Differential intolerance to loss of function and missense mutations in genes that encode human matricellular proteins

Abstract

Targeted gene disruption in mice has provided valuable insights into the functions of matricellular proteins. Apart from missense and loss of function mutations that have been associated with inherited diseases, however, their functions in humans remain unclear. The availability of deep exome sequencing data from over 140,000 individuals in the Genome Aggregation Database provided an opportunity to examine intolerance to loss of function and missense mutations in human matricellular genes. The probability of loss-of-function intolerance (pLI) differed widely within members of the thrombospondin, CYR61/CTGF/NOV (CCN), tenascin, small integrin-binding ligand N-linked glycoproteins (SIBLING), and secreted protein, acidic and rich in cysteine (SPARC) gene families. Notably, pLI values in humans had limited correlation with viability of the corresponding homozygous null mice. Among the thrombospondins, only THBS1 was highly loss-intolerant (pLI = 1). In contrast, Thbs1 is not essential for viability in mice. Several known thrombospondin-1 receptors were similarly loss-intolerant, although thrombospondin-1 is not the exclusive ligand for some of these receptors. The frequencies of missense mutations in THBS1 and the gene encoding its signaling receptor CD47 indicated conservation of some residues implicated in specific receptor binding. Deficits in missense mutations were also observed for other thrombospondin genes and for SPARC, SPOCK1, SPOCK2, TNR, and DSPP. The intolerance of THBS1 to loss of function in humans and elevated pLI values for THBS2, SPARC, SPOCK1, TNR, and CCN1 support important functions for these matricellular protein genes in humans, some of which may relate to functions in reproduction or responding to environmental stresses.

Introduction

Matricellular proteins were defined by Paul Bornstein in 1995 as extracellular regulators of cell function that modulate cell behavior by interacting with structural components of the extracellular matrix, cytokines, or proteases and by binding to specific cell surface receptors (Bornstein 1995; Murphy-Ullrich and Sage 2014). With some exceptions, matricellular proteins do not serve structural roles in the extracellular matrix. Rather, they are transiently expressed at specific stages of development, during tissue remodeling, or in response to acute injuries or chronic disease. Original members included thrombospondin-1 and -2, tenascins, the secreted protein acidic and rich in cysteine (SPARC) family, and osteopontin, a member of the small integrin-binding ligand N-linked glycoproteins (SIBLING) family. Matricellular proteins currently include additional families including the Cyr61/CTGF/NOV (CCN) gene family, short fibulins, galectins, and R-spondins (Elola et al. 2007; Knight and Hankenson 2014; Leask 2020; Murphy-Ullrich and Sage 2014; Nakamura 2018).

Human genetics and transgenic mice have provided complementary insights into the functions of matricellular proteins. Missense, regulatory, or inactivating mutations in genes encoding specific matricellular proteins including CCN6, SPARC, SMOC1, SPOCK1, TNXB, DMP1, DSPP, THBS1, THBS2, and COMP have been linked to inherited genetic disorders or disease risk in humans (Abouzeid et al. 2011; Bristow et al. 2005; Burke et al. 2009; Dhamija et al. 2014; Hurvitz et al. 1999; Mendoza-Londono et al. 2015; Okada et al. 2011; Posey et al. 2018; Rainger et al. 2011; Staines et al. 2012; Stenina et al. 2007; Topol et al. 2001). However, genes that serve critical roles during human fetal development may escape detection. Conversely, disruption of matricellular genes in mice by homologous recombination identified CCN1, CCN2, and SMOC1 to be essential for viability (Ivkovic et al. 2003; Mo and Lau 2006; Mo et al. 2002; Okada et al. 2011), but other homozygous null mice were viable, and some initially lacked an obvious phenotype (Bouleftour et al. 2016; Bradshaw 2009; Canalis et al. 2010; Hankenson et al. 2005a, b; Jones and Jones 2000; Kutz et al. 2005; Midwood and Orend 2009; Svensson et al. 2002). In some cases, important gene functions have been revealed when these mice were subjected to specific stresses (Calabro et al. 2014; Kim et al. 2018; Murphy-Ullrich and Sage 2014; Roberts et al. 2012; Soto-Pantoja et al. 2015; Stenina-Adognravi and Plow 2019).

Disruption of Thbs1, encoding thrombospondin-1 in mice, yielded viable mice that were fertile and appeared healthy except for lung inflammation (Lawler et al. 1998). The lung inflammation may relate to exposure to a specific pathogen because the lung phenotype was lost when the mice were rederived in a different vivarium (Isenberg et al. 2008a). Subsequent studies identified beneficial as well as detrimental effects of Thbs1 gene disruption on the ability of mice to survive exposure to specific pathogens or respond to a variety of physiological stresses (Arun et al. 2020; Martin-Manso et al. 2012; McMaken et al. 2011; Qu et al. 2018; Soto-Pantoja et al. 2015; Zhao et al. 2015). These studies illustrate how gene functions can be influenced by the environmental context. Such environmental stresses that could reveal important adaptive functions of matricellular protein genes may be absent in the highly controlled environment of a laboratory vivarium.

