Abstract

   We used epidemiological data from 21195 patients with cancer and 180407
   matched controls, including in-depth analyses in 216 cancer patients,
   to discover clinical and molecular determinants that predispose cancer
   patients to breakthrough infections. Patients with B cell malignancies,
   with differential expression of CD24, CDK14 and PLEKHG1, were most
   susceptible to breakthrough infections, suggesting that these patients
   may require more booster immunisations to ameliorate cellular responses
   and immune protection against COVID-19.

   Subject terms: Cancer, Vaccines, RNA vaccines
     __________________________________________________________________

   Patients with cancer may have decreased humoral and T cell responses to
   mRNA vaccines, exposing them to increased risk of severe
   COVID-19^[86]1–[87]7. The rates of breakthrough infections remain
   higher, even in fully vaccinated patients with cancer, indicating that
   regular boosters may be required to upkeep the longevity of
   immunity^[88]6,[89]8,[90]9. In October 2024, the Advisory Committee on
   Immunisation Practices recommended that moderately or severely
   immunocompromised individuals may receive additional doses based on
   shared clinical decision-making^[91]10. However, it remains uncertain
   whether all, or only a subset of patients, require regular booster
   doses to protect against SARS-CoV-2 infection.

   In this study, we used epidemiological data from 21195 patients with
   cancer and 180407 matched controls, as well as in-depth translational
   data from a subset of 216 patients, to discover clinical and molecular
   determinants that predispose to SARS-CoV-2 breakthrough infections.
   Transcriptomic and serological analysis was performed on a subset of
   patients to discover host characteristics associated with increased
   rates of breakthrough infections.

   Overall, 21,044 individual patients with cancer (7463 receiving active
   treatment and 13,732 cancer survivors) with 180,407 matched controls
   from the general population were included across the delta and omicron
   waves. Cancer survivors were defined as individuals with a previous
   diagnosis of cancer and not currently receiving any active cancer
   treatment. The demographic characteristics were detailed in
   Supplementary Table [92]1. More than 95% of the patients with active
   cancer, cancer survivors and the matched controls from the general
   population were fully vaccinated with at least two doses of mRNA
   SARS-CoV-2 vaccines (Supplementary Table [93]1). Despite similar
   vaccination rates, the cumulative incidence of SARS-CoV-2 infections
   among both active-treated patients with cancer and cancer survivors was
   higher than matched controls, especially after the easing of nationwide
   non-pharmacological interventions of infection control such as mask
   mandates (Supplementary Fig. [94]1a).

   To dissect the determinants associated with increased breakthrough
   SARS-CoV-2 infections in patients with cancer, we studied a cohort of
   216 patients with cancer who participated in the Oncovax study, where
   blood samples were drawn before the third vaccine dose was
   administered, at a median of 112 days (interquartile range 91–140)
   after the first dose (Fig. [95]1a, Supplementary Table [96]2). Patients
   received primary and booster doses of mRNA SARS-CoV-2 vaccines and were
   followed up for SARS-CoV-2 infections for 1 year. The predominant
   SARS-CoV-2 strain that was circulating was the Omicron variant. The
   third vaccine dose was given ~3 months after the second dose (Fig.
   [97]1a). Patients with breakthrough infection (n = 103) showed
   comparable vaccination rates to those without breakthrough infection
   (n = 113) (Supplementary Table [98]3). Consistent with the
   epidemiological observations in the larger cohort, patients with cancer
   on active treatment had similar breakthrough infection rates compared
   to cancer survivors (Fig. [99]1b). The types of active treatment,
   including chemotherapy, immunotherapy and targeted therapy, did not
   affect breakthrough infection rates (Supplementary Fig. [100]1b) or
   severity of symptoms (Supplementary Table [101]3). The differences in
   breakthrough infection rates were also not attributed to demographic
   characteristics, such as age, sex and ethnicity (Supplementary Table
   [102]3). Next, we examined if tumour types were associated with
   breakthrough infection rates. Patients with underlying haematological
   malignancies had significantly higher breakthrough infection rates
   within 12 months of primary vaccination, regardless of whether they
   were on active treatment (Fig. [103]1c). As the tumour type had the
   greatest effects on breakthrough infection rates (Supplementary Table
   [104]3), we subsequently focused our gene expression and immunological
   analyses on the patients with haematological malignancies.

