Prolonged replication of SARS-CoV-2 in immunocompromised hosts under inefficient immune pressure promotes intra-host diversification and selection of immune escape variants. Recently, a number of single case reports of such individuals have shown within host evolution with a focus on the adaptation of the spike protein to neutralising antibodies1–4. Selection pressure imposed by CD8 T-cell responses is a major driving force for viral adaptation in other viral infections such as HIV5. T-cell responses in SARS-CoV-2 infection are associated with immune control, but so far the potential of this novel coronavirus to escape these responses and its clinical implications remain unclear6. Here we investigate the case of a severely immunocompromised host with considerable viral diversification and show that most of the acquired mutations evade recognition by humoral and cellular immunity.
A woman in her seventies was hospitalised in spring 2020 with COVID-19 and severe ARDS. She was managed in our intensive care unit developing multiple complications and multi-organ failure until her death five months later. Her past medical history included follicular lymphoma treated with standard chemotherapy consisting of cyclophosphamide, doxorubicin, vincristine and prednisolone in combination with the B-cell-depleting anti-CD20 antibody obinutuzumab up to a month prior to presentation. As expected, standard ELISAs (Euroimmune) for SARS-CoV-2-specific IgG and IgA were negative throughout the course of disease. COVID-19-directed treatment included steroids and multiple doses of convalescent plasma (Fig. 1a). Therapy with remdesivir was deemed unsafe due to severe renal failure and known hypersensitivity to the vehicle sulfobutylether-β-cyclodextrin. Virological studies revealed persistently high SARS-CoV-2 RNA levels (Extended Data Fig. 1) with replication-competent virus in respiratory tract specimens until death 156 days following COVID-19 diagnosis.
In order to study intra-host viral evolution, sequential upper and lower respiratory tract specimens were used to generate 21 near full length genomes of SARS-CoV-2 spanning 150 days (Supplementary Table 1). The infecting strain was classified as B.1.1.29 (20B) bearing the spike mutation D614G and clustered with other sequences circulating at that time (Extended Data Fig. 2). Over the following 5 months the virus diversified extensively with 34 different mutations in coding regions emerging at >40% frequency abundance (Extended Data Fig. 3). To study viral evolution in adaptation to the host adaptive immune response, we focused our analyses on the 12 non-synonymous mutations that became fixed in the dominant variants until death (Fig. 1b).
We recognised three mutations located in the spike gene previously associated with antibody escape: A deletion at position 144 in the N-terminal domain (NTD) (mutation I), and S477N (mutation III) and E484K (mutation II) substitutions in the receptor binding domain (RBD) (Fig. 1c). We performed in vitro neutralisation assays testing the activity of serum samples taken from the donors of convalescent plasma against patient isolates. All but one donor showed robust neutralising responses against an early pandemic isolate carrying only the D614G spike mutation (MUC-IMB01) and an autologous isolate sampled at day 20, before the emergence of the additional mutations in the spike gene (Fig. 1d). However, the autologous variant isolated at day 83 carrying the del144 and E484K spike mutations was only weakly neutralised by one donor serum and none of the others tested, showing that the combination of these RBD and NTD alterations was sufficient to almost completely abrogate neutralising activity. Structural analysis illustrates that the two mutations in the RBD and the deletion in the NTD are located in epitopes targeted by exemplary neutralising antibodies REGN10993, Ly-COV555, and 4A8 (Fig. 1e). Taken together, these data indicate that the neutralising polyclonal antibodies administered with the convalescent plasma transfusions mediated immune selection pressure that drove adaptation of the viral spike protein to evade these responses.
We investigated whether the lack of an effective T-cell response might have contributed to the failure to clear SARS-CoV-2 infection. An IFN-γ ELISPOT assay that was performed with a fresh peripheral blood sample on day 127 showed that the patient had strong T-cell responses to antigens spike (N-terminal and C-terminal half), nucleocapsid, and ORF3a. These responses tended to be stronger than those of five immunocompetent convalescent donors at a median of 55 days post-infection (Fig. 2a). Flow cytometry on day 145 demonstrated reduced numbers of peripheral T cells (142/μl), most of them CD8+ (60%). Many CD8 T cells (31%) co-expressed activation markers CD38 and HLA-DR, whereas CD4 T cells were less activated (Fig. 2b).
