One hundred thirty-four germ line PU.1 variants and the agammaglobulinemic patients carrying them Free

PU.1 is a master hematopoietic transcriptional regulator that is critical for guiding multipotent progenitor cells through early stages of leukopoesis.^1-5 Complete PU.1 deficiency is embryonically lethal in mice because of myeloid and lymphoid developmental arrest.^2,6 Depending on the mouse model, more subtle decreases in PU.1 expression (20% to 65% of wild type) either induce acute myeloid leukemia (AML)^7-9 or selectively block early B-cell^6,10 and dendritic-cell (DC) development.^11,12 

Several lines of evidence indicate humans are also intolerant of altered PU.1 dose. First, 0 nonsense or frameshifting loss-of-function (LOF) PU.1 variants are listed in the Genome Aggregation Database (gnomAD) version 2.1.1, a large human reference comprising >125 000 whole exome and 15 000 whole genome sequences.¹³ The omission of heterozygous LOF PU.1 variants from gnomAD suggests considerable evolutionary constraint and high a degree of loss intolerance that is consistent with haploinsufficiency.^14,15 Indeed, in 2021 we reported the first 6 patients (5 males, 1 female) with PU.1-mutated agammaglobulinemia (PU.MA), a congenital haploinsufficiency disease that arrests B-cell development between the pro- and pre-B stages.¹⁶ Each patient harbored a novel, LOF PU.1 variant that either occurred de novo or was unphased. Principal disease features included susceptibilities to sinopulmonary and enteroviral infections secondary to stark deficiencies of circulating antibodies, B cells, and conventional DCs (cDCs). Although PU.1 is often somatically mutated or epigenetically repressed in human AML,^8,17 cancer was not initially observed in patients with PU.MA. Because of the small number of cases reported to date, questions related to disease penetrance, sex bias, cancer predisposition, and noninfectious complications persist.

We amassed a collection of 56 novel, protein-altering PU.1 variants from either patients with “mutation-negative” agammaglobulinemia, institutional clinical databases, or previously published reports.^16,18,19 To these, all rare (minor allelic frequencies [MAF] of <0.1%), nonsynonymous PU.1 variants from gnomAD 2.1.1 were added. The full complement of 134 PU.1 variants (supplemental Table 1, available on the Blood website) were subjected to a series of functional analyses (supplemental Figure 1) identifying 25 LOF variants. Herein, we share a comprehensive, functionally annotated atlas of human PU.1 variation that provides insights into the molecular determinates of PU.1 stability and transcriptional efficiency. We also describe the phenotypic, epidemiologic, and therapeutic details of a greatly enlarged PU.MA cohort now numbering 22 patients.

Blood donors provided written informed consent via research protocols approved by the institutional review boards of Children’s Hospital of Philadelphia, University of Pennsylvania, University of Virginia, Rockefeller University, Colorado Children’s Hospital, Children’s Hospital of Los Angeles, Indiana University, Belgrade Mother and Child Institute, Soroka University Medical Center, University of Helsinki, HUS Helsinki University Hospital, IRCCS Ospedale San Raffaele, Children's Hospital of Fudan University, Meyer Children's Hospital IRCCS, Tehran University of Medical Sciences, Monash Health, and Lausanne University Hospital.

Variants in the SPI1 gene (which encodes PU.1) were collected from multiple sources and annotated to the Matched Annotation from the National Center for Biotechnology Information, European Molecular Biology Laboratory-European Bioinformatics Institute (MANE) transcript NM_003120.3/ENST00000378538.8. Corresponding gene products were annotated to MANE protein NP_003111.2/ ENSP00000367799.4, a 270–amino acid protein. SPI1 gene variants included in the study were (1) protein altering, and (2) either rare (MAF < 0.1%) or novel (supplemental Table 1; supplemental Figure 2A). A total of 71 rare SPI1 variants were collected from gnomAD version 2.1.1, a source lacking paired clinical information.¹³ An additional 42 novel variants were identified in 1 of the following private patient databases: (1) Penn Medicine BioBank,²⁰ (2) Children’s Hospital of Philadelphia Center for Applied Genomics, or (3) Rockefeller University St. Giles Laboratory of Human Genetics of Infectious Diseases. Paired clinical and laboratory data were available on some but not all individuals from institutional databases. Fourteen SPI1 variants were identified using clinical whole-exome or whole-genome sequencing to analyze patients with agammaglobulinemia lacking pathogenic variants in other International Union of Immunological Societies genes (ie, “mutation-negative” agammaglobulinemia).²¹ To make the resource maximally comprehensive, 7 previously published, disease-associated PU.1 variants, were also included.^16,18 