Fig. 1. Baseline characteristics of patients with cancer associated with
SARS-CoV-2 breakthrough infections.

   [105]Fig. 1
   [106]Open in a new tab

   a Schematic of the study schedule for 273 patients with cancer on the
   Oncovax study who have at least three vaccine doses. Patients received
   primary and booster doses of mRNA SARS-CoV-2 vaccines and were followed
   up for SARS-CoV-2 infections for 1 year. Blood drawn pre-vaccine dose 3
   was sent for RNA sequencing and serological assays. Created in
   BioRender. Zhong, Y. (2025) [107]https://BioRender.com/z05h519. b
   Breakthrough infection rates among patients with cancer who are in
   active treatment compared to cancer survivors. c Breakthrough infection
   rates among patients with haematological malignancies and those with
   solid tumours. Chi-square test, **P < 0.01. d Volcano plot showing
   increased (red) and decreased (blue) expression of genes in patients
   with haematological malignancies compared to those with solid tumours.
   (Fold-change > 1.5; FDR-adjusted p value < 0.05). e Venn diagram of
   pathway enrichment modules that were significantly enriched in patients
   with haematological malignancies and with breakthrough infections. f
   Significantly enriched modules (P195, P122 and P123) (adjusted p
   value < 0.05) in patients with haematological malignancies and with
   breakthrough infections. g Network plot of leading-edge genes in the
   three modules P195, P122 and P123. CD24, CDK14 and PLEKHG1 were common
   leading-edge genes shared between the three modules. h Breakthrough
   infection rates among patients with different haematological and solid
   tumour cancer types. One-way ANOVA is used to evaluate significance.

   To understand the baseline characteristics associated with
   haematological malignancies and breakthrough infections, total RNA was
   extracted from the whole blood of 73 patients who had available samples
   at the pre-dose 3 time point. These patients had similar clinical
   features to the main Oncovax cohort of 216 patients (Fig. [108]1a,
   Supplementary Table [109]2). There were more genes that were
   significantly downregulated in patients with haematological
   malignancies compared to those with solid tumours (fold-change > 1.5,
   false discovery rate [FDR] adjusted p value < 0.05) (Fig. [110]1d). To
   further dissect the molecular pathways and immune cell types involved,
   we performed Gene Set Enrichment Analysis (GSEA) using the blood
   transcription modules (BTMs), a curated database comprising an
   integrated large-scale network of publicly available human blood
   transcriptomes^[111]11. B-cell modules were most negatively
   downregulated in patients with haematological malignancies compared to
   those with solid tumours (Supplementary Fig. [112]1c). On the other
   hand, the positively enriched pathways consisted of T cell and myeloid
   cell modules (Supplementary Fig. [113]1c). Furthermore, to identify the
   pathways that were also associated with the higher risk of breakthrough
   infections, we performed GSEA comparing patients with and without
   breakthrough infections regardless of the tumour type (Supplementary
   Fig. [114]1d). Interestingly, the B-cell modules were also negatively
   downregulated in patients with breakthrough infections (Supplementary
   Fig. [115]1d). Taken together, we identified that B-cell modules (P122,
   P123, P195) were negatively downregulated in patients with
   haematological malignancies and with breakthrough infections (Fig.
   [116]1e, f), with CD24, CDK14 and PLEKHG1 as the top common
   leading-edge genes that were represented in the B-cell modules (Fig.
   [117]1g). Moreover, to examine the types of B cells that express CD24,
   CDK14 and PLEKHG, we used the Human Protein Atlas and Monaco
   datasets^[118]12. The expression of CD24, CDK14 and PLEKHG1 was found
   to be most highly expressed in naïve and memory B cells (Supplementary
   Fig. [119]1e–g). Although CD24 can be detected in eosinophils, and
   CDK14 can be detected in basophils and neutrophils, these immune cells
   were not significantly enriched in patients with haematological
   malignancies and with breakthrough infections (Fig. [120]1f,
   Supplementary Fig. [121]1c, d).