Since the patient had strong T-cell responses to SARS-CoV-2 antigens and activated CD8 T cells, we hypothesised that mutations in SARS-CoV-2 might have emerged under selective pressure by CD8 T cells and had led to inactivation of T-cell epitopes. Therefore, we examined whether these mutations were located in potential CD8 T-cell epitopes presented by the patient’s HLA class I molecules (Supplementary Table 2). For epitope prediction, we used an approach based on simple anchor motifs (Supplementary Table 3). Out of twelve non-synonymous mutations that became fixed in dominant SARS-CoV-2 variants, nine altered the amino acid sequences of putative CD8 T-cell epitopes (Supplementary Table 4), including the eight earliest mutations (day 41-97). Both nucleocapsid mutations affected the same nonameric B*35:01-restricted candidate epitope with pre-mutation sequence TPSGTWLTY (Figure 3C). Because patient samples were limited, we next studied the effect of mutations on T-cell recognition by immunocompetent HLA-matched convalescent donors. We established peptide-stimulated T-cell cultures and tested their reactivity against original and altered epitope peptides. In HLA-B*35:01-positive donor 1, stimulation with nucleocapsid TPSGTWLTY peptide, but no other HLA-B*35:01 candidates, expanded a viable T-cell culture that strongly recognized TPSGTWLTY. The T332I exchange reduced T-cell recognition, whereas subsequent T325K substitution fully abolished recognition (Figure 3D). Thus, two sequential mutations altering this epitope were required to eliminate T-cell recognition.
T cells could also be expanded from two HLA-A*02:01-positive donors with peptide ALWEIQQVV from nsp8/ORF1ab, but none of the other six HLA-A*02:01 candidates. When these T cells were stimulated with 15-mer peptides encompassing ALWEIQQVV or its variant ALWEIQQFV (ORF1ab V4101F), only the original but not the variant was recognized (Figure 3E). From two HLA-A*01:01-positive donors, T-cell cultures were established by stimulation with mixed peptides CTDDNALAY and CTDDNALAYY from nsp9/ORF1ab. T cells from donor 2 recognized each peptide, and recognition of their variants was reduced (Fig. 3F). T cells from donor 3 responded to the decameric but not the nonameric peptide, and again mutation reduced recognition (Extended data Fig. 4). Therefore, CTDDNALAY(Y) may represent two T-cell epitopes recognized by the same or different T cells. T-cell responses to each of them are impaired by the ORF1ab T4164I mutation, which replaces an optimal anchor residue threonine in position 2 by suboptimal isoleucine.
To verify HLA restriction and test the impact of mutations on binding of HLA/peptide complexes to T cells, we stained T-cell cultures with HLA/peptide tetramers. The results confirmed that TPSGTWLTY was HLA-B*35:01-restricted and CTDDNALAY(Y) were HLA-A*01:01-restricted (Extended data Fig. 5). Cultures from donor 1 were dominated by CD8 T cells that bound to the HLA-B*35:01/TPSGTWLTY tetramer (77%), and consecutive mutation strongly reduced binding (down to 1.6% for the double mutant). Similarly, cultures from donor 2 contained substantial proportions of HLA-A*01:01/CTDDNALAY(Y) tetramer-binding CD8 T cells (18–23%), and mutant peptides produced lower tetramer binding. A late-stage cryopreserved PBMC sample of the patient (day 145) was available for tetramer staining. While a large fraction (29.7%) of CD8 T cells bound to the wild type HLA-B*35:01 TPSGTWLTY multimer, binding was reduced (12.4%) with the single mutant peptide and nearly abolished (0.5%) with the double mutant (Figure 3G). Thus, the patient had maintained high numbers of CD8 T cells specific for the original epitope 84 days after the second mutation, corroborating strong cellular immune pressure, and two mutations were required to fully prevent these T cells from recognizing their target epitope.