For single SPI1 plasmid transfections, mutated or unmuted pCMV-iRFP-2A-PU.1 (100 ng) were transfected into HEK293 PU.1 reporter lines as previously described.¹⁶ After 48 hours, cells were stained for viability (LIVE/DEAD kit, Invitrogen) and frequencies of enhanced green fluorescent protein (EGFP) expressing, live, near-infrared fluorescent protein (iRFP)⁺ cells were identified by flow cytometry. To determine interference, mutated pCMV-iRFP-2A-PU.1 (50 ng) and unmutated porcine cyotomegalovirus (pCMV)-mCherry-2A-PU.1 (50 ng) plasmids were cotransfected with the reporter plasmid. EGFP expression of mCherry⁺iRFP⁺ double-positive cells was captured at 48 hours. All reporter experiments were performed in triplicate and included same day, unmutated PU.1 controls.

Data were analyzed with GraphPad Prism (version 10.3.0) using either Student t tests, Fisher exact tests, or Anderson-Darling tests for normal data distribution.

Additional methods (flow cytometry, SPI1 plasmid construction, and immunoblots) are described in the supplemental Methods.

The 134 identified protein-altering PU.1 variants spanned the near entirety of the 270–amino acid molecule from residues 10 to 268. There were 112 single amino acid substitutions, 6 in-frame deletions, 9 frameshifting insertions/deletions, 6 premature stop codons, and a 439-kb chromosome 11 microdeletion encompassing the entirety of SPI1 (supplemental Table 1). One frameshifting variant, p.D48Afs∗82, was the consequence of 4 separate genetic changes in cis (c.159C>G, c.157T>G, c.147-155delTTACTGGGA, and c.143insCCCCC). This complex genetic lesion was confirmed by next-generation and Sanger sequencing. Nucleotide-specific combined annotation and dependent depletion (CADD) scores could be assigned to 132 PU.1 variants and ranged from 1.1, inconsistent with altered function, to 42, among the top 0.01% most deleterious nucleotide changes possible in the human genome (supplemental Table 1; supplemental Figure 2A-C).^22,23 Similarly, every PU.1 single amino acid substitution was assigned an AlphaMissense pathogenicity score²⁴ ranging from 0.058 (likely benign), to the maximum score of 1 (likely pathogenic; supplemental Table 1; supplemental Figure 2B,D). All studied variants were at least rare (MAF < 0.1%) and 56 were considered “novel” because they were absent from gnomAD version 2.1.1.

To determine protein stability, we separately overexpressed each PU.1 variant in HEK293 lines, size-separated by electrophoresis, and detected with a monoclonal N-terminal PU.1 detection antibody. On immunoblots, 114 PU.1 protein variants appeared as bands of similar size and normalized densities to unmutated PU.1 (supplemental Figure 3A; supplemental Table 1). Overall, 15 mutant PU.1 proteins were poorly detected by the monoclonal antibody with normalized band densities of <0.2. Poorly detected variants were primarily encoded by destabilizing indels or premature stop codons, but 2 (R30S and R30H) were single amino acid substitutions for the same acidic domain residue. Four PU.1 proteins, all single amino acid substitutions in the guanine-rich (GR) (T71M) or E26 transformation-specific (ETS) domains (L179P, L180F, and V241G) were detectable only as light bands with normalized densities between 0.2 and 0.6. All PU.1 variants either poorly detected or detected as light bands with the monoclonal antibody were separately probed with a polyclonal anti-PU.1 antibody (Figure 1A). Results between experiments were similar except for R30S and R30H variants (supplemental Figure 3B). R30S and R30H were only detected by the polyclonal antibody, albeit as semistable, light bands. These data indicate that Arg30 is part of the binding epitope for our anti-PU.1 monoclonal antibody and may also be an important contributor to PU.1 stability. Hence, variants at key positions spanning PU.1’s functional domains can be molecularly destabilizing.