   The enrichment of B-cell modules in patients with haematological
   malignancies and breakthrough infection led us to postulate that the
   type of haematological malignancies may influence the rates of
   breakthrough infections. We thus stratified the haematological
   malignancies by immune cell type. Consistent with the gene expression
   data, we observed that patients with B-cell malignancies, including
   Hodgkin’s lymphoma, acute lymphoblastic leukaemia, diffuse large B-cell
   lymphoma and follicular lymphoma, accounted for a greater difference in
   risk profile compared to those with myeloid haematological malignancies
   and other types of solid tumours (Fig. [122]1h). In addition, to
   examine if the B-cell malignancies affected the humoral responses to
   the mRNA vaccine, we measured the anti-spike (S) IgG titre and
   variant-specific neutralising antibody titres against SARS-CoV-2
   variants Wuhan-Hu-1, Beta, Delta, Omicron BA.2, Omicron XBB.1.16, and
   JN.1. Using 72 subjects that had similar clinical features as the
   entire cohort of 220 patients (Fig. [123]1a, Supplementary Table
   [124]3), we first compared the anti-S IgG or neutralising antibody
   titres between patients with and without breakthrough infection. No
   difference in anti-S IgG or neutralising antibody titres was observed
   (Supplementary Fig. [125]1h, i). Likewise, antibody titres were not
   different between patients with haematological malignancies and solid
   tumours (Supplementary Fig. [126]1j, k). Thus, the increased rates of
   breakthrough infections in patients with B-cell malignancies were not
   attributed to differences in antibody responses, suggesting that
   antibody titres alone cannot be used to recommend the need for booster
   vaccinations to protect against COVID-19.

   This study addresses the clinical characteristics of patients with
   cancer who are at higher risk of breakthrough infections using a
   comprehensive analysis of tumour types, cancer treatment, gene
   expression and serology data. While previous meta-analysis indicated
   that patients undergoing chemotherapy, targeted therapy and steroid
   usage had reduced seroconversion rates^[127]13, we did not observe
   differences in breakthrough infection rates and antibody production in
   our study, and these variabilities may arise due to differences in the
   types of vaccines used, host genetics, environmental factors or the
   sampling time point where serology was performed. Instead, we
   identified that tumour type, especially patients with B-cell
   malignancies, had a bigger influence on susceptibility to breakthrough
   infections, likely attributed to immunosuppression generated by B-cell
   lymphodepletion or aplasia. However, in our study, the increased
   breakthrough infections cannot be fully explained by compromised
   antibody responses to SARS-CoV-2, as these patients can produce
   detectable cross-reactive neutralising antibodies after mRNA
   vaccination. It is thus plausible that cellular immunity may be more
   important for long-term protection, especially against variants of
   concern and when antibody titre wanes over time^[128]14–[129]17.

   The findings that downregulation of CD24, CDK14 and PLEKHG1 were the
   leading-edge genes associated with both haematological malignancies and
   increased breakthrough infections highlighted the involvement of B
   cells in vaccine-mediated protection against SARS-CoV-2. That CD24,
   CDK14 and PLEKHG1 are more highly expressed in naïve and memory B cells
   may suggest that the responsiveness to SARS-CoV-2 may be implicated due
   to reduced availability and function of naïve and memory B cells. CD24
   is involved in B-cell development and the expression of CD24 declines
   as B cells mature^[130]18. CDK14 regulates cell cycle regulation,
   especially mitotic progression, which could be relevant for B-cell
   proliferation and expansion^[131]19. The Rho guanine nucleotide
   exchange factor PLEKHG1 regulates cell motility and morphology^[132]20.
   Collectively, these findings raise an interesting possibility that
   changes in B-cell frequencies, development and function in patients
   with B-cell malignancies can lead to reduced T cell responses and
   increased breakthrough infections. This is supported by mouse models
   demonstrating that T cell vaccine responses depend on B cells^[133]21.
   Future studies using single-cell RNAseq and deep immunophenotyping of
   naïve and memory B cells in patients with B-cell malignancies will shed
   deeper insights into the molecular underpinnings involved.