Despite our ability to control our environment, the ability to survive numerous environmental stresses including acute injuries and ongoing exposures to endemic and novel pathogens has played an important role in human evolution. Genetic diversity is critical for the long-term survival of any species facing such unpredictable challenges, and identifying relevant variations in specific genes is one goal of population genetics. The Exome Aggregation Consortium (ExAC) assembled a data set containing variant calls across 60,706 human exomes to globally examine the prevalence of missense and predicted loss of function (LoF) mutations (Lek et al. 2016). Known and previously unrecognized essential genes were identified by having significantly fewer LoF mutations than expected. This genomic variation data was expanded to include 141,456 individuals in the Genome Aggregation Database (gnomAD), which currently includes 125,748 deep-sequenced exomes and 15,701 full genome sequences from unrelated individuals (Karczewski et al. 2020). Individuals with severe pediatric genetic diseases and their first-degree relatives were excluded to better reflect the incidence of recessive disease-causing alleles. Here we analyzed gnomAD v2.1.1 data to examine the rates of missense and predicted LoF mutations in several families of matricellular protein genes. Focusing on gene families that include several paralogs provided useful controls because the expected frequencies of LoF mutants depends in part on the length of their coding regions (Lek et al. 2016).

Materials and methods

Data for missense and LoF mutants in genes that encode human matricellular proteins was accessed and analyzed using the ExAC browser (http://exac.broadinstitute.org) and, subsequently, the Genome Aggregation Database (gnomAD, https://gnomad.broadinstitute.org) (Karczewski et al. 2020; Lek et al. 2016). Mouse knockout phenotypes were obtained from mousephenotype.org or http://www.informatics.jax.org/ (Bult et al. 2019) and published studies where indicated. Data for clinical associations with variants in human genes was obtained from ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/).

Results

Variation in the frequencies of LoF mutations in the thrombospondin gene family

Previous studies demonstrated that mice with homozygous LoF mutants in any single THBS family gene are viable (Frolova et al. 2010; Hankenson et al. 2005a, b; Lawler et al. 1998; Svensson et al. 2002). Characterization of strains bearing multiple THBS gene knockouts indicated minimal functional cross-compensation between the five THBS genes (Posey et al. 2008). In contrast, the ExAC data from 60,706 individuals and the expanded human gnomAD dataset representing 141,456 individuals showed deficits in the observed versus expected numbers of individuals with LoF mutants for THBS1 and THBS2 (Table 1). The pLI for THBS1 was 1.00 in both datasets, indicating this gene to be highly loss intolerant. None of the individuals with THBS1 LoF mutant alleles were homozygotes. THBS2 also showed substantial loss intolerance, whereas the numbers of observed LoF mutants in THBS3, THBS4, and COMP did not differ significantly from the expected numbers. None of the THBS2, THBS3, or COMP LoF mutants were homozygous, whereas the THBS4 mutants in two individuals were homozygous frameshifts (p.Thr915GlnfsTer and p.Thr915GlnfsTer).

Table 1.

LoF mutants in the thrombospondin gene family

Gene	ORF (kb)	Expected LoF mutants	Observed LoF mutants	Observed/expected (90% range)	pLI	Null mouse phenotype
THBS1	3.5	56	7	0.13 (0.07–0.23)	1.00	viable
THBS2	3.5	59.5	13	0.22 (0.14–0.35)	0.56	viable
THBS3	2.9	57.4	41	0.71 (0.56–0.93)	0.0	viable
THBS4	2.9	51.6	36	0.7 (0.53–0.92)	0.0	viable
COMP	2.3	39.5	21	0.53 (0.38–0.77)	0.0	viable

Observed numbers of LoF mutants identified based on exome sequencing from 141,456 individuals in the Genome Aggregation Database (v2.1.1) were used to calculate the probability the indicated genes are loss-intolerant (pLI) as described (Karczewski et al. 2020; Lek et al. 2016). Where multiple isoforms exist, ORF length is presented for isoform 1 except where noted. Null mouse viability phenotypes are from http://www.informatics.jax.org/ (Bult et al. 2019).

LoF mutations in the SPARC gene family

Among the Sparc gene family members in mice, only Smoc1 is essential for viability based on studies of homozygous null mice (Okada et al. 2011). Homozygous nonsense and splice mutants in murine Smoc1 recapitulated the features of the SMOC1-dependent human autosomal-recessive disorder microphthalmia with limb anomalies (Waardenburg Anophthalmia syndrome), and mice bearing these mutations died shortly after birth. Despite being essential for normal ocular and limb development in mice and humans (Okada et al. 2011; Rainger et al. 2011), SMOC1 did not show a significant pLI in humans (Table 2). One caveat in interpreting this result is that the number of expected LoF mutants for SPARC family genes is smaller than for the THBS family because the former have much shorter coding regions. Because the 90% confidence range for all of the SPOCK and SMOC paralogs extends to significant observed/expected LoF ratios, future availability of exome data for a larger human population will be required to confirm or exclude significant pLI values for any of these genes.