PU.1’s ETS domain recognizes a purine-rich λB promoter/enhancer nucleotide sequence in genomic DNA.²⁵ To measure the transcriptional activity of all 114 stable and 6 semistable PU.1 variant proteins, each was separately overexpressed in a HEK293 PU.1 reporter line containing a pentameric λB nucleotide sequence positioned upstream of a minimal promoter and an EGFP gene transcriptional start site. Based upon frequencies of EGFP⁺ cells, each PU.1 variant was assigned a transcriptional index (TI) relative to wild-type PU.1. Reporter cells expressing no PU.1 (TI = 0) or wild-type PU.1 (TI = 1) served as negative and positive controls, respectively. To determine a lower limit for transcriptional activity below which a variant would be considered LOF, normally distributed gnomAD variant TI values were used (supplemental Figure 3C). Based upon the mean TI value (1.03) and standard deviation (0.11), the LOF TI boundary was conservatively set at 0.7, 3 standard deviations below the mean.

Based upon a 0.7 TI threshold, 11 stable or semistable PU.1 variants were considered LOF (Figure 1B-C), and all but 1 of these were located in the ETS domain.²⁶ The ETS, which spans amino acids 169 through 259, is encoded by exon 5 of the SPI1 gene. Consistent with regional constraint, there were far fewer gnomAD version 2.1.1 missense PU.1 variants in its central portion of the ETS (between Asp195 and Gly236) than expected by chance (2 observed vs 25.9 expected; missense constraint value = 0.08; P < 6.6 × 10⁻⁶ by the Fishers exact test).¹⁵ Furthermore, random single-nucleotide substitutions along the length of exon 5 (mean, 28.9; range, 6.5-49) yield, on average, higher CADD scores than counterpart replacements in other SPI1 exons (Figure 2D). All PU.1 variants demonstrating TIs of <0.7 (ie, LOF) in reporter assays possessed CADD scores of >25, but similar scores were also assigned to 22 other PU.1 variants with normal transcriptional activity (supplemental Figure 2C). AlphaMissense pathogenicity scores proved similarly accurate (supplemental Figure 2B,D), highlighting the high sensitivity and low specificity of both in silico tools.

Several stable LOF PU.1 ETS variants altered 1 of the 2 specific structures directly contacting genomic DNA, namely the “wing” between the β3 and β4 sheets (K245del),²⁷ or the α3 helix (A231Lfs∗15, L232Afs∗53 N234Tfs∗11). Notably, these frameshifting ETS variants were not only predicted to alter DNA binding but also to obliterate their downstream nuclear localization sequence (NLS). NLS loss caused cytoplasmic entrapment of frameshifting ETS variants in fractionated PU.1 immunoblots^16,27,28 (supplemental Figure 5). Deletion of the positively charged Lys245 residue (K245del) in the NLS resulted in similar yet less-pronounced effects. Missense variants outside the NLS and charge-neutral variants within it (eg, H211P) did not alter nuclear localization.¹⁶ Other LOF ETS missense variants deformed key regions undergirding DNA-contacting structures like the α1 helix (F175L, L180F), α2 helix (H211P), β1 sheet (W191R), and β3 sheet (V241G). R116W, a gnomAD variant in the PEST domain, was the only stable LOF PU.1 variant lying outside the ETS. Because it possessed the highest mean AF (0.00116%) and TI value (0.64) among LOF variants (supplemental Figure 2A,C), R116W may represent an expected tail value from the normal distribution (supplemental Figure 3C), not a disease-causing mutation. The semistable T71M, R30S, and R30H variants performed as well as unmutated PU.1 in reporter lines, suggesting normal transcriptional activity despite quantitatively less protein.