   In conclusion, our data show that the tumour type, particularly
   patients with B-cell malignancies, are at greater risk of breakthrough
   infections. These patients may require more frequent booster
   immunisations to generate an enhanced immune response, providing immune
   protection. In addition, antiviral approaches may be required for these
   patients since they may be more susceptible to symptomatic infection.
   Extended studies investigating the availability and function of
   SARS-CoV-2-specific memory B and T cells in patients with B-cell
   malignancies will be required to learn more about their role in
   protection against SARS-CoV-2 symptomatic infections.

   We acknowledge the limitations of our study. Firstly, as our study
   relied on reported SARS-CoV-2 infections to the Ministry of Health, it
   did not differentiate between symptomatic and asymptomatic infections;
   however, as there was no proactive surveillance performed for
   asymptomatic individuals during the period of the study, many of the
   infections captured in our study can be expected to be symptomatic
   SARS-CoV-2 infections. Second, we acknowledge the possibility of other
   confounders such as time spent in hospital and outdoors, healthcare
   seeking behaviour and the habits of using Personal Protective
   Equipment. Third, while we do not detect significant differences in
   neutralising antibody production between patients with haematological
   malignancies and those with solid tumours in our study, we do not
   exclude the possibility that these differences may become more apparent
   with bigger sample sizes, as indicated by a previous report showing
   that B-cell chronic lymphocytic leukaemia patients had suboptimal
   antibody responses^[134]22. Finally, our study suggested that impaired
   cellular immunity is a likely explanation for the differences in
   breakthrough infection rates in patients with haematological
   malignancies, so future studies that thoroughly investigate effects on
   T cell frequencies, activation and functions in these patients will be
   needed to confirm their role in vaccine-mediated protection.

Methods

Clinical cohort

   The multicenter cohort study was conducted by Singapore’s Ministry of
   Health, which included all patients who were on follow-up in the public
   healthcare system for diagnosis of cancer across the National Cancer
   Centre Singapore, National University Cancer Institute and Tan Tock
   Seng Hospital. Patients received the mRNA-based vaccines BNT162b2 and
   mRNA-1273. More details related to the study design have been
   previously described^[135]8.

   Controls were separately identified for the Delta (15 September 2021 to
   20 December 2021) and Omicron periods (6 January 2022 to 30 June 2023)
   for each patient with cancer through a 1:5 matching based on age, sex,
   race, and socioeconomic status (approximated using flat type). It is
   possible for an individual who was identified as a control during the
   Delta period to also serve as a control during the Omicron period,
   which is why the number of controls is not exactly five times the
   number of patients with cancer. People who took non-mRNA vaccines are
   not included in the numbers.

Cumulative incidence rates

   A competing risks regression model using Stata (version 18), adjusting
   for race, sex, socioeconomic status (approximated using flat type), and
   age categories, was conducted to estimate the cumulative incidence of
   SARS-CoV-2 infections over time. The dataset, structured on a
   time-to-event basis, treated SARS-CoV-2 infections as the primary event
   and non-COVID-19 related deaths as the competing event. Based on the
   regression results, cumulative incidence function (CIF) graphs for each
   group were generated to visualise the cumulative probability of
   infection from 15 September 2021 to 30 June 2023 and then superimposed
   into one chart.

Oncovax clinical study

   A subset of 216 patients with cancer were recruited prospectively from
   two hospitals of the National University Cancer Institute, Singapore:
   National University Hospital and Ng Teng Fong General Hospital,
   Singapore, between July 2021 and March 2022. All patients had a
   personal history of malignancy, were ≥21 years old, and were deemed by
   their primary physician to be suitable to receive SARS-CoV-2
   vaccination. Patients on active treatment were defined as those on
   systemic chemotherapy, tyrosine kinase inhibition or immunotherapy.
   Cancer survivors include those who are at least 3–5 months
   post-adjuvant chemotherapy on surveillance, those on adjuvant hormone
   therapy, or those on radiotherapy alone. Clinicopathological data
   (including age at diagnosis, sex, type of cancer, and anti-neoplastic
   treatment at enrolment) were collected and de-identified. All patients
   received mRNA SARS-CoV-2 vaccination with either the Pfizer BNT162b2
   vaccine or the Moderna mRNA-1273 vaccine. As this was an observational
   study, primary vaccination and booster doses of vaccine were given at
   the discretion of patients and primary physicians. Subjects were
   matched with administrative data on SARS-CoV-2 vaccinations and
   infections reported to the Ministry of Health for the purposes of
   monitoring disease transmission and vaccination uptake under the
   Infectious Disease Act. While symptomaticity of SARS-CoV-2 infections
   was not available from this database, as the national directive during
   the time of the study was to perform diagnostic testing for symptomatic
   individuals, the vast majority of these cases would be symptomatic
   infections. Administrative data on vaccination included the date and
   brand of vaccinations, while data on SARS-CoV-2 infections include the
   notification date of infection, and whether cases required treatment in
   hospital, Supplemental oxygen, intensive care unit admission and/or
   resulted in death. Blood was taken immediately before the first vaccine
   dose, 3–8 weeks after the first dose, 3 months after dose 1 or just
   before the third dose, and at 6- and 12-month post-dose 1.