We demonstrate emergence of immune escape mutations against antibodies as well as CD8 T-cell responses in an immunocompromised individual with persistent, ultimately fatal COVID-19. Several previous case reports have described rapid viral diversification in the setting of insufficient immune control1–3. These studies have focused on the adaptation of the SARS-CoV-2 spike protein and identified several mutations located in neutralizing epitopes of the RBD and NTD. Strikingly, also in our case two mutations in the RBD and one in the NTD of spike emerged that have previously been shown in vitro to reduce neutralising activity of antibodies targeting these regions7–9. Our data shows sensitivity of the initial autologous virus to neutralising antibodies of transfused convalescent plasma that was abrogated by the subsequent mutant isolate. This further corroborates that these mutations were selected by and resulted in escape against the transfused polyclonal antibodies. We propose judicious use of convalescent plasma in the management of immunocompromised COVID-19 patients with prolonged infection given its limited therapeutic efficacy and the risk of selection of immune escape variants10.
Notably, the antibody escape mutations that evolved in this patient are also present in defined variants of concern/interest: E484K in Beta and Gamma, the deletion at position 144 in Alpha, and the mutation S477N in Iota (B.1.526.2) and Kappa (B.1.620)11. The independent emergence of such homoplastic mutations in separate individuals and their selection advantage to become predominant variants in different populations demonstrates convergent evolution in adaptation to the human immune system4.
Our results show that selective pressure by CD8 T cells shapes SARS-CoV-2 evolution in the setting of immunosuppression. In our case, CD8 T-cell escape was initiated prior to antibody escape, and prior to administration of convalescent plasma. Subsequent antibody escape may have been facilitated by impaired cellular immunity. It was proposed that mutations in SARS-CoV-2 variants have little impact on overall CD4 or CD8 T-cell reactivity in convalescents or vaccinees12. However, individual T-cell epitopes and their variants may strongly affect control of infection in immunocompromised patients with reduced T-cell diversity, as observed for other complex viruses13. Evidence is now accumulating that CD8 T-cell responses to ancestral SARS-CoV-2 are blunted by mutations occurring in certain viral variants14,15. Of note, most of the CD8 immune escape mutations observed in the present patient have independently emerged in multiple globally circulating lineages (Supplementary Table 6), which strongly suggests that CD8 T cells contribute to SARS-CoV-2 evolution in the wider population.
Recent data has demonstrated the role of CD8 T cells in recovery from COVID-19 in patients with haematological malignancies16. However, a direct effect of CD8 T cells on virus evolution and immune evasion in a patient has not been shown prior to the present study. It is well established that strong CD8 T-cell responses are directed to nucleoprotein epitopes including HLA-B*35:01/TPSGTWLTY6,17–19. CD8 responses to nsp8/9/10 were less prominent in peptide-screening studies, in contrast to our results suggesting their substantial effect in this patient17,18. However, unbiased screening with intracellular expression fragments covering SARS-CoV-2 identified each of the epitopes HLA-A*02:01/ALWEIQQVV and HLA-A*01:01/CTDDNALAYY among the three most frequently detected CD8 T-cell responses for their HLA restriction19, consistent with a significant role of these epitopes and the non-structural proteins nsp8 and nsp9 in T-cell control of infection. CD8 T cells specific for epitopes from structural and non-structural proteins of a virus may functionally complement each other, since they recognize cells at different stages and contexts of infection20. In particular, CD8 T cells that recognize epitopes from regulatory proteins expressed early in virus-infected cells may be able to eliminate such cells before virions can be produced.
In conclusion, we find that detailed characterization of the interaction between SARS-CoV-2 and the host in immunocompromised, particularly B-cell-depleted patients, provides a unique opportunity to follow viral adaptive evolution during the course of chronic infection. Selection pressure shaping SARS-CoV-2 evolution is mediated not only by antibody responses, but importantly also by CD8 T-cell responses.