Il cavallo marino e l’energia

Il cavallo marino e l’energia

Le misure metaboliche in vitro dello studio norvegese  in cui fu evidenziato un possibile blocco al livello del piruvato deidrogenasi nei pazienti ME/CFS, sono state effettuate con il dispositivo Seahorse XFe96 della Agilent. Questo apparecchio (grande come una stampante da tavolo) permette di misurare in tempo reale – in vitro – il metabolismo energetico di cellule prelevate da pazienti (ad esempio linfociti). Come è spiegato nel video che segue, il tutto si riduce a due misure:

  1. una misura del consumo di ossigeno, che fornisce una stima del funzionamento mitocondriale;
  2. una misura della concentrazione di protoni, che fornisce una stima del funzionamento della glicolisi.

Mi risulta che l’Università degli Studi di Firenze sia in possesso di questo apparecchio (vedi qui).

Il Seahorse è attualmente impiegato nello studio di Avindra Nath (NIH) su 40 pazienti con ME/CFS post-infettiva (PI-ME/CFS). Come gruppo di controllo per questa ricerca, oltre a 20 persone sane, sono state selezionate anche 20 persone che hanno avuto la Lyme e sono completamente guarite.

seahorse

L’algoritmo di Needleman-Wunsch

L’algoritmo di Needleman-Wunsch

Allineamento di proteine

Cercare similitudini fra sequenze di peptidi è estremamente utile per stabilire rapporti evolutivi fra proteine (e quindi fra gli esseri viventi che le sintetizzano), per progettare vaccini, per studiare fenomeni di autoimmunità e altro. In questo post presento un software che si occupa proprio del confronto fra due sequenze proteiche. La scrittura di questo programma mi ha tenuto compagnia durante molti mesi, segnandomi la rotta fra le costanti ricadute e riacutizzazioni della mia malattia, malattia che fin’ora non mi ha ancora abbandonato per un solo giorno. Mentre scrivevo questo e altri codici, che ne sono lo sviluppo, inventavo anche un modo per aspettare il domani.

dihedral-angle-psi-2
Figura 1. Una sequenza di tre amminoacidi, con in evidenza il legame peptidico e l’angolo psi del legame fra il carbonio C-alpha e il C-beta di un amminoacido. Disegno di Paolo Maccallini.

Needleman e Wunsch

L’algoritmo di Needleman-Wunsch descrive un procedimento automatico che consente di calcolare il migliore allineamento possibile fra due sequenze di amminoacidi (Needleman SB, Wunsch CD, 1969). Questo metodo permette di svolgere in modo relativamente veloce e ingegnoso il confronto fra tutti gli allineamenti fra le due sequenze, considerando ogni possibile numero di lacune, in ogni possibile posizione. Il suo scopo è quello di scegliere fra questi allineamenti il migliore, ovvero quello che garantisce il ‘punteggio’ più alto, essendo tale punteggio calcolato utilizzando delle matrici quadrate di dimensione 20, dette ‘matrici di sostituzione’. Questo compito è non banale e comporta l’esame di un numero di allineamenti pari a

NeW 1.png

dove k e m sono le lunghezze dei due peptidi. Si tratta di numeri molto elevati, infatti posto ad esempio k=4 e m=6, si ottengono circa 215 allineamenti diversi. In genere però si ha a che fare con peptidi di centinaia di amminoacidi, il che comporta milioni di possibili allineamenti tra cui cercare il migliore. Ebbene, l’algoritmo di Needleman e Wunsch permette di effettuare questa analisi senza dover considerare direttamente ogni allineamento possibile. In figura 2 e 3 si ha una descrizione grafica di questo algoritmo.

new
Figura 2. L’algoritmo di Needleman-Wunsch, con l’indicazione delle variabili che ho usato nel mio codice per implementarlo. Disegno di Paolo Maccallini.
new2
Figura 3. La matrice TBM (trace back matrix) per uno specifico allineamento. Disegno di Paolo Maccallini.

Il mio programma

Il mio software per l’allineamento globale fra due proteine si chiama NeW_6 ed è scritto in Octave. Il programma presenta la stessa funzionalità di due analoghi prodotti di largo impiego, che sono LALIGN dello Swiss Institute of Bioniformatics e EMBOSS Needle dell’Europen Bioinformatic Institute. In particolare il mio programma ha le seguenti caratteristiche:

  • permette all’utente di scegliere fra un set di comuni matrici di sostituzione;
  • ha un modello di lacuna del tipo a+b(x), dove a è la penalità della lacuna iniziale, b quella delle lacune di estensione e x è il numero di lacune;
  • prevede lacune alla fine delle sequenze.