   The National Healthcare Group Domain Specific Review Board (DSRB)
   provided ethical approval for this study (reference number 2021/00523).
   All individuals enrolled in this study provided written informed
   consent approved by DSRB and in compliance with the Declaration of
   Helsinki principles.

Whole blood processing and bulk RNA sequencing

   Whole blood samples were received on dry ice and stored at −80 °C until
   processing. Before RNA extraction, samples were incubated at room
   temperature (RT) for 2 h. RNA was isolated using the PAXgene Blood RNA
   Kit (Qiagen, 762174) following the manufacturer’s protocol. RNA
   integrity was assessed using an Agilent Bioanalyzer RNA 6000 Nano chip
   (Agilent, 5067-1511). For library preparation, 300 ng of RNA underwent
   globin and rRNA depletion with the NEBNext Globin & rRNA Depletion Kit
   (NEB, E7755X), followed by library construction using the NEBNext Ultra
   II RNA Library Prep Kit (NEB, E7775L). The library size and
   concentration were verified via HT DNA Extended Range LabChip (Perkin
   Elmer, 760517) and pooled to equal concentrations before sequencing on
   the Illumina NovaSeq 6000. Raw sequencing results were processed and
   the raw counts were generated using the standard nf-core RNASeq
   pipeline version 3.6.

Bioinformatics analysis

   73 samples at the pre-dose 3 time point (P7) were included in
   bulk-RNAseq analysis. Three samples were excluded as one patient had an
   additional vaccine dose before P7 and two patients did not come for the
   P7 visit. Two pair-wise comparisons were made: One comparing the
   differences between cancer types—haematological malignancies versus
   solid organ tumours. The other comparison involved a comparison between
   patients with and without breakthrough infection. Raw counts were
   filtered for genes with at least 10 reads for a minimum of 16 samples,
   and 10 reads for a minimum of 36 samples in cancer type and COVID-19
   breakthrough status, respectively, resulting in 15577 and 14497 genes.
   Samples were normalised through DESeq2’s DESeq function
   (v1.42.0)^[136]23, where each sample’s size factors and gene-wise
   dispersions were estimated and subsequently fitted onto a negative
   binomial distribution model. The Wald statistical test was conducted in
   both comparisons for downstream differential expression analysis. The
   lfcShrink function was used to shrink the fold-changes to prevent false
   positives arising from large fold-differences in low read counts. For
   the volcano plot, significantly different expression levels at
   pre-vaccination time-points were determined based on fold-change > 1.5
   and FDR-adjusted p value < 0.05, using the Benjamini-Hochberg Step-Up
   FDR-controlling Procedure.

Pathway enrichment analysis

   GSEA was performed with clusterProfiler (v4.10.0)^[137]24 using
   fold-change values as the input. Blood transcription module
   (BTM)^[138]11 database, curated by Li et al., was used to understand
   the immune cell types involved in the comparisons. The top 10 positive
   and negative significantly enriched pathways (adjusted p value < 0.05)
   in the comparisons were reported. Venn diagrams were plotted with
   ggvenn (v0.1.10) to determine pathways that were commonly enriched in
   haematological malignancies and with breakthrough infections. To
   determine the gene expression levels of CD24, CDK14 and PLEKHG1 in the
   different immune cells, we leveraged on the Human Protein Atlas
   single-cell sequencing derived from the Human Protein Atlas and Monaco
   datasets. The weblinks with the information are as follows:

   [139]https://www.proteinatlas.org/ENSG00000272398-CD24/immune+cell[140]
   https://www.proteinatlas.org/ENSG00000058091-CDK14/immune+cell[141]http
   s://www.proteinatlas.org/ENSG00000120278-PLEKHG1/immune+cell.