Table 2.

LoF mutations in the SPARC gene family

Gene	ORF (kb)	Expected LoF mutants	Observed LoF mutants	Observed/expected (90% range)	pLI	Null mouse phenotype
SPARC	0.9	16	2	0.12 (0.05–0.39)	0.89	Viable
SPARCL1	2.0	28.8	19	0.66 (0.46–0.97)	0.0	Viable
SPOCK1	1.3	23.4	4	0.17 (0.08–0.39)	0.83	Viable
SPOCK2	1.3b	24.9	6	0.24 (0.13–0.47)	0.24	Pendinga
SPOCK3	1.3b	25.4	8	0.32 (0.18–0.57)	0.01	Viable
SMOC1	1.3	24.4	8	0.33 (0.19–0.59)	0.01	Neonatal lethal
SMOC2	1.4	25.3	9	0.36 (0.21–0.62)	0.0	Viable

^aMouse registered but phenotype not currently available at www.mousephenotype.org

^bIsoform 2

Despite being nonessential for development in mice (Gilmour et al. 1998; Roll et al. 2006), SPARC and SPOCK1 had high pLI values in the human data (Table 2). Humans lacking SPARC have not been reported to date, but missense mutations in SPARC cause osteogenesis imperfecta, type XVII (Mendoza-Londono et al. 2015). Most studies of SPARC in humans have focused on its role in various cancers and their metastatic spread (Nagaraju et al. 2014). Abnormalities in Sparc null mice include cataract formation and rupture of the lens capsule in the eye, severe osteopenia, and accelerated closure of dermal wounds (Bradshaw 2009; Gilmour et al. 1998).

Humans lacking SPOCK1 have not been reported to date, but a missense mutation in SPOCK1 (p.D80V) was identified in a patient with developmental delay, agenesis of the corpus callosum, and microcephaly (Dhamija et al. 2014). SPOCK1 encodes the proteoglycan testican-1, which is involved in neurogenesis and epithelial-to-mesenchymal transition (Roll et al. 2006; Sun et al. 2020). The lack of obvious abnormalities in Spock1 null mice may be due to functional redundancies that were reported with testican-2 (Spock2) and testican-3 (Spock3) (Roll et al. 2006).

LoF mutations in the CCN gene family

CCN1 was the only member of the CCN gene family with an elevated pLI (Table 3). This is consistent with the embryonic and perinatal lethal phenotype of a homozygous Ccn1 LoF mutant in mice (Mo and Lau 2006; Mo et al. 2002). As with the SPARC family, the relatively short coding sequences of CCN genes may require exome data from a larger population to reliably confirm or exclude intolerance to LoF, especially for CCN3 and CCN6 where the 90% range for observed/expected LoF mutants extends below the standard 0.3 pLI threshold. Loss of Ccn2 limits viability in mice based on its roles in bone and lung development (Baguma-Nibasheka and Kablar 2008; Ivkovic et al. 2003; Kawaki et al. 2008). None of the coding variations for CCN2 currently in ClinVar have a known clinical relevance, but analysis of additional exomes may determine whether human functions of CCN2 parallel those identified in the null mice. Three siblings diagnosed with early-onset parkinsonism were homozygous for a p.D82G mutation in CCN3 (Bentley et al. 2020), which suggests a pathophysiological function that would not be detected by this LoF screen. The same mutation occurred once as a heterozygous variant in the gnomAD dataset (Karczewski et al. 2020; Lek et al. 2016).

Table 3.

LoF mutations in the CCN gene family

Gene	ORF (kb)	Expected LoF mutants	Observed LoF mutants	Observed/expected (90% range)	pLI	Null mouse phenotype
CCN1 (Cyr61)	1.1	17.2	3	0.17 (0.08–0.45)	0.71	Embryonic or perinatal lethal
CCN2 (CTGF)	1.0	12.4	7	0.56 (0.32–1.06)	0.0	Perinatal lethal
CCN3 (NOV)	1.1	14.7	6	0.41 (0.22–0.81)	0.01	Viable
CCN4 (WISP1)	1.1	17.3	12	0.69 (0.44–1.13)	0.0	Viable
CCN5 (WISP2)	0.8	9.8	8	0.82 (0.48–1.46)	0.0	Viable
CCN6 (WISP3)	1.2	17.1	8	0.47 (0.27–0.84)	0.0	Viable

Previous studies indicated that the phenotypes associated with LoF mutants in CCN6 diverge between humans and mice (Kutz et al. 2005). Inactivating mutations in human CCN6 cause an autosomal recessive skeletal disorder, progressive pseudorheumatoid dysplasia (Hurvitz et al. 1999), whereas comparable mutants of Ccn6 in mice had no skeletal phenotype (Kutz et al. 2005). In humans with pseudorheumatoid dysplasia, a nonsense variant of CCN6 was always in cis with a G83E missense allele (Supplemental Table 20 in (Lek et al. 2016)).

