Genomic determinants of response and resistance to inotuzumab ozogamicin in B-cell ALL

Patient characteristics and response to InO by subtype

We studied 85 adult patients with R/R (n=81, 95%) or newly diagnosed (n=4, 4.7%) B-ALL treated with InO monotherapy (n=55, 65%) or InO in combination with low-intensity chemotherapy (n=30, 35%). This was a retrospective study, and the inclusion criterion was availability of pre-InO and/or post-InO tumor samples, either as banked cells or nucleic acids from bone marrow or peripheral blood-derived tumor cells. All patients had detectable disease (>5% blasts) in the bone marrow at the time of receiving treatment. Demographic and clinical features are summarized in Table 1. The median age was 42 years (range 18-84). The median number of prior therapies was two (range 0-5) and 55% patients received InO as their second or subsequent course of salvage therapy. The cohort was divided into responders (those who achieved complete remission [CR, n=26] or complete remission with incomplete hematologic recovery [CRi, n=25]) and non-responders (NR, either no response [n=31] or early death [n=3]). Thirty-one (61%) responders subsequently underwent allogeneic hematopoietic stem cell transplantation (HSCT) after attaining remission with InO (Figure 1, Supplementary Table 1). Of the 35 patients with post-InO relapse, 11 (31%) relapsed during InO therapy, 7 (20%) after completion of InO without undergoing HSCT, and 17 (49%) after completion InO and after subsequent HSCT. Samples were obtained pre-InO only (n=42), pre-& post-InO (n=28), and post-InO only (n=15). The post-InO samples included samples from patients who achieved CR/CRi but subsequently relapsed (n=27), and patients who were refractory to InO (n=16).

• Download figure
• Open in new tab

Figure 1. Response to inotuzumab.

Swimmer plot for InO responders (n=51). Each bar represents from the start of InO therapy to last follow up. Once attaining CR/CRi, patients were off InO therapy, and 31 responders subsequently received hematopoietic stem cell transplantation (HSCT).

Overall, 51 of 85 patients (60%) achieved CR/CRi with InO treatment. We investigated whether genomic subtypes were associated with response to InO. Genetic subtypes were determined by analysis of gene rearrangements, DNA copy number alterations, somatic single nucleotide variation (SNV) or insertion/deletion (indel) mutations and gene expression data (Supplementary Table 1). The most common (>10%) subtypes were BCR::ABL1-like (29%), BCR::ABL1-positive (14%), low hypodiploid (13%), and KMT2A-rearranged (KMT2A-R, 11%), consistent with the known distribution of genomic subtypes observed in adult B-ALL.¹⁶ Less frequent B-ALL subtypes (<5% each) were grouped together as “other subtypes” (n=23) for outcome analysis, and comprised TCF3::PBX1, DUX4-rearranged, BCL2/MYC, CDX2/UBTF, hyperdiploid, MEF2D-rearranged, ZNF384-rearranged, ETV6::RUNX1, ZEB2/CEBP, PAX5 P80R, and unclassified (B-other) cases. The “other subtypes” were predominantly standard-to-intermediate risk based on molecular classification of risk assessment, while BCR::ABL1-like, low hypodiploid and KMT2A-R were considered high risk.¹⁶ As expected, the CR/CRi rates were generally lower in the BCR::ABL1-positive, BCR::ABL1-like, low hypodiploid and KMT2A-R subtypes than in “other subtypes” (Supplementary Table 2). The event-free survival (EFS) and overall survival (OS) rates varied among subtypes (Supplementary Figure 1A-B). Compared to “other subtypes” group, EFS was shorter in BCR::ABL1-positive (P=0.02) and low hypodiploid (P=0.006), but the differences were not significant in BCR::ABL1-like (P=0.5) or KMT2A-R (P=0.1).