Lo sviluppo del programma, nonché il codice, si trovano in questo PDF, dove è possibile seguire vari esempi applicativi e varie versioni del programma stesso, nonché dei test in cui ne confronto l’output con i programmi attualmente in uso.

Applicazioni

Perché scrivere un programma che esiste già? Mi è servito per penetrare i metodi di allineamento fra due sequenze di amminoacidi, ma soprattutto mi ha permesso di costruire programmi più complessi (non inclusi nel PDF di cui sopra) che sto attualmente utilizzando per risolvere problemi di immunologia. In cerca di una soluzione.

Testing the lymphocyte transformation test for Lyme disease

Testing the lymphocyte transformation test for Lyme disease

In questo articolo dimostro che un test LTT per malattia di Lyme che utilizzi come uno degli antigeni la OspC (proteina integra) di B. burgdorferi sensu stricto può teoricamente risultare positivo (falso positivo) in soggetti con aumentata permeabilità intestinale.

Abstract

Some lymphocyte transformation tests (LTT) popular in Europe for the diagnosis of Lyme disease, use full length OspC of B. burgdorferi as one of their antigens and request a positive stimulation index against only one or two antigens, in order to be considered positive. In what follows, we demonstrate that, in case of patients with gut bacteria translocation, such a test has a theoretical risk of false positive results.

Lymphocyte transformation test

Lymphocyte transformation test (LTT) is an assay which allows to measure the activity of peripheral blood Th cells against specific antigens. T cell activation starts shortly after infection, with T cells proliferation and the production of cytokines (such as INF-γ) which regulate the adaptive immune response (Sompayrac, 2012). As T cell response vanishes after the resolution of the infection (Kaech, et al., 2007), LTT may be useful in providing a proof of active infection. When a LTT assay is performed, Th cells from peripheral blood of a patient are exposed to proteins from a particular pathogen. If a significant reaction is noted, which could be either Th cells proliferation or INF-γ expression, the assay is considered positive and suggestive of an active infection by that particular pathogen. The response is expressed through a number, often referred to as stimulatory index (SI). In Lyme disease, several attempts have been made in order to obtain such a tool, either by T cells proliferation assays or by INF-γ measures (Dressler, et al., 1991), (Chen, et al., 1999), (Valentine-Thon, et al., 2007), (von Baehr, et al., 2012), (Callister, et al., 2016 May). Nevertheless, this procedure has not been fully recognized as useful at present and neither the European guidelines (Stanek, et al., 2011) nor the CDC (Centers for disease control and prevention, 2015) recommend the use of this kind of test.

TCR.png
Figure 1. Presentation of an antigen to a helper T cell by MHC II molecule.

Th cells activation and cross-reactive T cell epitopes

Th cells are activated when their T cell receptors (TCR) recognize a complementary antigen presented by MHC II molecules (see Figure 1) (Sompayrac, 2012). Peptides presented by MHC II to T helper cells are exclusively linear epitopes, and they have a length between 13 and 17 amino acids (Rudensky, et al., 1991). Various experiments have demonstrated that peptides with 5 identical amino acids in a sequence of 10 have good chances to represent cross-reactive T cell epitopes (Root-Bernstein, 2014). That said, the algorithm described above for the LTT test is not free from the risk of false positive results, as on the surface of an antigen there might be one or more linear epitopes of 10 amino acids which share at least 5 amino acids with some epitope of 10 amino acids from another pathogen. This risk is particularly high when the assay uses complete proteins as antigens, and when a high SI for only a few antigens is required in order to have a positive result of the test.

