Targeted high-throughput sequencing for genetic diagnostics of hemophagocytic lymphohistiocytosis

Coverage analysis

A custom targeted resequencing panel was designed to cover 12 genes in which mutations
have been associated with HLH or lymphoproliferative disorders (Table 1). The panel consisted of 355 primer pairs, with an amplicon size range of 125–175 bp,
covering up to 97.3 % of the desired target. Specific primers could not be designed
for 1125 bp, due to repetitive regions. Following sequencing, analyses revealed that
some amplicons repeatedly failed to generate adequate coverage, as determined by a
?10× cut-off of mean coverage across samples (Fig. 1a; Additional file 1). Excluding two amplicons that failed in nearly all samples (mean coverage ?10×),
the effective coverage of our initial target sequences was estimated to be 96.6 %
of the initial regions of interest, with an average coverage per gene over exonic
and splice-site regions of 98 % (Table 1). To ensure clinical efficacy, we calculated the proportion of previously reported
mutations (based on the Human Gene Mutation database (HGMD), accessed June 2015) with
adequate coverage. Overall, 98.6 % of the mutations listed in HGMD were covered by
our design (Table 1).

Fig. 1. Analysis of coverage efficiency and variant filtering strategy. a Heatmap indicating coverage for each individual amplicon (355 amplicons) in each
patient sample. The coverage is categorized in 10×, 10–50× or 50× coverage. Patient
samples and amplicon are shown in columns and rows, respectively. Patient samples
from both the validation and the prospective cohorts are included. Rows are sorted
by position. Columns are sorted by average coverage for each patient. On the right
side the bar plot is shown the number of samples with lower coverage (10× and 10–50×).
The detailed coverage information of all amplicons is reported in Additional file
1. b Flowchart of filtering strategy in the prospective HLH cohort. For each step the
bar chart on the right shows the proportions of different types of variants. MAF minor allele frequency, UTR untranslated region

Assay validation

To validate our gene panel, we sequenced gDNA from 13 patients with previously identified
genetic defects (Table 2). The patients carried a wide spectrum of mutations located in different genes (Table 2). In order to assess the reliability of the method for detection of homozygous exonic
deletions, we also included a patient with a 298-bp homozygous exonic deletion of
STXBP2. We could identify all 18 small genetic aberrations upon read inspection in IGV.
Nonetheless, the variant calling software detected only 17 out of 18 small genetic
aberrations (Table 2). The RAB27A c.148_149delinsC InDel, located in a homopolymer nucleotide stretch, was instead
erroneously called as a synonymous variant (c.148AC; Fig. S1a in Additional file
2). The exonic deletion was easily detected by visual assessment of the coverage over
the amplicons (Fig. S1b in Additional file 2).

Table 2. Disease-causing mutations used in the validation phase

We next sought to estimate the overall sensitivity of the variant calling strategy
by assessing all exonic polymorphisms (n?=?56) previously identified in the 13 control samples (Additional file 3). In total, 74 variants (n?=?18 mutations and n?=?56 polymorphisms) were used for the sensitivity analysis. Out of these, 72 variants
were properly called. The overall sensitivity was 97.3 % (95 % confidence interval
90.7–99.2, Wilson score method).

Prospective cohort of HLH patients

Following validation, we sequenced a cohort of 58 prospectively recruited HLH patients
(Fig. 2a). The median age at diagnosis of HLH was 3 years, ranging from a few days to 70 years
(interquartile range?=?0.4–13.2 years; Table 3). Eight patients were above 18 years of age at diagnosis of HLH. The included patients
were of different ethnic origin, including 43 % from Turkey. Parental consanguinity
was reported in 24 cases. Interestingly, six patients also suffered from albinism
and eight patients had a familial history of HLH or unexplained siblings’ death in
childhood (Table 3). Hyperferritinemia, splenomegaly and hypertriglyceridemia and/or hypofibrinogenemia
were the most common findings in our cohort. Soluble interleukin-2 receptor (sCD25)
was elevated in all seven patients tested (Table 3).