LoF mutations in the tenascin gene family

TNR was the only member of the tenascin gene family with an elevated pLI (Table 4). Because Tnr null mice are viable, and the associated null phenotypes involve altered cognitive functions (Weber et al. 1999), the rationale for a deficit in LoF mutants in humans is unclear. Tenascin-R is primarily expressed in the central nervous system, and homozygous deletion of TNR was found in a patient with intellectual disability (Dufresne et al. 2012). The SNV rs6686722 in TNR was associated with attention deficit hyperactivity disorder in a hypothesis-free genome-wide association study (Hawi et al. 2018).

Table 4.

LoF mutations in tenascins

Gene	ORF (kb)	Expected LoF mutants	Observed LoF mutants	Observed/expected (90% range)	pLI	Null mouse phenotype
TNC	6.6	91.4	30	0.33 (0.24–0.45)	0.00	Viable
TNXB	2.0b	21.8	7	0.32 (0.18–0.6)	0.02	Viable
TNN	3.9	59.6	49	0.82 (0.65–1.04)	0.00	Pending*
TNR	4.1	72.4	16	0.22 (0.15–0.34)	0.52	Viable

*Mouse registered but phenotype not currently available at www.mousephenotype.org

^bIsoform 2

LoF mutations in the SIBLING gene family

None of the SIBLING family genes are essential for viability in mice, and reported disease-associated mutations in humans generally have postnatal effects (Bouleftour et al. 2016; Staines et al. 2012). Correspondingly, elevated pLI values were not found for these genes in the gnomAD data (Table 5). Mutations in DMP1 cause autosomal recessive hypophosphatemic rickets, type 1 and osteomalacia (Feng et al. 2006; Lorenz-Depiereux et al. 2006). Mutations in DSPP are associated with dentinogenesis imperfecta and dentin dysplasia (Song et al. 2008; Zhang et al. 2001).

Table 5.

LoF mutations in SIBLING gene family members

Gene	ORF (kb)	Expected LoF mutants	Observed LoF mutants	Observed/expected (90% range)	pLI	Null mouse phenotype
SPP1	0.9	9.7	9	0.93 (0.56–1.59)	0.0	Viable
DMP1	1.5	18.1	13	0.72 (0.47–1.14)	0.0	Viable
DSPP	3.9	17.6	13	0.74 (0.48–1.17)	0.0	Viable
MEPE	1.6	7	4	0.57 (0.28–1.29)	0.01	Viable
IBSP	1.0	17.5	17	0.97 (0.67–1.46)	0.0	Viable

Osteopontin (SPP1), Dentin matrix protein 1 (DMP1), Dentin sialophosphoglycoprotein (DSPP), Matrix extracellular phosphoglycoprotein (MEPE), and Bone sialoprotein (IBSP).

Distribution and frequencies of missense mutations in THBS family genes

In the ExAC data, deficits in LoF mutants positively correlated with deficits in missense SNVs (Lek et al. 2016). Consistent with this global correlation and the pLI data in Table 1, THBS1 had the highest Z-score in the thrombospondin gene family for observed/expected missense mutants in the gnomAD data, and THBS2 was the second highest (Table 6). Notably, all members of the THBS family had deficits in observed versus expected missense SNVs that exceeded the 90% range.

Table 6.

Missense mutation frequencies in matricellular protein genes

Gene	Expected missense SNVs	Observed missense SNVs	Observed/expected (90% range)	Z score
THBS1	721.4	516	0.72 (0.67–0.77)	2.72
THBS2	758.7	587	0.77 (0.72–0.83)	2.21
THBS3	574.9	455	0.79 (0.73–0.85)	1.78
THBS4	553.9	474	0.86 (0.79–0.92)	1.21
COMP	454.8	348	0.77 (0.7–0.84)	1.78
SPARC	180.5	139	0.77 (0.67–0.89)	1.10
SPARCL1	340.3	337	0.99 (0.91–1.08)	0.06
SPOCK1	241.7	190	0.79 (0.7–0.89)	1.18
SPOCK2	249	199	0.8 (0.71–0.9)	1.13
SPOCK3	235.9	219	0.93 (0.83–1.04)	0.39
SMOC1	246.8	214	0.87 (0.78–0.97)	0.74
SMOC2	284.5	254	0.89 (0.81–0.99)	0.64
CCN1	216.2	201	0.93 (0.83–1.04)	0.37
CCN2	182.8	163	0.89 (0.78–1.01)	0.52
CCN3	201.3	192	0.95 (0.85–1.07)	0.23
CCN4	245.5	236	0.96 (0.86–1.07)	0.22
CCN5	156.8	144	0.92 (0.8–1.05)	0.36
CCN6	190.2	184	0.97 (0.86–1.09)	0.16
TNC	1287.7	1296	1.01 (0.96–1.05)	-0.08
TNXB	246.9	245	0.99 (0.89–1.1)	0.04
TNN	781.6	824	1.05 (0.99–1.12)	-0.54
TNR	813	685	0.84 (0.79–0.9)	1.60
SPP1	176.5	166	0.94 (0.83–1.07)	0.28
DMP1	267.6	255	0.95 (0.86–1.06)	0.27
DSPP	676.8	596	0.88 (0.82–0.94)	1.10
MEPE	276.8	268	0.97 (0.88–1.07)	0.19
IBSP	173.2	156	0.9 (0.79–1.03)	0.46