OspC and Pseudomonas aeruginosa

We have used BLAST from NCBI (National Library of Medicine), with OspC from Borrelia burgdorferi (strain ATCC 35210 / B31 / CIP 102532 / DSM 4680) identified by the swiss-prot ID Q07337 () as query sequence, settings being as follows: expected threshold of 10, BLOSUM62 as substitution matrix, and a word of 3 amino acids. We have built a custom database with the main Phyla of the human gut microbiome observed in a healthy population, which are Bacteroides, Firmicutes, Proteobacteria, Verrucomicrobia, Actinobacteria, Tenericutes, and Euryarchaeota (Giloteaux, et al., 2016). One of the possible matches that BLAST gives back is the following alignment between the query sequence and the outer membrane protein G (OprG) of Pseudomonas aeruginosa (PDB ID: 2X27):

OspC_OmpG.png

As you can see, we have 6 identical amino acids in a peptide 10 amino acids long. This means that this peptide from Borrelia burgdorferi could theoretically binds a Th cell previously activated by P. aeruginosa. As peptides presented by antigen presenting cells through MHC II are digested in the cytoplasm of these leukocytes, they do not need to be surface exposed in order to elicit a response in  Th cells, and don’t need to belong to a surface protein, either. But in an LTT assay, T cells are simply exposed to antigens, this means that the peptide 111-120 of OspC has to be surface exposed, otherwise it will not engage the T cell receptor. As you can see from Figure 2 this is actually the case. The 3D structure of OspC from B. burgdorferi strain B31 used for that picture has been experimentally determined with X rays and a resolution of 2,51 Å in 2001 (Kumaran, et al., 2001) and its MMDB ID is 15958 (). The conclusion from this data is that Th cells from a patient with an active infection by P. aeruginosa could proliferate and produce INF-γ when exposed to OspC from B. burgdorferi. In other words, a patient with an active P. aeruginosa infection would come out to have a positive LTT test for OspC.

OspC.png
Figure 2. Peptide 111-120 (in yellow) of OspC from B. burgdorferi (B31) is surface exposed.

Gut bacteria translocation

A disrupted mucosal barrier of the bowel, with consequent translocation of bacteria from the gut to the peripheral blood, has been described in patients with liver diseases (Zhu, et al., 2013), chronic HIV infection (Openshaw, 2009), Crohn’s disease (Wyatt, et al., 1993), and in myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) (Giloteaux, et al., 2016). In ME/CFS it has been possible, in particular, to demonstrate the translocation of Pseudomonas aeruginosa, among other gram-negative enterobacteria. In fact serum concentration of IgA against lipopolysaccharides from P. aeruginosa and other enterobacteria has been found to be significantly greater in ME/CFS patients than in normal volunteers (Maes, et al., 2007). Thus in ME/CFS patients the adaptive immune system usually reacts against pathogens which exit from the gut, and in particular we know that it reacts against P. auruginosa.

Lyme disease LTT test and ME/CFS

We will now discuss the theoretical risk of false positive results, in the ME/CFS population, of a specific LTT test for Lyme disease, widely used in Europe. This test provides results for the reaction of Th cells against three sets of antigens:

  1. Borrelia burgdorferi ‘Fully Antigen’, which are antigens from burgdorferi B31;
  2. Borrelia burgdorferi ‘OSP-mix’, which are OspA, OspC, and DbpA from the three main European pathogenic species of Borrelia (B. burgdorferi, B. garinii and B. afzelii);
  3. Borrelia burgdorferi LFA-1, which is a linear epitope shared by OspA (B. burgdorferi ss) and by some human leukocytes.

As you can see, OspC from B.burgdorferi sensu stricto is present in both the first and the second set of antigens. Based on what discussed above, we are now able to predict that Th cells from the peripheral blood of a patient with P. aeruginosa translocation due to gut increased permeability, might react in both these assays, thus leading to a false positive LTT test for Lyme disease.

Conclusion

ME/CFS patients are among the main users of this kind of tests, because of the similarities between Lyme disease and the clinical picture of ME/CFS (Gaudino, et al., 1997). ME/CFS patients have a high prevalence of increased gut permeability and gut microbiome translocation (Giloteaux, et al., 2016), and their immune system reacts against P. aeruginosa in many cases (Maes, et al., 2007). Thus, each LTT test for Lyme disease which uses full length OspC from B. burgdorferi ss as antigen, could theoretically lead to a high rate of false positive results in this population of patients. The Lyme disease LTT test discussed above, which is currently popular in Europe, is one of such tests. More researches are warranted in order to confirm or exclude the theoretical danger of cross reaction of Lyme disease LTT tests with gut microbiome. Moreover, on the basis of what here presented, it might be possible to develop an LTT test specific for the diagnosis of gut bacteria translocation.