To determine whether any of the 11 stably expressed LOF PU.1 mutants interfered with unmutated PU.1, plasmids encoding mutant and wild-type proteins were cotransfected into the PU.1 reporter line to model balanced PU.1 expression within patient cells. No mutant PU.1 proteins impeded the ability of unmutated PU.1 to drive EGFP expression (supplemental Figure 4). Hence, all described LOF PU.1 variants convey effects via haploinsufficiency.

Of the 25 identified LOF PU.1 variants, 1 was exclusive to gnomAD and therefore disease associations could not be investigated. Four LOF variants were identified in institutional databases, but all lacked sufficient clinical annotation to be included in this study. The remaining 20 LOF variants were carried by 1 of 22 newly identified or previously published patients with agammaglobulinemia.^16,18 In addition to their PU.1 variant, 3 newly identified patients (H.II.1, Q.II1, and U.II.1) carried potentially confounding heterozygous variants of unknown significance in other International Union of Immunological Societies inborn error of immunity genes,²¹ specifically IKZF2, IGHM, and WDR1 (Table 1). Although likely affected, the deceased mother of H.II.1¹⁹ and the deceased sister of L.II.1 were not included in our patient cohort because neither could be genotyped. To maximize our variant data set, published and unpublished cases were combined to create an enlarged cohort of 22 patients with PU.MA from 21 unrelated families (Table 1; Figure 2), hailing from 13 countries in North America, Central America, Europe, Asia, and Oceania (supplemental Figure 6).

The original description of PU.MA included 5 affected males and 1 female. The enlarged 22-member cohort remained mostly (77%) male, surprising for an autosomal disorder (Table 1). The mean age of symptom onset and agammaglobulinemia diagnosis for patients with PU.MA was 4.7 and 8.5 years, respectively. Both values were significantly greater than those previously reported in a large cohort of US patients with X-linked agammaglobulinemia (XLA)²⁹ (2 and 3.3 years; P < .01 and P < .0001, respectively; t test; Figure 3). Although 10 patients with PU.MA were diagnosed in the first 2 years of life, a cluster of 12 patients received diagnoses later in childhood (n = 5), adolescence (n = 3), young adulthood (n = 3), and even middle age (n = 1). In 5 such patients (F.II.1, H.II.1, M.II.1, N.II.1, and S.II.1) symptoms occurred before age 2 years but diagnosis was delayed. The 7 remaining late diagnoses developed infectious symptoms later in life, sometimes well into adulthood (range, aged 4-33 years). Several lines of evidence such as preceding hypogammaglobulinemia (C.II.1 and O.II.1), lingering vaccine titers at diagnosis (C.II.1 and G.II.1), residual circulating B cells (N.II.1 and Q.II.1), and an apparent lack of early childhood sinopulmonary infections (all), indicated that there could have been prolonged periods of immune competence before the onset of stark B-cell aplasia.

The 22 patients with PU.MA belonged to 1 of 21 unrelated families (Figure 2A). Within families, LOF PU.1 variants were phased by genotyping all willing first- and second-degree relations. In this manner, 10 definitive instances of vertical inheritance were identified. PU.1 variants occurred de novo in 6 patients from families without immune deficiency histories. The mode(s) of inheritance for remaining PU.1 variants could not be ascertained because of the unavailability of genomic DNA from both parents.

Of the 11 relatives harboring heterozygous LOF PU.1 variants, 6 were clinically well with laboratory evaluations documenting serum antibody and circulating B-cell concentrations within age-adjusted normal ranges. Five additional relatives with LOF PU.1 variants displayed either decreased B cells (T.I.1 and T.II.1), mild hypogammaglobulinemia (K.I.1 and S.II.2), or selective immunoglobulin A (IgA) deficiency (O.III.1). Hence, among 33 patients with a verified or presumed LOF PU.1 variant, 22 were agammaglobulinemic, 5 exhibited milder humoral immune deficiencies, and 6 were apparently unaffected, indicating a disease penetrance of 81.8% with incomplete expressivity.