Higher positive Z-scores indicate increased selective pressure to limit missense mutations

The frequency of missense variants across the THBS1 coding sequence is presented in Fig. 1. One of the most frequent variants is at N700. The N700S variant was associated with increased risk for early myocardial infarction (Topol et al. 2001), and biochemical studies established that this variant decreases the affinity for calcium binding (Stenina et al. 2005) and destabilizes the protein (Carlson et al. 2008). In addition to altering the secretion or stability of thrombospondin-1, missense mutations could interfere with its interactions with other proteins that mediate its functions, including multiple cell surface receptors (Resovi et al. 2014). Except for more frequent variations at G454 and R517 in sequences identified to be recognized by CD36 (Dawson et al. 1997; Tolsma et al. 1993), all variations in sequences previously implicated in interactions of thrombospondin-1 with its integrin and non-integrin receptors (Calzada and Roberts 2005) occurred at frequencies < 10^–4. Furthermore, the frequencies of rare variants in these sequences were similar to those for SNVs occurring throughout the coding sequence. Therefore, the frequency of missense SNVs in these putative functional sequences is insufficient to infer protection of specific receptor binding sites.

Fig. 1.

Missense mutations of putative functional sequences in human thrombospondin-1 (P07996). Yellow highlighted regions indicate peptide sequences reported to engage the indicated thrombospondin-1 receptors or ligands. Residues are colored based on variant frequency: no variants (black), > 10^–6 (blue), > 10^–5 (green), > 10^–4 (orange), and 10^–3 to 10^–1 (red)

Missense SNV frequencies in other matricellular gene families

The loss-intolerant SPARC family members SPARC and SPOCK1 also had elevated Z-scores for deficits in missense SNVs (Table 6). As noted previously, missense mutations in SPARC (E263K, R166H) cause osteogenesis imperfecta, type XVII (Mendoza-Londono et al. 2015), and a p.D80V missense mutation in SPOCK1 was associated with developmental delay, agenesis of the corpus callosum, and microcephaly (Dhamija et al. 2014). Only one SPARC p.Arg166His allele (frequency 3.98 × 10^–6) was found in gnomAD. SPARC p.E263K and SPOCK1 p.D80V variants were not found in the gnomAD data. Consistent with Smoc1 being essential for viability in mice and potential cross compensation between SMOC1 and SMOC2 (DeGroot et al. 2019), SMOC1 and SMOC2 had similar moderate deficits in missense SNVs (Z = 0.74 and 0.64, respectively).

Among the tenascins, only TNR had a deficit in observed versus expected missense SNVs that exceeded the 90% range (Z = 1.60, Table 6). Several TNR missense variants (C155S, T166A, N180H, T592A) were previously linked to familial Parkinson disease, but their pathologic significance remained uncertain (Farlow et al. 2016), The variant p.Cys155Ser occurred in 13 alleles in gnomAD with a frequency of 4.63 × 10^–5, p.Thr166Ala in 1175 alleles (4.2 × 10^–3), p.Asn180His in 1164 alleles (4.12 × 10^–3), and p.Thr592Ala in 53 alleles (1.98 × 10^–4). The high frequencies of these variants raise caution regarding their disease relevance. Missense variants in TNR (p.Arg1192Trp, p.Ala397Thr) were also linked to a nonprogressive neurodevelopmental disorder with spasticity and transient opisthotonos, with the former variant being clinically significant (Wagner et al. 2020). The R1192W variant was not found in the gnomAD data, which supports its role in disease. The A397T variant occurred in 4 alleles with a frequency of 1.59 × 10^–5, suggesting need for further investigation.

In the SIBLING family, only DSPP had a deficit in observed versus expected missense SNVs that exceeded the 90% range (Z = 1.10, Table 6). Multiple deletion and missense mutations in DSPP including p.Ala15Val, p.Pro17Thr, p.Val18Phe, and p.Pro19Leu variants result in dentinogenesis imperfecta, deafness, and autosomal dominant nonsyndromic sensorineural 39 (de La Dure-Molla et al. 2015; Liang et al. 2019). Among these residues only p.Pro17Ser had a single variant in gnomAD. The occurrence of multiple pathogenesis-associated missense mutations in DSPP may account for the overall deficit in mutations in this gene among healthy individuals.

