loci were calculated using differential binding with the thresholds of fold change > 1.2 and q < 0.05. Genomic distribution of loci was determined using the ChIPseeker framework.

Gene set enrichment analysis (GSEA).

GSEA analysis was generated from the GSEA preranked mode. The input file was log₂ (fold change of gene expression (SHMT2 versus vector)). The KEGG and hallmarks of cancer gene sets were used for the analysis.

Global DNA methylation analysis.

Dilutions of DNA (100, 50 and 25 ng μl⁻¹) in 5 μl of 0.1 M NaOH were denatured at 95 °C for 5 min. After neutralizing with 0.1 volume of 6.6 M ammonium acetate, 1 μl of each DNA was spotted on a nitrocellulose membrane and air-dried. The membrane was cross-linked by ultraviolet light, blocked with 5% milk overnight at 4 °C and incubated with anti-5-methylcytosine (5mC) antibody (1:500; clone 87G31; Millipore) for 1 h at room temperature. After incubation with horseradish peroxidase mouse secondary antibody (1:1,000; catalog no. SC-516102; Santa Cruz Biotechnology) for 1 h at room temperature, DNA was detected by enhanced chemiluminescence (Amersham). As DNA loading control, the same membrane was stained with 0.04% methylene blue for 5 min and washed for 5 min at room temperature.

DNA methylation analysis at base-pair resolution.

ERRBS.

Genomic DNA (200 ng) was modified by bisulfite treatment according to the manufacturer’s instructions (MethylDetector; Active Motif). DNA fragments were PCR-amplified and the direct sequencing reaction was performed using standard conditions (Applied Biosystems). The primers used were 5′-GTTGATAGTAGAAAGTTGTTTTGG and 5′-CTCTAACTCAAAATCATTCTATTCT. CpG methylation-sensitive digestion of genomic DNA was carried out with the EpiJET DNA Methylation Analysis Kit based on MspI/HpaII digestion (Thermo Fisher Scientific) according to the manufacturer’s instructions using 200 ng of purified genomic DNA. Quantitative PCR was performed using SYBR Green Select Master Mix (Thermo Fisher Scientific) according to the manufacturer’s instructions. The primers for common PCR were as follows: forward: 5′-CTCTGACTCTGGTCTGAAGT-3′; reverse: 5′-GTCTCCTGCTTCGTGTTATC-3′. The methylation level was quantified using the 2^−ΔC_T method and presented as the percentage cytosine methylation at the target gene site.

Analysis of differentially methylated CpG sites (DMCs).

R (https://www.R-project.org/) v.3.3.2 was used for the data analyses; methylKit v 0.5.3⁵⁷ was used for the analysis of differential methylation. The significance of DMCs was determined using a chi-squared test. For multiple hypothesis testing, the significance cutoff was 0.01. A methylation change >25% or <25% was used for significance.

Genomic annotation.

The genomic annotation reference files (CpG islands and RefSeq genes) for the DMC distribution analyses were obtained from the University of California Santa Cruz Genome Browser (https://genome.ucsc.edu/) using the December 2011 (GRCm38/mm10) assembly. Promoters were defined as transcription start sites ±1 kilobase (kb). CpG shores were defined as 2 kb flanking CpG islands, subtracted by any regions overlapping with nearby CpG islands. CpG shelves were defined as 2 kb flanking CpG shores, subtracted by any regions overlapping with nearby CpG islands and shores.

Integrative analysis.

DMCs annotated to gene promoters (transcription start sites ±2 kb) were considered for the integrative analyses with differentially expressed genes.

Metabolomics.

Sample preparation.

For the serine tracing experiment, isolated B cells from 3 different VavP-Bcl2; SHMT2 or VavP-Bcl2; vector tumors were cultured in serine-free RPMI medium containing 10% FCS, 1% l-glutamine and 1% penicillin-streptomycin for 30 min and 0.8 mM of l-serine (¹³C₃, 99%, ¹⁵N) (catalog no. CNLM-474-H-0.5; Cambridge Isotope Laboratories) was added for 3 h. The viability of these cells was tested by ViaCount before and after adding serine ¹³C₃; 2 ml of SU-DHL-4 cells (5 × 10⁵ ml⁻¹) with shSHMT2 or empty vector cells were seeded as triplicates in serine-free RPMI medium containing 10% FCS, 1% l-glutamine and 1% penicillin-streptomycin at 5 × 10⁵ cells per 2 ml for 30 min; subsequently, 0.8 mM of l-serine (¹³C₃, 99%, ¹⁵N) was added for 6 h. Cells were washed three times with 1 ml of cold 0.9% NaCl and polar metabolites were extracted in 1 ml of cold 80% methanol containing internal standards (catalog no. MSK-A2-1.2; Cambridge Isotope Laboratories). After extraction, samples were nitrogen-dried and stored at −80 °C until analysis by LC–MS.