Cleavage of perforin within NK cells in ME/CFS

Cleavage of perforin within NK cells in ME/CFS

Paolo Maccallini

In questo studio porto dei dati a favore di un incremento dell’attività proteolitica all’interno delle NK, nei pazienti CFS. In particolare ipotizzo che la ridotta citotossicità delle NK sia dovuta alla digestione della perforina citoplasmatica, riconducibile a qualche enzima proteolitico, secondo una dinamica simile a quella descritta alcuni anni fa per l’Rnase-L.

Abstract
Rnase-L and perforin are two proteins involved in host defense against viral infections. In chronic fatigue syndrome (ME/CFS), a disabling pathology of unknown etiology, the pathway which leads to the activation of Rnase-L and the genetic expression of NK cells perforin have been reported to be both up-regulated. Moreover, while researchers have described the presence of truncated forms of Rnase-L in ME/CFS patients, another work documents a depleted intracellular perforin level in NK cells from ME/CFS subjects. In this paper, I propose the hypothesis that the same proteolytic activity is involved in both Rnase-L and NK perforin cleavage. This idea is based on the identification of a significant local alignment between their two primary structures, which involves sub sequences that are surface exposed in both tertiary structures. This peptide divides Rnase-L into two truncated forms, one of which with the same length as the one described in ME/CFS patients. Moreover, it represents a possible B cell epitope for human perforin and this might provide an explanation for the low level of intracellular perforin reported in NK cells from CFS patients, as discussed below.

Rnase-L

As a consequence of a viral threat, our own cells which are infected by pathogens start releasing INF-α, a cytokine which activates natural killer cells (NK), main actors of the innate immune system anti-viral response (Sompayrac, 2012). INF-α has other functions as well, and one of them is to promote the synthesis of the active form of Rnase-L, which is an enzyme implicated in the degradation of viral RNA (Bastide, et al., 2002). Rnase-L is a protein 741 amino acids long, with a mass of 83,533 kDa (ID: Q05823 ). It has been demonstrated that in ME/CFS a novel form of this molecule is present, with a weight of only 37 kDa, which is considered to be a dysfunctional variant of the 83 kDa peptide (Suhadolnik, et al., 1997). It accumulates in peripheral blood mononuclear cell (PBMC) extracts from ME/CFS patients and has been considered a potential diagnostic marker (De Meirleir, et al., 2000). Some years later, an increased proteolytic activity in PBMC extracts from CFS patients, has been demonstrated to be responsible for the accumulation of the truncated form. More precisely, researchers found that proteases in CFS PBMC extract, as well as purified human leukocyte elastase, give rise to two fragments of Rnase-L, the one already found in previous studies and another one of 30kDa. The 37 kDa fragment includes the N-terminal end of the full size Rnase-L, while the 30 kDa fragment includes the catalytic site in the C-terminal part (Demettre, et al., 2002). Moreover, it was previously found that the whole pathway which leads to Rnase-L activation is up-regulated in ME/CFS patients with low molecular weight Rnase-L (Suhadolnik, et al., 1994).

 Perforin

 Perforin (ID: P14222 ) is an enzyme of 555 amino acids, with a weight of 61,7 kDa, used by NK cells and T CD8+ cells to induce apoptosis of infected cells, a process called cytotoxicity. More in detail, perforin forms pore-like structures to facilitate the entry of granzymes into the target cell, and granzymes activate several apoptosis pathways that ensure effective killing of the target cell (Sompayrac, 2012). It has long been recognized a cytotoxicity defect in NK cells from CFS patients, since first attempts in understanding the immunology of ME/CFS (Caligiuri, et al., 1987). It was subsequently possible to demonstrate that low cytotoxicity of NK cells from ME/CFS patients is associated with a low level of perforin within NK cells (Maher, et al., 2005). Surprisingly, a higher than normal perforin gene expression was found some years later, despite both the low NK cytotoxic activity and intracellular level of perforin in ME/CFS patients. In fact, in a 2011 study, the level of perforin mRNA was found to be elevated in NK cells from CFS patients (Brenu, et al., 2011).