Intolerance to LoF mutations in thrombospondin-1 receptors

The pLI values for known thrombospondin-1 receptors were examined to identify genetic evidence for which receptors mediate critical functions of thrombospondin-1 (Table 7). CD36 was the first receptor linked to the anti-angiogenic activity of thrombospondin-1 (Dawson et al. 1997), but LoF mutations in CD36 occurred at 2.8 times the expected rate for this gene (pLI 0.0, Table 7). The elevated frequency of LoF mutations is consistent with prior reports that genetic deficiencies associated with loss of CD36 expression on red blood cells (Nak^a-negative phenotype) are common in Asian and African populations (Curtis and Aster 1996; Hirano et al. 2003). Type 1 LoF CD36 mutations in these populations may be related to resistance to malaria (Chilongola et al. 2009; Liu et al. 2020), which would provide a rationale for the high number of observed LoF mutants. Conversely, deletion of CD36 has been linked with cardiovascular disease and to insulin resistance (Miyaoka et al. 2001; Yuasa-Kawase et al. 2012). These functions of CD36 may be independent of its role as a thrombospondin-1 receptor, thereby accounting for the divergence of the pLI values for these genes.

Table 7.

LoF mutants in genes encoding thrombospondin-1 receptors

Gene	Expected LoF mutants	Observed LoF mutants	Observed/expected (90% range)	pLI	Null mouse phenotype
CD36	23.4	66	2.82 (1.77–2)	0.0	Viable
CD47	16.9	2	0.12 (0.05–0.37)	0.92	Viable
PTPRJ (CD148)	62.4	29	0.46 (0.34–0.63)	0.0	Viable/lethal
CACNA2D1 (α2δ1)	74.4	11	0.15 (0.09–0.24)	1.00	Viable
STIM1	31.3	6	0.19 (0.1–0.38)	0.78	Peri- and postnatal lethal
LRP1	246.9	8	0.03 (0.02–0.06)	1.00	Embryonic lethal
ITGA3	58.9	22	0.37 (0.27–0.53)	0.0	Perinatal lethal
ITGA4	62.9	19	0.3 (0.21–0.44)	0.0	Embryonic lethal
ITGA6	61.6	23	0.37 (0.27–0.53)	0.0	Perinatal lethal
ITGB1	38.7	6	0.15 (0.08–0.31)	0.98	Embryonic lethal
ITGAV	62.9	19	0.3 (0.21–0.44)	0.0	Viable
ITGB3	37.4	12	0.32 (0.2–0.52)	0.0	Reduced viability

CD47 mediates signaling functions of CD47 in several cell types (Soto-Pantoja et al. 2015). In contrast to CD36, CD47 had a low frequency of LoF mutations and high loss intolerance (pLI = 0.92, Table 7). CD47 is not essential for viability in mice (Lindberg et al. 1996; Soto-Pantoja et al. 2015), and loss of CD47 expression has only been reported in human red blood cells in the context of mutations in protein 4.2 (EPB42) that cause hereditary spherocytosis (Bruce et al. 2002). CD47 also has an independent function as the counter-receptor of SIRPα (Barclay and Van den Berg 2014), and SIRPA also had elevated loss-intolerance (pLI = 0.67, o/e = 0.19 (0.09–0.43). Therefore, the basis for loss intolerance in human CD47 remains unclear, but it is a candidate for playing a significant role in the loss intolerance of THBS1. Analysis of the frequency of coding variants in CD47 revealed only rare variations in residues involved in its interaction with the counter-receptor SIRPα (Hatherley et al. 2008) and those subject to post-translational modifications including the glycosylation required for THBS1-dependent signaling (Kaur et al. 2011) (Fig. 2).

Fig. 2.

Missense mutation frequencies in human CD47. Residues involved in SIRPα binding (yellow highlight) generally lack mutations. Residues subject to posttranslational modifications are highlighted in gray. Thrombospondin-1 signaling requires posttranslational modification of S⁸² (Kaur et al. 2011)

Thrombospondin-1 binding increases the enzymatic activity of the membrane-bound tyrosine phosphatase CD148, encoded by PTPRJ (Takahashi et al. 2012). PTPRJ is a tumor suppressor gene, and allelic loss or loss of heterozygosity (LOH) occurs in human sporadic colorectal, lung and breast carcinomas and non-Hodgkin’s lymphomas (Aya-Bonilla et al. 2013; Ruivenkamp et al. 2002). Two LoF variants that result in frameshift and insertion of a premature stop codon in PTPRJ (g.48131608A > G (c.97-2A > G and g.48158556delG (c.1875delG) are associated with an inherited autosomal recessive thrombocytopenia (Marconi et al. 2019). Consistent with Ptprj-deficient mice being viable, fertile, and without any anatomical abnormalities (Trapasso et al. 2006), PTPRJ did not exhibit an elevated pLI (Table 7). Therefore, this interaction is unlikely to account for the loss intolerance of THBS1.