LC–MS.

Analysis was conducted on a Q Exactive benchtop orbitrap mass spectrometer equipped with an Ion Max Source and a Heated Electrospray Ionization (HESI-II) probe, which was coupled to a Dionex UltiMate 3000 UPLC system (Thermo Fisher Scientific). External mass calibration was performed using a standard calibration mixture every 7 d. Dried polar samples were resuspended in 100 μl of water and 2 μl were injected into a ZIC-pHILIC 150 × 2.1 mm (5 μm particle size) column (EMD Millipore). Chromatographic separation was achieved using the following conditions: buffer A consisted of 20 mM of ammonium carbonate and 0.1% ammonium hydroxide; buffer B was acetonitrile. The column oven and autosampler tray were held at 25 and 4 °C, respectively. The chromatographic gradient was run at a flow rate of 0.150 ml min⁻¹ as follows: 0–20 min, linear gradient from 80 to 20% B; 20–20.5 min, linear gradient from 20 to 80% B; 20.5–28 min, hold at 80% B. The mass spectrometer was operated in full-scan, polarity switching mode with the spray voltage set to 3.0 kV, the heated capillary held at 275 °C and the HESI probe held at 350 °C. The sheath gas flow was set to 40 units, the auxiliary gas flow was set to 15 units and the sweep gas flow was set to 1 unit. MS data acquisition was performed within a range of 70–1,000 m/z, with the resolution set at 70,000, the AGC target at 10⁶ and the maximum injection time at 20 ms. Relative quantitation of polar metabolites was performed with XCalibur QuanBrowser 2.2 (Thermo Fisher Scientific) using a 5 ppm mass tolerance and referencing an in-house library of chemical standards. Metabolite levels were normalized to the total valine amount for each condition.

ATP content measurement.

The ATP content of cells was measured using the CellTiter-Glo 2.0 Cell Viability Assay (Promega Corporation) as recommended by the manufacturer.

Cell death assay (annexin V assay).

Cells were stained with APC-annexin V (catalog no. 640920; BioLegend) and 7-aminoactinomycin D (7-ADD) as recommended by the manufacturer. Analysis was performed with a BD LSR Fortessa cell analyzer and FlowJo software.

Immunoblot analysis.

Immunoblotting was performed according to standard procedures⁵⁸. Membranes were probed with the indicated primary antibodies (all Cell Signaling Technology unless stated otherwise): anti-SHMT2 (catalog no.12762); anti-SHMT1 (catalog no. 12612); anti-β-actin (catalog no. A5441; Sigma-Aldrich); anti-c-Myc (catalog no. 9402); anti-caspase-3 (catalog no. 9662, polyclonal); and anti-cleaved caspase-3 (catalog no. 9661S).

Quantitative PCR with reverse transcription (qRT–PCR) analysis.

qRT–PCR was performed as described by Ortega-Molina et al.⁸. β-actin (Actb (mouse, endogenous control, cat. no. 4352933E) and ACTB (human, endogenous control, cat. no. 4333762F); Thermo Fisher Scientific) was used as the housekeeping gene for input normalization of the data. The TaqMan gene expression assays used were: Shmt2 (Mm00659512); Ptprm (Mm00436095); Sash1(Mm00620677); Tusc1 (Mm02344017); SHMT2 (Hs1059263); TUSC1 (Hs00995084); PTPRM (Hs01124552); and SASH1 (00323932). Data were obtained using the QuantStudio 7 Flex Real-Time PCR System.

Drug treatment.

We used ABT-199 (CTA-199; Chemietek), BIX-01294 (B9311; Sigma-Aldrich) and 5-Aza (A2385; Sigma-Aldrich) for drug treatment. SHIN1 was provided by A. Friedman (Raze Therapeutics). Cells were seeded at 10,000 per 100 μl of media and each drug was added as 0, 0.001, 0.01, 0.1, 1 and 10 μM for 48 and 72 h. Cell viability was measured using the ViaCount assay. For combination therapy, we treated 4 different cell lines with a constant concentration of ABT-199 (250 nM) and different concentrations of BIX-01294 (0, 0.625, 1.25, 2.5, 5 and 10 μM). The combination index was determined using the CompuSyn v.1.0 software.