The hypothesis

You would have noticed that both intracellular NK perforin and Rnase-L share the same pattern of a depleted level and an up-regulated pathway, in ME/CFS patients. In the case of Rnase-L, it has been possible, as mentioned, to further specify that low level of full length enzyme is due to the presence of a low molecular weight Rnase-L. I propose the hypothesis that perforin shares the same fate as Rnase-L of being digested by some proteases within NK cells, in ME/CFS patients. In order to legitimate this hypothesis, I am going to demonstrate the presence of a local similarity between primary structures of Rnase-L and perforin. I will show then that this shared peptide is a possible target for proteolytic activity as it is surface exposed in both these molecules. Moreover, I will provide a possible explanation for the finding of a low level of perforin in NK cells from MECFS patients.

In search for similarities between Rnase-L and perforin

Now, we will perform a local alignment between human perforin (ID: P14222 ) and Rnase-L (ID: Q05823 ) in search for some common peptides between the two primary structures through SIBI’s LALIGN (), using the following settings (default settings):

rnase-l-1

Remember that LALIGN uses a gap model op_gap + ex_gap(x), where x is the length of the gap; thus an opening gap of -12 with an extending gap of -2 corresponds to an opening gap of -14 with an extending gap of -2 in a gap model of the type op_gap + ex_gap(x-1). LALIGN gives us the following best alignment:

rnase-l-2

For settings in Eq. 1, LALIGN calculates a value of 0.1887 for λ, and of 0.02798 for K. Thus, according to local alignment statistic (Karlin, et al., 1990) we have that the number p of local alignments between two random sequences of 741 and 555 amino acids, with a score at least S, is

rnase-l-3

Consider now that the score  is calculated according to BLOSUM50, with a scale ln(2)/3. In order to calculate the score S to put in Eq. 2, we have to change the scale of the scoring matrix:

rnase-l-4

According to Eq. 2, the number p of local alignments with a score at least 61.22 is

rnase-l-5

A local alignment is usually considered to have good significance when it reaches a p of 0.05 (Altschul, 1991), nevertheless two times that threshold value could still be of interest.

rnase-l 6.png

The 30kDa Rnase-L fragment

Demettre and colleagues were able to identify that the N-terminal end of the 30kDa Rnase-L fragment was 500-HLADFDKSI-508  in case of digestion by human elastase, and 492-LIDSKKAAH-500 when cleavage was obtained through PBMC extract from CFS patients (Demettre, et al., 2002). Interesting enough, in both cases we have a fragment which may be produced when the target of the proteolytic activity is the local alignment found for Rnase-L and perforin. A way to further verify this hypothesis is to see if the peptide found in the local alignment is surface exposed and thus available to proteolytic enzymes. Rnase-L structure has been experimentally determined by X-ray diffraction method, with a resolution of 2.31 Å (Han, et al., 2014). Its ID in the molecular modelling database (MMDB) is 118306, while its protein data bank (PDB) ID is 4OAU. You can find molecule’s 3D structure in this page (). I have downloaded its data in ASN.1 format and I have read it with Cn3D 4.3.1 () and then I have searched for the peptide of the local alignment indicated above (Table 1) and highlighted it in yellow. As you can see (Figure 1) this peptide is surface exposed in its central residues, even if it has not a good protrusion from the  molecule’s surface, on the contrary. Nevertheless, it may still be available for protelytic enzymes.

A possible explication for low perforin content

As mentioned, NK cells from CFS patients have a low perforin content (Maher, et al., 2005), associated to a high perforin gene expression (Brenu, et al., 2011). NK perforin content has been evaluated with the use of a phycoerythrin labeled anti-perforin antibody with an assay described in a previous study (Maher, et al., 2002). According to my hypothesis, perforin is digested with a cleavage between residues 329 and 353. If the epitope of the anti-perforin antibody used by Maher and colleagues to detect NK perforin included part of this peptide, the antibody would fail to detect perforin, even if truncated forms of this enzyme were present. Unfortunately, we don’t know the epitope of human perforin bound by the anti-perforin Ab used by Maher and colleagues. We thus predicted both linear and conformational B cells-epitopes by using Ellipro (Ponomarenko, et al., 2008), a free software which calculates the level of protrusion (protrusion index) of each residues from protein central ellipsoids of inertia which include a specified portion of the protein itself, according to Thornton’s algorithm (Thornton, et al., 1986). As the structure of human perforin has not been experimentally determined yet, we allowed Ellipro to predict a 3D model using as a template the experimental structure of murine perforin (), found in 2010 with X-ray method and a resolution of 2,75 Å (Law, et al., 2010). Through this analysis, we found several linear and conformation epitopes, with our peptide present in three of them. In particular, residues 329-335 are included in a big conformational epitope. This means that this peptide is both an available target for proteolytic enzymes and a possible epitope for an anti-perforin antibody. In Figure 2 you can appreciate the surface exposure of peptide 329-335 in a model built by ModBase using murine perforin () as a template. This model has been downloaded in .PDB format, then converted to ASN1 data, using VAST search () page and hence read with Cn3D 4.3.1.