The auxiliary subunit of voltage-gated calcium channels α2δ1 interacts with several members of the THBS gene family that regulate channel function including thrombospondin-1 (Eroglu et al. 2009; Taylor and Harris 2020). The low frequency of LoF for CACNA2D1 resulted in a pLI of 1.0 (Table 7), which is consistent with clinical pathologies associated with CACNA2D1 mutations. Missense mutations in the extracellular region of α2δ1 at c.2867C > A p.S956T, c.2126 G > A p.S709N, and in the Cache domain (c.1648 G > T p.D550Y) have been associated with early repolarization syndrome (ERS) and inherited Brugada syndrome /J-wave syndromes that cause sudden cardiac death (Burashnikov et al. 2010). Loss of function mutants of CACNA2D1 were also reported in a patient with Short QT syndrome (Templin et al. 2011) and associated with epilepsy and intellectual disability (Vergult et al. 2015). Because THBS4 also interacts with and regulates α2δ1 but was not loss intolerant, the relevance of this receptor to the elevated pLI for THBS1 is unclear.

Stromal interaction molecule (STIM1) is an essential regulator of store-operated Ca2 + entry (SOCE) and Ca²⁺ release activated Ca²⁺ (CRAC) channels by binding to ORAI1 (Feske 2010). In addition to its intracellular roles in calcium signaling, some STIM1 is on the cell surface and interacts with thrombospondin-1 (Ambily et al. 2014; Duquette et al. 2014). STIM1 had a deficit in LoF mutations that indicates loss intolerance (pLI = 0.78, Table 7), which is consistent with the perinatal lethality of the null mouse (Varga-Szabo et al. 2008). LoF mutations in human STIM1 and ORAI1 abolished CRAC and SOCE channel currents and are associated with severe combined immunodeficiency, congenital myopathy, and anhydrotic ectodermal dysplasia (Feske 2010; Lacruz and Feske 2015). Therefore, STIM1 is also a candidate to play a role in the loss intolerance of THBS1.

Low density lipoprotein receptor-related protein 1 (LRP1) is a scavenger receptor that mediates endocytosis of multiple ligands including lipoproteins and Thrombospondin-1 (Gonias et al. 2004). LRP1 modulates downstream intracellular signaling controlling cell survival associated with tissue remodeling in response to injury via a thrombospondin-1-calreticulin complex (Pallero et al. 2008). LoF mutations in LRP1 were rare, indicating a high degree of loss intolerance (pLI = 1.0, Table 7), consistent with the embryonic lethal phenotype of Lrp1 null mice at embryonic implantation (Herz et al. 1992). The promiscuity of this receptor precludes assessment of its relevance to the loss intolerance of THBS1.

Several integrin heterodimers act as thrombospondin-1 receptors including α3β1, α4β1, α6β1, and αvβ3 (Calzada and Roberts 2005; Resovi et al. 2014). Of the genes encoding their respective subunits, only ITGB1 showed an elevated (pLI = 0.98, Table 7). Although Itgb1 is essential for embryonic development in mice, the role of β1-integrins as receptors for multiple ECM proteins including members of the THBS, CCN, and tenascin families precludes assigning a specific role for any individual integrin in the loss intolerance for THBS1 (Humphries et al. 2006).

Discussion

The elevated pLI values observed for several matricellular protein genes infer that LoF mutations in those genes confers a significant survival or reproductive disadvantage in humans (Lek et al. 2016). When homozygous knockout of the murine ortholog indicates an essential role in development, as reported for CCN1, loss-intolerance for the human gene is consistent with a similar role in human embryonic development. On the other hand, loss-intolerant matricellular genes such as THBS1, THBS2, SPARC, SPOCK1, and TNR that are not essential for murine embryonic development may have critical roles in human postnatal survival or in adult reproductive function. In the case of THBS1, the selective pressure to maintain this gene may have more than one origin. Studies in mice and primates indicated specific roles for Thbs1 in reproduction (Bender et al. 2019; Greenaway et al. 2007) and in the ability of adult mice, rats and pigs to repair dermal wounds and survive exposure to ischemic injuries or genotoxic stress (Isenberg et al. 2007, 2008a, b; Soto-Pantoja et al. 2015). Loss of Thbs1 in mice also alters their survival following exposure to several pathogens (Arun et al. 2020; Binsker et al. 2019; Lawler et al. 1998; Martin-Manso et al. 2012; Qu et al. 2018). Loss of Thbs1 impairs survival for some of these stresses while improving survival or recovery from other stresses. These animal studies suggest that LoF mutants in human THBS1 could alter the probability of surviving acute injuries and infections, some of which could lead to a decrease in longevity or success in reproduction. The multiplicity of functions for thrombospondin-1 and other matricellular proteins suggests that no single function will account for the selection against individuals carrying LoF or missense mutants.