Fig. 2. Clinical, genetic and functional characteristics of the patients included in the prospective
cohort. a Heatmap of clinical and functional features of the implementation cohort in relation
to HLH-2004 diagnostic criteria 2]. The patients are ordered based on age at diagnosis of HLH. A family history refers
to a positive family history for HLH or unexplained siblings’ death in childhood.
b The different molecular diagnoses achieved in the prospective cohort according to
age group at diagnosis of HLH. c NK cell cytotoxic activity, displayed as lytic units at 25 % specific lysis, in healthy
controls and patients from the implementation cohort grouped in diagnosed (n?=?10) and undiagnosed (n?=?13) (significance level *p??0.05, ****p??0.0001). d Intracellular expression of perforin, CD107a, granzyme A and B, and SAP in PBMCs
of HLH patients from the implementation cohort. Patients are grouped by their molecular
diagnosis (FHL2, n?=?4; FHL3-4,GS2,CHS, n?=?9; No diagnosis, n?=?19). The data are expressed as percentage of normalized median fluorescence intensity
(MFI) in comparison with healthy controls. Exocytic activity of CD3
^?
CD56
⁺
NK cells (e) and CD8
⁺
CD57
⁺
T cells (f) was measured as percentage of CD107a
⁺
cells in healthy controls and HLH patients from the implementation cohort. P815 target
cells were used alone and in combination with anti-CD16 antibody for NK cells and
anti-CD3 antibody for CD8 T cells. Exocytic activity of NK cells was also measured
using K562 target cells. Patients are grouped by their molecular diagnosis (FHL2, n?=?4; FHL3-4,GS2,CHS, n?=?9; No diagnosis, n?=?20). The controls used were both local and transport controls

Table 3. Clinical characteristics of HLH patients included in the prospective cohort

Overall, 246 genetic variants were identified in the prospective cohort. Filtering
for variants with a possible impact at the protein level, and with a minor allele
frequency 0.05 in the Exome Aggregation Consortium dataset 35], 71 potentially pathogenic variants were selected for further analysis (Fig. 1b). After manual curation, 19 variants (single-nucleotide variants or small indels),
either in homozygous or compound heterozygous state, were classified as disease-causing
(Table 4; Additional file 4). One additional disease-causing mutation was found upon read inspection in IGV,
namely c.148_149delinsC in RAB27A (P53). The same disease-causing mutation was also missed by the variant calling program
in our validation study (Table 2; Fig. S1a in Additional file 2).

Table 4. Details of disease-causing mutations identified in the prospective cohort

In addition to analysis of single nucleotide variants and small indels, we performed
coverage analysis to identify larger homozygous deletions (Fig. 1a). We identified a hemizygous deletion of XIAP in P26 and a large homozygous deletion of STX11 in P56 (Fig. S1c, d in Additional file 2).

Molecular diagnoses

In total, we identified and validated 22 unique disease-causing mutations located
in six different genes (Table 4), achieving a molecular diagnosis in 22 patients (overall diagnostic rate 38 %, 22
of 58). The diagnostic yield was higher, 65 % (13 of 20), in the group of patients
with HLH presentation before one year of age compared with 24 % (9 of 37) among the
older patients (Fig. 2b). Excluding adult cases of HLH (n?=?8), the diagnostic rate was 44 % (22 of 50) among the pediatric cases. Interestingly,
the oldest patient with primary HLH in this cohort, aged 16 years (P48), had compound
heterozygous variants in PRF1, c.272CT (p.Ala91Val) and c.1288GT (p.Asp430Tyr). The variant p.Ala91Val has been
associated with later onset of disease when in trans to other PRF1 mutations 36].

We identified patients with biallelic mutations in PRF1 (n?=?7), UNC13D (n?=?6), STX11 (n?=?4), RAB27A (n?=?2) and LYST (n?=?2) as well as a patient with a hemizygous mutation in XIAP (Fig. 2b, Table 4). The mutational spectrum was broad, including missense, nonsense, and splicing mutations,
indels, small and large deletions. Eight of the 22 identified mutations are novel
(Table 4). Mutations were identified in five out of six patients presenting with albinism
and HLH. Interestingly, one such patient with albinism (P1) was diagnosed with a homozygous
UNC13D splice-site mutation (c.570-1GA). No patients were found to have mutations in STXBP2, SH2D1A, ITK, MAGT1, AP3B1, and BLOC1S6.

Finally, in 36 out of 58 patients, biallelic variants that could explain the disease
phenotype were not detected. The age at diagnosis of HLH was significantly higher
compared with genetically diagnosed patients (Wilcoxon rank sum test, p?=?0.06). Moreover, 61 % (22 of 36) were diagnosed with a concomitant disease known
to predispose to secondary HLH. The most commonly associated diseases were EBV infection
(n?=?10), other infections (n?=?4), and hematological cancer (n?=?3). In contrast, an infectious trigger was reported in only 4 of 22 (18 %) of the
genetically diagnosed patients. Thus, the group of undiagnosed patients had a higher
frequency of known triggers of HLH (Fisher’s exact test, p?=?0.002). Of note, seven out of the eight adult HLH cases (88 %) were associated
with a known trigger of HLH, suggesting that these may truly represent secondary HLH
cases.