Peripheral blood cells from patients with PU.MA were broadly analyzed via flow cytometry. Compared with age-matched healthy controls, patients possessed very low mean circulating concentrations of B cells (599 vs 12.15 cells per μL; P < .0001) and cDCs (31.5 vs 5.2 cells per μL; P < .0001; Figure 4A-B). Notably, circulating plasmacytoid DC (pDC) concentrations were also significantly lower in patients than healthy donors (9.2 vs 3.4 cells per μL; P < .001), an immunophenotypic feature not previously appreciated. Consistent differences in natural killer cells, CD4 cells, CD8 cells, or monocyte subsets were not appreciated (supplemental Table 2). Other than the transient reductions associated with pre–immunoglobulin replacement therapy (IRT) infections, neutropenia was not observed.

The chief symptomatic feature of patients with PU.MA, such as all patients with agammaglobulinemia, was recurrent bacterial sinopulmonary infections including otitis, sinusitis, and/or pneumonia, which were universally experienced by patients before initiation of IRT (Table 1). Radiographic evidence of bronchiectasis from recurrent respiratory infections was common. Hematogenous spread of streptococcal, Haemophilus, and staphylococcal septicemia and osteomyelitis, arthritis, and meningitis were each reported. Staphylococcal skin infections and periorbital cellulitis also occurred. One patient was treated for Salmonella typhi enterocolitis. Notably, bacille Calmette-Guérin (BCG) vaccination was documented in at least 10 patients (Table 1); none experienced regional lymphadenitis or systemic BCG-osis.

Viral infections, most notably enteroviruses, affected multiple patients. One (D.II.1) of 5 patients receiving live-attenuated oral polio vaccine developed iatrogenic paralytic poliomyelitis. Patient A.II.2 experienced reactive arthritis during a prolonged period of echoviremia. Influenza A virus myocarditis was observed in patient I.II.1 before the initiation of IRT. Nearly all patients experienced severe acute respiratory syndrome coronavirus 2 infections while on IRT, and clinical courses were symptomatically mild.

Although not previously reported, gastrointestinal disease was common in newly identified patients with PU.MA. Eight patients (36%) experienced an inflammatory bowel disorder, either as frank entercolitis, a sprue-like disease, a perirectal abscess, or autoimmune gastritis. There were 2 patients with liver granulomas and 1 instance of chronic liver disease. Inflammatory disease outside of the gastrointestinal tract included juvenile idiopathic arthritis (n = 2); type 1 diabetes mellitus (n = 2); and single cases of immune thrombocytopenia, psoriasis, hidradenitis suppurativa, and autoimmune lipodystrophy. In addition to the single instance of infectious paralytic poliomyelitis, 5 other patients experienced neurocognitive disorders. Three were given developmental, behavioral, and/or autism spectrum diagnoses in the first years of life. One patient (O.II.1) was diagnosed with Parkinson disease without dementia in her sixth decade. Patient L.II.1 developed rapidly progressive and ultimately, fatal neurocognitive decline at age 38 years.¹⁸ 

Despite the association of somatic SPI1 variants with human myeloid leukemias,¹⁷ none of the 22 patients with PU.MA in our cohort developed blood cancers despite a combined 514 patient-years. Similarly, neither leukemia nor lymphoma was observed in any of the 6 healthy or 5 antibody-deficient relatives carrying germ line LOF PU.1 variants. One nonhematopoietic malignancy, a meningioma, was identified in patient O.II.1 in her seventh decade of life. Although 12 SPI1 variants included in this study also appear in the Catalogue Of Somatic Mutations In Cancer database (supplemental Table 3), all encoded stable proteins that proved transcriptionally active (TI range, 0.88-1.12) in PU.1 reporter assays.