As KMT2A-R ALL may undergo lineage switch following CD19-directed therapy^17–19, we examined the expression of CD22, CD19 and myeloid lineage markers in KMT2A-R patients with post-InO samples available (n=3, 2 of which also had paired pre-InO samples). The KMT2A-R cases demonstrated post-InO reduced CD22 RNA expression (47%-59% decrease) and percentage of CD22+ blasts (75%-90% decrease), with no change in the level of CD19 (supplementary Figure 1C). One sample (SJALL079501) had elevated RNA expression of myeloid lineage genes (MPO, ITGAM [CD11b], ANPEP [CD13], CD33, CD14) and the stem cell marker KIT (CD117), without detectable changes of protein expression on flow cytometry or lineage switch at post-InO relapse. This patient experienced lineage switch and relapse with acute myeloid leukemia after subsequent therapy with blinatumomab. No signs of lineage switch were observed for the other two KMT2A-R samples.

Multiple mechanisms of CD22 antigen escape

No somatic CD22 mutations were observed in pre-InO baseline samples. After InO treatment, acquired CD22 mutations were present in 11% (3/27) of post-InO relapsed samples, but not in post-InO refractory samples (0/16). One patient with acquired CD22 mutations (SJALL058834) was treated with InO monotherapy, and the other two (SJALL058975 and SJALL074541) with InO in combination with low-intensity chemotherapy. In all three cases, the relapse occurred after completion of InO and after subsequent HSCT. There were multiple CD22 mutations per sample (range 3-4) with a total of 10 mutations identified (Table 2). Mutations occurred in the extracellular and cytoplasmic domains of CD22, with exon 2 that encodes the first of the 7 immunoglobin (Ig) domains as the mutational hotspot (Figure 2A). The identified mutation types included frameshift indel (n=4), nonsense SNV (n=2), missense SNV (n=3) and stoploss SNV (n=1) (Figure 2B). The sum of variant allele frequency (VAF) was proportional to the blast% (Table 2), suggesting all or nearly all cells harbored a CD22 mutation. We identified multiple mechanisms of CD22 escape by genetic mutations as described below, including protein truncation, epitope alteration, and protein destabilization.

• Download figure
• Open in new tab

Figure 2. Multiple mechanisms of CD22 antigen escape.

(A) Protein domain plot of all post-InO acquired CD22 mutations (n=10) identified in this cohort. (B) Wild-type (WT) and the predicted mutant CD22 protein structures. Grouped by mutation types. (C) Clinical flow plots of paired pre- and post-InO patient samples. (D) Antibodies directed to different CD22 epitopes (InO and SHCL1 that bind the first Ig domain of CD22, and RFB4 that binds the third). (E-F) Functional characterization of CD22 p.R75T and p.C529Y. CD22 protein expression and binding were performed using antibodies directed to different CD22 epitopes. (G) Stack plots depicting relative abundance of alternative splicing of CD22 exon1 in pre-InO baseline samples, comparing responders (EFS>12m and EFS<12m) and non-responders. (D) Stack plots depicting relative abundance of alternative splicing of CD22 exon1, comparing pre-vs post-InO samples.

TP53 mutations were associated with primary and acquired resistance to InO

We characterized baseline genomic alterations in genes known to be mutated in leukemia (Figure 3A). Alterations in transcriptional regulators (excluding B-lineage transcription factors) were more common in responders than non-responders (54% vs 23%, P=0.01). Samples from non-responders were enriched with alterations in pathways of cell cycle (77% vs 54% in responders, P=0.05) and Ras (35% vs 13%, P=0.04). Notably, TP53 mutations were more common in non-responders than responders at pre-InO baseline (48% vs 23%, P=0.04, Figure 3A). TP53 mutations were identified in 34% (24/70) of the cases with pre-InO samples. Although 9 of TP53-mutated cases (37.5%) attained CR/CRi, 8 relapsed within 12 months, and TP53-mutated cases had shorter EFS and OS comparing to cases without TP53 mutations (Figure 3B). Among patients who were refractory or relapsed after InO, TP53 mutant clones emerged or expanded, suggesting TP53 mutations promoted acquired resistance to InO. Of 14 cases with TP53 mutations in post-InO samples, two acquired TP53 mutations post InO (one was KMT2A-R, two mutations, sum of VAF 0.7; one was ZNF384-rearranged, VAF 0.39, Figure 3C); three had TP53 mutation VAF increased post InO (Figure 3C); and in 9 cases, mutations were hemizygous and/or maintained high VAF during disease progression (Supplementary Figure 2).