rnase-l-7

Conclusion

We have proposed the hypothesis that the low level of intracellular perforin in NK cells from ME/CFS patients is due to the same proteolytic activity which underlays cleavage of Rnase-L in this patients group. We have proposed a possible proteases target shared by Rnase-L and perforin and we have shown how this theory matches with available experimental data. As for the reasons of this increased proteolytic activity, we quote Demettre and colleagues, who noted that: “proteases play an important role in numerous physiological responses and, in particular, in inflammation and apoptosis” (Demettre, et al., 2002).

References

Altschul, SF. 1991. Amino acid substitution matrices from an information theoretic perspective. June 5, 1991, Vol. 219, 3, pp. 555-65.

Bastide, L, et al. 2002. Interferon and the 2-5A/Pathway. [book auth.] P Englebienne and K De Meirleir. Chronic fatigue syndrome, a biological approach. Boca Raton : CRC press, 2002, 1, pp. 1-3.

Brenu, EW, et al. 2011. Immunological abnormalities as potential biomarkers in Chronic Fatigue Syndrome/Myalgic Encephalomyelitis. J Transl Med. May 2011, Vol. 9, 81.

Caligiuri, M, et al. 1987. Phenotypic and functional deficiency of natural killer cells in patients with chronic fatigue syndrome. J Immunol. Nov 15, 1987, Vol. 139, 10, pp. 3306-13.

De Meirleir, K, et al. 2000. A 37 kDa 2-5A binding protein as a potential biochemical marker for chronic fatigue syndrome. Am J Med. Feb 99-105, 2000, Vol. 108, 2.

Demettre, E, et al. 2002. Ribonuclease L proteolysis in peripheral blood mononuclear cells of chronic fatigue syndrome patients. J Biol Chem. Sep 20, 2002, Vol. 277, 38, pp. 35746-51.

Han, Y, et al. 2014. Structure of human RNase L reveals the basis for regulated RNA decay in the IFN response. Science. Mar 14, 2014, Vol. 343, 6176, pp. 1244-8.

Law, RH, et al. 2010. The structural basis for membrane binding and pore formation by lymphocyte perforin. Nature. 2010 Nov 18;468(7322):. Nov 18, 2010, Vol. 468, 7322, pp. 447-51.

Maher, KJ, et al. 2002. Quantitative fluorescence measures for determination of intracellular perforin content. Clin Diagn Lab Immunol. Nov 2002, Vol. 9, 6, pp. 1248-52.

Maher, KJ, Klimas, NG and Fletcher, MA. 2005. Chronic fatigue syndrome is associated with diminished intracellular perforin. Clin Exp Immunol. . Dec 2005, Vol. 142, 3, pp. 505-11.

Ponomarenko, J, et al. 2008. ElliPro: a new structure-based tool for the prediction of antibody epitopes. BMC Bioinformatics. Dec 2, 2008, Vol. 9, 514.

Sompayrac, Lauren M. 2012. How the Immune System Works. Fourth Edition. s.l. : Wiley-Blackwell, 2012.

Suhadolnik, RJ, et al. 1997. Biochemical evidence for a novel low molecular weight 2-5A-dependent RNase L in chronic fatigue syndrome. J Interferon Cytokine Res. . Jul 1997, Vol. 17, 7, pp. 377-85.

Suhadolnik, RJ, et al. 1994. Changes in the 2-5A synthetase/RNase L antiviral pathway in a controlled clinical trial with poly(I)-poly(C12U) in chronic fatigue syndrome. In Vivo. . Jul-Aug 1994, Vol. 8, 4, pp. 599-604.

Thornton, JM, et al. 1986. Location of ‘continuous’ antigenic determinants in the protruding regions of proteins. EMBO J. . Feb 1986, Vol. 5, 2, pp. 409-13.