All patients with PU.MA were IRT dependent (Table 1). One of 3 premature patient deaths resulted from an IRT-preventable infection stemming from noncompliance (D.II.1). Immune modulating treatments successfully used for inflammatory bowel disorder included mesalamine, mycophenolate, and tumor necrosis factor inhibition (adalimumab). A series of therapies (methotrexate, hydroxychloroquine, etanercept, tocilizumab, abatacept, and canakinumab) were trialed with limited success in 1 patient with recalcitrant juvenile idiopathic arthritis (S.II.1).

Two patients underwent hematopoietic stem cell transplantation (HSCT). The first, patient A.II.2 received fully matched stem cells from a PU.1-sufficient, matched sibling donor in his second year of life. Five years after transplantation, he remains IRT dependent because of undetectable serum immunoglobulin (IgA) and IgM despite naïve B-cell reconstitution and >99% donor peripheral blood engraftment in B-cell, T-cell, and myeloid cell compartments. The second patient (L.II.1) who received HSCT received stem cells from her healthy sister (L.II.2) at the age of 16 but failed to engraft.¹⁸ Unbeknown at the time of HSCT, the donor harbored the same PU.1 variant as the recipient.

Herein, we functionally assessed the stability and transcriptional efficiency of 134 rare and novel PU.1 variants. Several LOF variants we identified localized to contact points between the ETS and double-stranded genomic DNA or potentially deformed protein structures undergirding this DNA-binding apparatus.³⁰ Other ETS domain mutations interfered with nuclear importation. Several destabilizing missense variants localized to intrinsically disordered domains of PU.1 (acidic, GR, and PEST), which are structurally uncharacterized. Unlike missense variants within the ETS, which overwhelmingly altered PU.1’s transcriptional efficiency, counterparts outside it mostly did not, even when they decreased protein stability. Regardless of location, all tested LOF variants conveyed effects via haploinsufficiency.

Heterozygous LOF PU.1 variants were carried by 1 of 22 patients with PU.MA and 11 family members. Of relatives, 6 were apparently healthy and 5 displayed deficiencies of ≥1 antibody isotypes. Based on these data, we estimate the disease penetrance of LOF PU.1 variants to be 81.8% with variable expressivity. Although it was the only means available, our use of familial aggregation data to estimate penetrance is a limitation of our study. Family aggregation is susceptible to ascertainment bias-associated inflation,^31,32 especially in small pedigrees enriched in de novo and unphased mutations such as those presented here. An opposing force confounding penetrance estimates in the opposite direction is late-onset disease; PU.1 variant carriers now counted as unaffected could later develop antibody deficiency or frank agammaglobulinemia. The delayed and progressive B-cell loss seen in some patients with PU.MA evoke IKAROS or HELIOS haploinsufficiencies.^33,34 All 3 disorders, which are caused by decreased doses of key hematopoietic transcription factors, suggest that rigid adult B-cell gene networks may be more flexible during childhood. Surprisingly, in PU.MA there is not an obvious correlation between PU.1 variants and disease onset age. For instance, many patients with PU.MA first presenting as adults possessed pure LOF variants (E34X and Y121X) that completely abolished protein expression. Similarly, the L180F variant passed from parent to child in “Family O” caused disease at vastly different ages, 34 years and 1 year, respectively.