Spike (S)-binding IgG quantification

   Anti-S IgG was quantified using enzyme-linked immunosorbent assay
   (ELISA), as previously described^[142]10. Briefly, two high-binding
   96-well ELISA microplates (Greiner) were coated with 1 μg/mL Wuhan-Hu-1
   S hexapro protein diluted in PBS and incubated at RT for 45 min. Plates
   were washed with PBS-T (0.05% Tween-20) and incubated with blocking
   buffer (PBS with 3% BSA) at RT for 1 h. Plasma was diluted 200× and
   5000× in blocking buffer, while a standard antibody (anti-SARS-CoV-2 S
   RBD Neutralising Antibody, Acrobiosystems) was serially diluted 10× for
   the standard curve. Blocked plates were washed and then incubated with
   diluted plasma and antibody standard in duplicates at RT for 1 h.
   Plates were washed and then incubated with HRP-IgG secondary antibody
   (Life Technologies) at 10,000× dilution at RT for 1 h. Lastly, plates
   were washed and then detection reagent 3,30,5,50-Tetramethylbenzidine
   (TMB) (Thermofisher) was added. Reaction was quenched with 1 M
   Sulphuric acid/phosphoric acid. Sample optical density was measured
   with a spectrophotometer at 450 nm, and concentrations in U/ml were
   interpolated from the standard curve using GraphPad Prism 10.0.2.

Neutralising antibody titres quantification

   Neutralising antibodies against VOCs were measured with a pseudotyped
   virus neutralisation assay (pVNT), as previously described^[143]17.
   Briefly, human lung carcinoma epithelial (A549, ATCC CRM CCL-185) cells
   were grown and maintained in RPMI 1640 supplemented with 10% FBS. The
   human ACE2 gene in the pFUGW vector was introduced into A549 cells by
   lentivirus transduction1 and maintained in RPMI 1640 supplemented with
   10% FBS and 15 µg/ml of blasticidin. Human embryo kidney (HEK293T, ATCC
   CRL-3216) cells were grown and maintained in DMEM supplemented with 10%
   FBS. SARS-CoV-2 2 parental (Wuhan-Hu-1), Beta, Delta, Omicron BA.2,
   Omicron XBB.1.16 and JN.1 full-length spike pseudotyped viruses were
   produced by transfecting 20 µg of pCAGGS spike plasmid into 5 million
   HEK293T cells using FuGENE 6 (Promega). At 24 h post-transfection, the
   transfected cells were infected with VSV∆G luc seed virus at an MOI of
   5 for 2 h. After two PBS washes, infected cells were replenished with
   DMEM 10% FBS supplemented with 1:5000 diluted anti-VSV-G mAb (Clone
   8G5F11, Kerafast). Upon 80% cytopathic effect, pseudotyped viruses were
   harvested by centrifugation at 2000 × g for 5 min. Pseudoviruses (~3
   million RLU) were pre-incubated with four-fold serial diluted test
   serum in a final volume of 50 μl for 1 h at 37 °C, followed by
   infection of A549-ACE2 cells. At 20–24 h post-infection, an equal
   volume of ONE-Glo luciferase substrate (Promega) was added, and the
   luminescence signal was measured using the Cytation 5 microplate reader
   (BioTek) with Gen5 software version 3.10. The 50% neutralising titre
   (NT50) was interpolated using GraphPad Prism 10.0.2.

Statistical analysis

   Statistical analyses were performed with Prism 10.3.1 software,
   including unpaired t-test, chi-squared test and one-way ANOVA. RNAseq
   and pathway enrichment analysis was performed with R, using the DESeq
   and clusterProfiler packages. Figure [144]1a was created in BioRender,
   while all other figures were created using R; every element of all
   figures were created by the authors of this manuscript.

Supplementary information

   [145]Supplementary File^ (584.6KB, pdf)

Acknowledgements