Correlation between genetic and functional findings

Results from at least one NK cell or CD8
⁺
T cell functional assay were available from 33 patients, including 13 patients with
a molecular diagnosis and 20 patients without a definitive diagnosis. Substantiating
the genetic findings, immunological analyses revealed defective NK cell cytotoxicity
and virtually absent perforin expression in the four patients with biallelic PRF1 mutations studied (Fig. 2a, c, d). Moreover, CD8
⁺
T-cell and NK cell exocytosis was defective in patients with mutations in UNC13D, STX11, RAB27A, and LYST (Fig. 2a, e, f). Interestingly, a patient with RAB27A mutations presented with defective NK and CD8
⁺
T cell exocytosis, but normal NK cell cytotoxicity, while a patient with LYST mutations displayed defective NK cell cytotoxicity but only abnormal NK and CD8
⁺
T-cell exocytosis (Fig. 2a, d, e). Overall, all patients with a genetic diagnosis for which functional data were available
displayed a functional defect by at least one diagnostic assay.

Among the undiagnosed patients, seven patients displayed defective NK cell cytotoxicity
(10 LU; n?=?6) and/or exocytosis (5 % CD107a
⁺
NK cells following K562 target cell incubation; n?=?4). Three such cases belonged to the adult HLH cohort (Fig. 2a). Therefore, 4 out of 15 pediatric HLH cases (26 %) for which functional data were
available displayed a defective NK cell function. In a few cases, the low NK cell
cytotoxicity could reflect the low percentage of NK cells in PBMCs (Additional file
5). Notably, none of the undiagnosed patients with defective exocytosis against K562
target cells displayed concomitant defective NK cell exocytosis following anti-CD16
stimulation or defective CD8
⁺
CD57
⁺
T-cell exocytosis following anti-CD3 stimulation (Fig. 2e). This result contrasted patients with biallelic UNC13D, STX11, STXBP2, RAB27A, or LYST mutations, which all displayed defective exocytosis in response to all stimuli. A
greater variability has been noted in assays quantifying NK cell exocytosis in response
to Fc receptor engagement or CD8
⁺
CD57
⁺
T-cell exocytosis in response to T-cell receptor engagement 20]. Taken together, it is possible that impaired K562 target cell-induced exocytosis
in these patients does not reflect mutations in proteins generally required for cytotoxic
lymphocyte exocytosis.

Contribution of monoallelic mutations as cause of HLH

In patients without an established molecular diagnosis based on biallelic or hemizygous
mutations, we identified seven different monoallelic variants in nine patients with
damaging predictions by either SIFT or PolyPhen-2 (Additional file 6). These were regarded as monoallelic variants of unknown significance. Three patients
carried the variant PRF1 c.272CT p.Ala91Val in a heterozygous state. One of these patients also carried an
additional rare variant with pathogenic prediction in STXBP2 (c.1034CT, p.Thr345Met), a combination previously reported in two patients with
HLH 37]. The monoallelic variants were identified among both pediatric (n?=?7) and adult (n?=?2) patients. Four out of the nine patients with monoallelic variants were reported
to have a known trigger of HLH and only one had a positive family history of unexplained
siblings’ death in childhood. Overall, 25 % of HLH patients without an established
molecular diagnosis carried at least one variant with an in silico damaging prediction.

To interpret findings in patients without an established molecular diagnosis and provide
an overview of genetic variations in genes linked to HLH, we examined the frequency
of variants in genes included in our panel among 2504 unrelated individuals from the
1000 Genomes project. Discarding intronic variants outside splice-site regions and
synonymous variants, 1956 individuals carried at least one variant with a minor allele
frequency lower than 0.05. Applying more strict filters (i.e., at least one damaging
prediction by either SIFT or PolyPhen-2), 636 individuals (25.4 %) were identified
as carrying at least one possibly damaging variant (Additional file 7). The majority of variants were found in LYST and UNC13D, likely reflecting gene size (Fig. S3a, b in Additional file 8). Limiting the analysis to FHL genes, 413 individuals carried at least one possibly
damaging variant. Surprisingly, monoallelic variants in genes linked to HLH were thus
not enriched in patients with HLH lacking a molecular diagnosis (Fig. S3c, d in Additional
file 8). Nonetheless, the PRF1 c.272CT (p.Ala91Val) variant, in a heterozygous state, showed a weak enrichment
(Fisher’s exact test, p value?=?0.07) in patients with HLH but no biallelic mutations. However, a larger
cohort of HLH patients lacking biallelic mutations is required to study this association
further. Of note, two individuals out of 2504 were homozygous for possibly damaging
variants in HLH-related genes, namely PRF1 c.272CT (p.Ala91Val) and UNC13D c.1579CT (p.Arg527Trp).