The defining immunophenotypic features of PU.MA are stark B-cell, cDC, and pDC deficiencies, but infectious susceptibilities of patients (primarily sinopulmonary and enteroviral infections) are very close to other forms of agammaglobulinemia and not to monogenic DC defects. Historically, the hallmark of human cDC deficiency has been severe mycobacterial infections, including BCG-osis, but at least 9 patients in our PU.MA cohort were safely administered live-attenuated BCG vaccine, suggesting intact mycobacterial immunity. Key differences between PU.1 haploinsufficiency and previously described monogenic cDC deficiencies are the other cell types affected. Although PU.MA narrowly alters B-cell and DC development, patients with GATA-binding factor 2 (GATA2),³⁵ interferon regulatory factor 8 (IRF8),³⁶ and signal peptide peptidase-like 2A (SPPL2A)³⁷ deficiency exhibit defects in other cell types critical to cellular immunity (eg, monocytes and/or T cells). Together with patients with feline McDonough sarcoma-related tyrosine kinase 3 ligand (FLT3LG) deficiency,³⁸ who also do not experience BCG infections despite cDC deficiencies, patients with PU.MA support the proposition that DCs are dispensable for human mycobacterial defense. Similarly, patients with pDC deficiency with PU.MA did not experience the same severe viral respiratory infections with severe acute respiratory syndrome coronavirus 2 or seasonal influenza infections than patients with the prototypical pDC defect, IRF7 deficiency.^39,40 This observation suggests that IRF7-mediated type 1 interferon responses in respiratory epithelial cells, not pDCs, may be most critical for immunity to airborne respiratory viruses.⁴¹ 

The noninfectious symptoms of PU.MA closely align with other forms of pure B-cell aplasia such as XLA.⁴² In particular, the high incidence of gastrointestinal disease across the spectrum of congenital agammaglobulinemia, regardless of gene, indicates the importance of secreted antibodies such as IgA to gut/microbiome détente. For less clear reasons, a considerable and similar fraction of patients with PU.MA and XLA also exhibit a spectrum of neurological conditions. One potential explanation is that damage from diagnosed or undiagnosed central nervous system infections underlie a spectrum of acute and chronic neurocognitive dysfunction. Another possibility is that microglia, which highly express both PU.1 and BTK,^43-45 could be altered in these patients. Histological analyses of affected central nervous system tissues may be revelatory, but none are currently unavailable.

Finally, despite extensive literature suggesting that human somatic PU.1 mutations in myeloid precursors or mature B cells promote myeloid leukemia^8,17 or Waldenström macroglobulinemia,⁴⁶ respectively, no blood cancers were observed in the described 22 patients with PU.MA or their 11 relatives carrying germ line LOF PU.1 variants. Although certainly not cancer predisposing to the same degree as germ line SAMD9^47,48 and SAMD9L mutations,^47,49 our cohort remains too small to conclude that inborn LOF PU.1 variants convey 0 malignancy risks. Notably, mouse studies indicate that PU.1 expression losses of ∼80% consistently induce AML^8,9 whereas decreases of 35% do not, except on a mismatch repair-deficient background.⁷ Hence, PU.1 may theoretically pose the greatest leukemic hazard to humans carrying either (1) homozygous PU.1 LOF variants (if not embryonically lethal), (2) dominant interfering PU.1 variants (if these exist), or (3) PU.1 haploinsufficiency combined with ≥1 other cancer risk gene alleles.⁵⁰ In this study, we did identify 1 case of adult-onset meningioma. Although this may represent a potential risk signal for nonhematopoietic cancers in patients with PU.MA, we think that to be unlikely because the meninges do not express PU.1.

There is growing interest in using cellular therapies to treat agammaglobulinemia, especially in the United States, where the high price of life-long IRT renders it cost-ineffective relative to HSCT.⁵¹ The 2 known attempts at treating PU.MA with HSCT resulted in either graft rejection or ongoing IRT dependence. Why did they fail? For patient A.II1 full trilineage engraftment was achieved after transplant, but endogenous antibody production remained undetectable. This unanticipated failure suggests that PU.1-expressing cells that are not easily replaced by HSCT, such as tissue-resident macrophages,⁵² may be important for reconstituting humoral responses. In contrast, graft rejection observed in patient L.II.1 could be explained by a lack of competitive advantage for donor cells over recipient cells because both carried the same heterozygous LOF PU.1 variant. Although a small sample size, these failures represent a cautionary signal to be considered when weighing risks and benefits of lifelong IRT vs HSCT for patients with PU.MA.