• Download figure
• Open in new tab

Figure 3. TP53 mutations in primary and acquired resistance to InO.

(A) Baseline genomic alterations in major leukemia genes, comparing responders vs non-responders. (B) EFS and OS of TP53WT vs TP53mut patients. The elapsed observation time is plotted on the curve as a circle. Symbols for both events and censored observations are plotted so that each subject was shown. Censored observations can be clearly seen as circles along the horizontal portion of the curve. Medians were estimated with the Kaplan-Meier method. P values were determined by log-rank test. (C) Dynamic changes in TP53 VAF in 5 cases with paired pre- and post-InO samples. The VAF for each mutation was shown, along with the sample blast count at the corresponding time point. The time elapsed from start of InO to treatment failure (relapse/refractory) was shown in days. EFS, Event-free survival; OS, overall survival; VAF, variant allele frequency.

Hypermutation-driven CD22 antigen escape from InO

We examined the number and type of post-InO acquired mutations to investigate which DSBs repair pathways were leveraged to mitigate InO toxicity. Single base substitutions (SBS) were the primary mutation type in all but one patient (Figure 4A), suggesting in most cases the rapid but indel-prone DSBs repair pathways (NHEJ, a-EJ or SSA) was not primarily employed for InO-induced DNA damage; instead, HR-mediated repair, a higher fidelity mechanism, may have been utilized. Based on the density histogram of the number of acquired mutations per sample in our cohort (Figure 4A), we divided patients into post-InO low-mutators (n=21, 36-252 acquired mutations), moderate-mutators (n=3, 447-759 acquired mutations), and hyper-mutators (n=2, 1737-2433 acquired mutations). Patients who were refractory to InO and those who relapsed during InO therapy were all low-mutators; in contrast, patients who relapsed after completion of InO were low-, moderate-, or hyper-mutators (Figure 4A). Notably, the two hyper-mutators (SJALL058834 and SJALL074541) and one moderate-mutator (SJALL058975) acquired multiple CD22 mutations post-InO (Figure 4A and 4B), suggesting accelerated acquisition of mutations was one driving force behind CD22 antigen escape to InO.

• Download figure
• Open in new tab

Figure 4. Post-InO acquired determinants of resistance.

(A) The number (top) and oncoprint (bottom) of post-InO acquired mutations. The dotted lines (at 300 and 900, number of post-InO acquired mutations) were used as cutoffs to divide patients into post-InO low-mutators, moderate-mutators and hyper-mutators. (B) Fishplots showing acquisition of mutations in CD22 and KMT2D. (C) The mutational signatures of post-InO acquired mutations in hyper-mutators SJALL058834 (>10bp insertions) and SJALL074541 (SBS6, MMR-deficient). (D) A simplified schematic of compromised G1/S DNA damage checkpoint by acquired mutations in InO-treated patients. Created with BioRender.com.

Mutational signature analysis revealed that the hypermutation was generated either by a-EJ repair in response to InO-induced DSBs, or by tumor-intrinsic mismatch repair (MMR) deficiency. The primary mutation type of post-InO acquired mutations in hyper-mutator SJALL058834 were indels (86%, Figure 4A), most commonly insertions of >10 nucleotides (74%, Figure 4C). These features were not present in its pre-InO sample (7% mutations were indels, 1.7% were insertions >10 nucleotides), indicating this post-InO hypermutation signature was treatment induced. Of the four DSBs repair pathways, a-EJ was known to create long insertions,¹¹ suggesting SJALL058834 tumor cells might have resisted InO through a-EJ DNA repair. In contrast, hyper-mutator SJALL074541 acquired SBS6 signature mutations attributable to defective MMR (Figure 4C). Its pre-InO sample harbored a loss of function mutation in MLH1 (MLH1 p.L155fs) and was therefore tumor-intrinsic MMR deficient, which was recognized as the cause of hypermutation and corresponding mutational signature.