Statistics and reproducibility.

Sample sizes for comparisons between cell types or between mouse genotypes followed Mead’s recommendations⁴⁸. Samples were allocated to their experimental groups according to their predetermined type (that is, mouse genotype); therefore, there was no randomization. The investigators were not blinded to allocation during the experiments and outcome assessment. Survival in the mouse experiments was represented with Kaplan–Meier curves and significance was estimated with the log-rank test. In the case of Figs. 2a, 3i and 7a, only mice that developed lymphomas were considered; mice that did not develop lymphomas were censored and indicated with ticks in the Kaplan–Meier curves. qPCR data were obtained from independent biological replicates (n values are indicated in the corresponding figure legends). Normal distribution and equal variance was confirmed in the large majority of data; therefore, we assumed normality and equal variance for all samples. On this basis, we used an unpaired, two-tailed Student’s t-test to estimate statistical significance. For contingency analysis (proportion of H3K4me1/H3K4me2 peaks), we used the chi-squared test.

Reporting Summary.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability

Sequencing (RNA-seq, proton sequencing, ChIP–seq and ERRBS) data that support the findings of this study have been deposited in the Gene Expression Omnibus under accession codes GSE142336 and GSE139523. Previously published microarray data that were reanalyzed in this study are available under accession code GSE132929. Histone mass spectrometry data were deposited to MassIVE under accession code MSV000085251. Source data for Figs. 1–7 and Extended Data Fig. 2 have been provided as Source Data Files 1–12. All other data supporting the findings of this study are available from the corresponding author upon reasonable request. Source data are provided with this paper.

Code availability

Any custom computer code or algorithm previously used to generate results that are reported in this paper and central to its main claims are available upon request.

Extended Data

Extended Data Fig. 1 |. De novo serine synthesis pathway in human B cell lymphoma.

a, Bar graphs presenting the functional gain (red) or loss (blue) of SHMT2 located on chromosome 12 by DNA copy number analysis of 568 DLBCL and 176 FL tumors. b, The frequency of loss (blue), gain (red) or diploid (black) status of serine synthesis pathway genes in 568 human DLBCL and 176 human FL tumors. c, Diagrams demonstrating the overlaps of amplification (red) and loss (blue) of SHMT2 vs other serine biosynthesis pathway enzymes (PSPH, PHGDH, PSAT1, SHMT1) in 568 DLBCL and 176 FL tumors. d, Bar graph showing the frequency of SHMT2 amplification in different subtype (GC-like, ABC-like or unclassified) of 249 human DLBCL tumors.

Extended Data Fig. 2 |. SHMT2 promotes lymphomagenesis in vivo.

a, Diagram of FL mouse model. Fetal VavPBcl2 HSCs were transduced by MSCV-GFP plasmid carrying SHMT2 cDNA or empty vector and injected to lethally irradiated female mice. b, Representative graphs of flow cytometry analysis comparing GFP⁺ HSCs before injection vs GFP⁺ splenic lymphoma cells from VavPBcl2;vector- and VavPBcl2;SHMT2- induced tumors collected 5 months after injection. c, Dot plot representing the initial GFP⁺ cells in hematopoietic stem cells before injection vs GFP⁺ cells enriched in splenic cells collected from VavP-Bcl2;vector (N = 5 mice) and VavP-Bcl2;SHMT2 (N = 10 mice) tumors. Two-tailed Student’s t-test was used to determine statistical significance; VavP-Bcl2;vector: P(HSCvsLymphoma)= 0.443, NS; VavP-Bcl2;SHMT2: P(HSCvsLymphoma)=0.0006. d, Representative images of histology studies of VavPBcl2; vector and VavPBcl2;SHMT2 lung. The slides were stained with H&E, and antibodies for B220, TUNEL, Ki67, PNA. This experiment was independently repeated three times with similar results. Scale Bars, 500 nm. e, tumor clonality analysis on B220⁺ cDNA collected from VavPBcl2;vector vs VavPBcl2;SHMT2 tumors. Each lane corresponds to one tumor. This experiment was independently repeated two times with four independent samples in each genotype with similar results f, Immunoblot against SHMT2, SHMT1 and ACTIN in DLBCL cell lines carrying two different short hairpins against SHMT2. This experiment was independently repeated two times with similar results. The uncropped images of the original blots are presented in Source Data Extended Data File 12. The numerical data for this figure are presented in Source Data Extended Data File 2.