The authors thank the patients and their families. They thank the Children’s Hospital of Philadelphia’s Center for applied genomics, the Flow Cytometry Core, and the University of Pennsylvania’s Penn Medicine Biobank. The authors thank Maria Teresa de la Morena, Vivian Hernandez-Trujillo, and the United States Immunodeficiency Network for assistance accessing their published XLA data; and Eira Leinonen for blood sampling. The authors acknowledge Orbicia Riccio, Elisa Calistri, Francesca Quaranta, Martina Cortimiglia, and George McGillivray for assistance with genetic diagnoses.

This work was supported by the National Institutes of Health (NIH), National Institute of Allergy and Infectious Diseases (grants AI14602, AI184976, and AI179680 [N.R.]; and P01AI061093 [J.-L.C.]), NIH, National Heart Lung and Blood Institute (grant HL155178 [G.M.K.P.]), and the NIH, Clinical and Translational Science Award program (grant UL1TR001866 [J.-L.C]). Additional funding provided by the Jeffrey Modell Foundation Translational Research Program (N.R.), Jeffrey Modell Foundation Centre Melbourne (M.C.v.Z. and E.S.J.E.), Sanford Children’s Genomic Medicine Consortium (C.M.D.), the Swiss National Science Foundation (grant 310030_197695 [F.C.]), the Finnish Medical Foundation (J.G.), Emil Aaltonen foundation (J.G.), Academy of Finland (J.G.), Stiftelsen Frimurare Barnhuset I Stockholm (J.G.) and Sigrid Juselius Foundation (J.G.), the Knut and Alice Wallenberg Foundation (Q.P.-H.), the Swedish Research Council (Q.P.-H.), Swedish Cancer Society (Q.P.-H.), the Knut and Alice Wallenberg Foundation (Q.P.-H.), the Sayer family (N.R.), and the Dr. Steven Douglas Memorial Fund (N.R.). The Laboratory of Human Genetics of Infectious Diseases (J.-L.C.) is supported by the Howard Hughes Medical Institute; the “Investissement d’Avenir” program launched by the French Government and implemented by the Agence Nationale de la Recherche (grant ANR-10-IAHU-01); the Integrative Biology of Emerging Infectious Diseases Laboratory of Excellence (grant ANR-10-LABX-62-IBEID), ANR KREM-AIF (grant ANR-21-CE17-0014-03); the St. Giles Foundation; the National Center for Advancing Translational Sciences; the Integrative Biology of Emerging Infectious Diseases Laboratory of Excellence (grant ANR-10-LABX-62-IBEID), ANR KREM-AIF (grant ANR-21-CE17-0014-03); the SCOR Corporate Foundation for Science; William E. Ford, General Atlantic’s chairman and chief executive officer; and Gabriel Caillaux, General Atlantic’s copresident, managing director and head of business in Europe, the Middle East, and Africa and the General Atlantic Foundation; Institut National de la Santé et de la Recherche Médicale; Paris Cité University; and the Imagine Institute.

Contribution: A.V.C.K., C.L.C., L.Y.C., G.M.K.P., and N.R. conceived and designed the study; A.V.C.K., L.Y.C., D.S., E.C.C., B.E.N., E.O., S.Y., and S. Rachimi generated and analyzed data; S.D.R., M.G.L., K.E.S., N.P., E.C., M. Stojanovic, G.P., S.P., J.C., R.M.F., F.C., T.A., K.T., C.M.D., L.M., G.S., A.B., A.N., J.L., K.K., R.D., M. Seppänen, M.V., T.M., J.G., H.H., D.M., K.B., B.B., C.A., S. Ricci, M.P., J.S., Q.P.-H., H.A., S.O., E.S.J.E., M.C.v.Z., and J.-L.C. identified genetic variants, generated clinical info, and provided patient samples; A.V.C.K., L.Y.C., and N.R. interpreted data and wrote the manuscript; and all authors reviewed and approved final manuscript.

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

Correspondence: Neil Romberg, Division of Immunology and Allergy, Children's Hospital of Philadelphia, Department of Pediatrics, Institute for Immunology, Perelman School of Medicine, University of Pennsylvania, 3615 Civic Center Blvd, Philadelphia, PA 19104; email: rombergn@chop.edu.