Resistance to InO by compromised G1/S checkpoint or mitigation of replication stress

We observed acquired loss-of-function mutations post-InO in genes that regulate the G1/S DNA damage checkpoint, in addition to the previously mentioned TP53 mutations (Figure 4D). SJALL058848 acquired missense mutation ATM p.V2424G, a known loss-of-function pathogenic variant that impairs ATM kinase activity.²² SJALL058975 acquired the nonsense p.R58X mutation in the CDKN2A tumor suppressor upstream of TP53. As the G1/S checkpoint is often compromised during tumorigenesis,²³ our observations suggested that acquired mutation-induced defects in G1/S DNA damage checkpoint contribute to InO resistance. Importantly, as tumor cells with compromised G1/S checkpoint depend on the G2/M checkpoint,²³ this vulnerability may be exploited through the use of selective molecularly targeted agents (e.g., G2/M checkpoint inhibitors) in combination with InO.

Another frequently post-InO mutated gene was KMT2D (MLL4, n=3, Figure 4A). SJALL058838 (ZNF384-rearranged) had a pre-existing loss-of-function KMT2D mutation (p.R2678X) and acquired another mutation (KMT2D p.A3292fs) post-InO, resulting in disruption of both KMT2D alleles (Figure 4B). Loss of KMT2D inhibits recruitment of the MRE11 nuclease to stalled replication forks, resulting in replication fork protection and cell survival by mitigation of replication stress; this is a known resistance mechanism to DNA damaging agents such as PARP inhibitor.²⁴

Genome wide CRISPR screening identified InO resistance and sensitization targets

We undertook genome wide CRISPR/Cas9 screening of the NALM-6 (DUX4-rearranged) cell line to understand genomic determinants of response and resistance to InO. Using false discovery rate (FDR) <0.1 at passage 5 (P5) and/or passage 10 (P10), we identified 274 genes whose loss led to InO sensitization (negatively selected) and 102 genes whose loss led to InO resistance (positively selected) (Supplemental Figure 3A, Supplementary Table 3). Gene set enrichment analysis (GSEA) identified multiple pathways relevant to InO sensitization, including DNA damage and DNA repair pathways (Supplemental Figure 3B). In addition, ABCC1, encoding an ABC transporter protein, was among the top sensitizing targets (Supplemental Figure 3A), suggesting ABCC1 mediated drug efflux was an important mechanism for InO clearance. Interestingly, DNTT (encoding terminal deoxynucleotidyl transferase, TdT) came out as the top resistance gene (Supplemental Figure 3A, 3C). The canonical function of TdT is to generate lymphocyte antigen receptor diversity by adding nontemplated nucleotides to DNA 3′ terminal ends during V(D)J recombination.²⁵ Recent studies demonstrated that TdT was the key factor driving the DSBs repair outcome, and DNTT knockout in Jurkat cells shifted insertion-dominant repair to deletion-dominant repair.²⁶ Another study reported that TdT-negative T cell acute lymphoblastic leukemia/lymphoma (T-ALL/LBL) patients had poor therapeutic responses; TdT loss decreased apoptosis, induced the accumulation of chromosomal abnormalities and tolerance to abnormal karyotypes.²⁷ DNTT loss might modulate the DSBs repair response to InO and result in increased tolerance to InO-induced DNA damage, leading to InO resistance. We observed decreased DNTT RNA expression in patient samples at post-InO (P=0.03, mean of different -1.19 log2[counts per million], 95% CI [-2.23, -0.11], Supplemental Figure 3D); no post-InO acquired DNTT mutations were identified in this dataset. TdT is commonly utilized in clinical flow and can be prospectively evaluated as a novel marker of InO response. CD22 arose as the resistance gene as expected. Moreover, loss of CHEK2, the crucial G1/S checkpoint gene, led to InO resistance (Supplemental Figure 3A and 3C). This supported the observation and hypothesis from patient data (Figure 4D) that tumor cells resisted InO by compromise of the G1/S DNA damage checkpoint.