Although the expectation–maximization algorithm used in calculating the pLI values compensates for the influence of coding sequence length on the expected number of LoF mutants, genes with longer coding sequences remain more likely to achieve a significant pLI at a given sample number (Lek et al. 2016). Therefore, the predictions of loss-intolerance are more reliable for the thrombospondin and tenascin family members with longer ORFs than for CCN or SIBLING family genes. Despite this limitation, the CCN1 data demonstrated loss-intolerance in humans, as expected based on its essential role in mouse embryonic development (Mo and Lau 2006; Mo et al. 2002). Additional factors including the breadth of tissue expression also correlate broadly with obtaining an elevated pLI (Lek et al. 2016), suggesting that matricellular genes with more restricted tissue expression or organ-specific essential functions are less likely to be detected in the gnomAD data. Temporal differences in the expression of matricellular proteins may also be a factor, as was documented for THBS1 versus THBS2 during dermal wound repair (Agah et al. 2002).

Significant loss intolerance was not demonstrated for some of the genes with shorter ORFs such as SMOC1, but the 90% confidence range extended beyond the cutoff for significant loss-intolerance, consistent with its essential role in murine embryonic development and other evidence that SMOC1 plays an important role in human embryonic development (Okada et al. 2011; Rainger et al. 2011). As the number of available human genomes in gnomAD increases, additional matricellular genes may be identified to be significantly LoF-intolerant.

The overall deficit in missense mutations in THBS1 and THBS2 are consistent with selection pressures to maintain the functional integrity of these proteins. However, the distribution of missense mutations in THBS1 did not clearly identify specific residues or regions of the protein that mediate these functions beyond those residues previously identified through genome-wide association studies linking one polymorphism in THBS1 with cardiovascular disease. In contrast, multiple residues in DSPP that are subject to disease-causing missense mutations were invariant in the gnomAD data. Thus, another use for this broad population data is to validate the absence of previously identified disease-causing mutations in a large healthy population. Further analyses of missense mutations in other matricellular genes may also provide insights into specific interactions that mediate functional roles of these proteins in human development and disease.

Recent advances in understanding the complex effects of LoF mutations on gene regulation may help explain the observed divergence between pLI values for human matricellular protein genes such as THBS1, SMOC1 and CCN2 and the viability of mice bearing LoF mutants in the corresponding murine orthologs. One advantage of the gnomAD LoF analysis over the knockout mouse studies is that pLI values are derived from multiple independent LoF mutants in each human gene, whereas the mouse phenotypes are typically based on a single gene knockout strategy. Gene knockout strategies can have unanticipated effects that extend beyond the targeted gene. As was reported for Thbs3 in mice, matricellular gene regulation may involve elements located within an adjacent essential gene (Collins et al. 1998). Using different methods to inactivate a gene can also result in contradictory phenotypes by triggering compensatory responses including transcriptional adaptation (Kontarakis and Stainier 2020). A subset of mutations causing premature transcript termination can result in gain of function rather than LoF alleles (Coban-Akdemir et al. 2018), An analysis of the ExAC data using an algorithm to predict such dominant gain of function alleles did not identify potential gain of function alleles for THBS1, SMOC1 and CCN2 (Coban-Akdemir et al. 2018). However, potential gain of function alleles were identified for DMP1 and DSPP in the ExAC data (Supplemental Table 4 in Coban-Akdemir et al. 2018). Based on these data, clinically relevant gain of function mutants in SIBLING family genes should be further investigated.

Abbreviations

α2δ1

Precursor of the α2 and δ subunits of voltage-dependent calcium channel, encoded by CACN2D1 

CCN

Cyr61/CTGF/NOV gene family

CD148

Membrane-bound tyrosine phosphatase, encoded by PTPRJ

COMP

Cartilage oligomeric matrix protein

DMP1

Dentin matrix protein 1

DSPP

Dentin sialophosphoglycoprotein

ExAC

Exome Aggregation Consortium

gnomAD

Genome Aggregation Database

IBSP

Bone sialoprotein

LoF

Loss of function

LRP1

Low density lipoprotein receptor-related protein 1

MEPE

Matrix extracellular phosphoglycoprotein

pLI

Probability of loss-of-function intolerance

SIBLING

Small integrin-binding ligand N-linked glycoproteins

SIRPα

Signal regulatory protein-α

SMOC

Secreted modular calcium-binding protein

SNV

Single nucleotide variation

SPARC

Secreted protein, acidic and rich in cysteine

SPOCK

Sparc/osteonectin, CWCV, and kazal-like domains proteoglycan (Testican)

SPP1

Osteopontin

STIM1

Stromal interaction molecule 1

THBS

Thrombospondin

TN

Tenascin

Funding

This work was supported by the Intramural Research Program of the NIH/NCI (ZIA SC009172).

Availability of data and materials

All data is contained in the manuscript or the indicated public databases.

Compliance with ethical standards

Conflict of interest

The authors have no relevant financial or non-financial interests to disclose.