Patients and clinical specimens

We retrospectively studied 85 adult B-ALL patients treated at MD Anderson Cancer Center. The selection criteria were patients with pre-InO and/or post-InO tumor samples (banked cells/DNA/RNA from bone marrow or peripheral blood-derived tumor cells) available. The study was approved by the institutional review boards of MD Anderson Cancer Center and St Jude Children’s Research Hospital, with informed consent.

Transcriptome sequencing

Total stranded transcriptome sequencing (RNA-seq; 100bp paired-end reads) was performed using the TruSeq Stranded Total RNA library preparation kit and sequencing performed using HiSeq 4000 or NovaSeq 6000 platforms (Illumina) to generate a minimum of 100 million reads per sample library with a target of greater than 90% mapped reads. The low input RNA library preparation kit (NuGen Ovation V2) was used for samples with limiting material (2-100 ng). The adapters in sequencing reads were trimmed with Trim Galore (v0.4.4; https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/, -q 20 –phred 33 –paired). The trimmed sequencing reads were mapped with STAR (v2.7.9a)³⁰ to human genome GRCh38. The expected gene counts calculated using RSEM³¹ for each sample were compiled to one gene count matrix. Only genes annotated as level 1 or 2 by GENCODE (v31) were kept in the downstream analysis.

Whole genome and exome sequencing

Whole exome sequencing (WES) of genomic DNA was performed using the TruSeq DNA Exome library preparation kit (Illumina). Sequencing (100bp paired-end reads) was performed using the NovaSeq 6000 platform (Illumina) to an minimum haploid coverage of 100x. Whole genome sequencing (WGS) libraries were prepared using the HyperPrep library preparation kit (Roche) and sequenced using the NovaSeq 6000 System (Illumina) to a target depth of 800 million paired-end 150bp reads per sample for 30x average haploid coverage. The paired end sequencing WGS and WES reads were mapped with BWA-MEM³² to human genome GRCh38. The alignment quality was assessed using Qualimap. Genomic data is publicly available and has been deposited in the European Genome Phenome Archive, Accession EGAS50000000067.

Genetic subtyping

Subtypes were determined by integrating gene expression, rearrangement, copy number and SNV/indel as previously described.^33,34. Samples without RNA-seq were usually assigned to the B-other subtype; however, in cases without RNA-seq but with WGS, subtype-defining rearrangements if detected by WGS were used to assign subtypes. Cytogenetic and FISH data collected on the clinical trials was compared against the results of genomic analyses, and discrepancies were resolved by in depth review.

Mutation analysis

Variants were called by Mutect2 (v4.1.2.0)³⁵ using the tumor-only calling approach. Multiple filtering steps were applied to exclude potential calling artifacts. The variants passing the filtering steps fulfilled the following criteria: coverage depth of the variant > 10, variant allele frequency > 0.01, alternative allele count > = 4, allele population frequency in public databases < 0.01 (gnomadAD, 1000 genomes, ExAC and Exome Sequencing Projects), mappability > 0.7, not co-localized with repeat elements and GC percentage between 0.4 and 0.6. The variant annotation was performed using Annovar.³⁶ Oncoprint was generated using ProteinPaint.³⁷ Post-InO acquired mutations were identified by using pre-InO sample to filter its paired post-InO samples. Two-dimensional MAF plots were used to visualize the relationship between pre- and post-InO samples. Clonal evolution was visualized using fishplot.³⁸ Post-InO acquired mutations were used to analyze post-InO mutational signatures. Mutational signatures were profiled by fitting SNV and indel counts per 96 tri-nucleotide contexts to the COSMIC signatures version 3.2. Signatures with <5% overall contribution were excluded from the summary.

Copy number analysis using whole genome sequencing

Chromosome level copy numbers were estimated using coverage analysis of WGS. Control-FREEC (Control-Free Copy number caller)³⁹ software was used to estimate genome-wide copy-numbers using the mode without control sample. Read counts were corrected by GC content and mappability. Window size was automatically adjusted using coefficientOfVariation value of 0.08. Boundaries of focal copy number alterations (CAN) were modified if they overlapped with a structure variant call from Delly (v0.7.7).⁴⁰ Recurrent CNAs were analyzed using GISTIC2. ⁴¹

CD22 mutations modeling and characterization

The protein structural model of full-length CD22 was obtained from AlphaFold2⁴² and mutations were introduced using foldX²¹ in order to assess effects on protein stability. CMV-CD22-IRES-GFP vector was purchased from GeneCopoeia, Inc. CD22 SNV/indel mutations were generated using Q5 Site-Directed Mutagenesis Kit (New England Biolabs) and the full mutation sequences were validated by Sanger sequencing. CMV-IRES-GFP empty vector, CD22 WT vector, and CD22 mutant vectors were transfected into HEK-293T cells using FuGENE HD (Promega). Transfected HEK-293T cells were dissociated using TrypLE Express Enzyme (Gibco, Thermo Fisher Scient ific) before staining. CD22 binding experiments were performed using antibodies directed to different CD22 epitopes (supplementary Table 4). For flow cytometry, cells were stained with InO complexed with FabuLight AF647 Goat Anti-Human IgG, Fcγ fragment specific (Jackson Immunoresearch), CD22 clone SHCL1 (BD Biosciences), or CD22 clone RFB4 (Thermo Fisher Scientific). The flow gating strategy was summarized in supplementary Figure 4. Western blot was performed using antibodies specific to the N-terminus (clone E7L6Z, Cell signaling) or C-terminus (polyclonal, BosterBio) of CD22 protein and imaged using Odyssey CLx Imager (LI-COR Biosciences).

CD22 exon junction analysis

CD22 junction spanning reads were obtained from the STAR “*_SJ.out.tab” result files based on its chromosome location (chr19: 35329187 – 35347361), and the exon position was annotated using CD22 isoform NM_001771.4 by BEDTools.⁴³ The CD22 Δex2* isoforms that skipped exon2 (e2) and variably connected exon1 (e1) to several possible downstream exons (exon3 [e1_e3], exon3-alt [e1_e3alt], exon4 [e1_e4], exon5 [e1_e5], exon6 [e1_e6], exon7 [e1_e7]) were previously reported.¹⁰ Each entry was normalized by dividing by the total number of junction-spanning reads as previously described.¹⁰ Samples with low CD22 expression (< 20 reads) were excluded from the analysis as the exon junction usage could not be accurately estimated due to low number of reads.

Statistical analysis

Differencesbetween groupswere tested using two-sided Fisher’s exact test. Median PFS and OS were estimated using the Kaplan-Meier method, with the P values, hazard ratio (HR) and corresponding 95% confidence intervals (CIs) calculated using log-rank method. GraphPad Prism 8.3.0 was used for statistical analysis.

Genome-wide CRISPR/Cas9 screen

Whole genome CRISPR/Cas9 screen was performed as previously described.⁴⁴ Briefly, human CRISPR Knockout Pooled Library (Brunello)^45–47 (Addgene #73179) lentivirus was obtained from the Center for Advanced Genome Engineering at St. Jude. NALM-6 cells were infected with the genome-wide sgRNA library at 20%-40% transduction efficiency to ensure no more than one sgRNA per cell, and infections were performed in three biological replicates per cell line and achieved a target representation of ∼500 cells per sgRNA. At 48 hours post transduction, transduced cells were selected by 0.5µg/ml puromycin for 7 days. After 7 days, from each replicate, at least 40×10⁶ cells were harvested for the baseline counts (Day 0 control), and remaining cells were divided into two equal groups. One group was treated with 1 ng/ml InO (half maximal inhibitory concentration, IC50) while the other group served as no-treatment control. Cells were passaged every 2-3 days with complete media change. Cells from the baseline and the endpoints (Passage 5 and 10) were subjected to genomic DNA extraction using QIAamp Blood Maxi Kit (Qiagen). Genomic DNA from each passage were amplified as per Doench et al.⁴⁵ and sequenced. CRISPR KO screens were analyzed using MAGeCK (0.5.9.5).⁴⁸ Pathway enrichment analysis was performed using Pre-ranked Gene Set Enrichment Analysis (GSEA) against ontology gene sets (C5 version c5.all.v2022.1.Hs.symbols.gmt) using L2FC of entire genes in the CRISPR library.